7/27/2004 --------- [bash-3.0 released] 7/28 ---- array.c - in array_insert(), make sure the value to be added is non-NULL before calling savestring() on it builtins/reserved.def - fix description of `CDPATH' lib/readline/display.c - when expanding a prompt that spans multiple lines with embedded newlines, set prompt_physical_chars from the portion after the final newline, not the preceding portion. Bug reported by "Ralf S. Engelschall" make_cmd.c - explicitly declare `lineno' in function prologue for make_case_command builtins/evalfile.c - include `trap.h' for declaration for run_return_trap bashline.c - fix a `return' without a value in enable_hostname_completion general.c - include test.h for extern declaration for test_eaccess externs.h - add declaration for zcatfd tests/{history,histexp}.tests - unset HISTFILESIZE to avoid problems if a value of 0 is inherited from the environment 7/30 ---- bashline.c - small changes to glob_expand_word to perform tilde expansion before attempting globbing builtins/Makefile.in - fix the install-help target to not cd into the `helpfiles' subdirectory, so a value of $INSTALL_DATA containing a relative pathname (e.g., .././support/install.sh) remains valid 7/31 ---- subst.c - new function, mbstrlen(s), returns length of a multibyte character string include/shmbutil.h - new macro, MB_STRLEN(s), calls mbstrlen or STRLEN as appropriate builtins/trap.def - small change so that a first argument that's a valid signal number (digits only -- no symbolic names) will be treated as a signal and reverted back to the original handling disposition. Fixes debian complaints subst.c - call MB_STRLEN instead of STRLEN where appropriate in parameter_brace_expand_length to handle multibyte characters properly - call MB_STRLEN instead of strlen in verify_substring_values so that negative substrings of strings with multibyte chars work properly 8/1 --- jobs.c - describe_pid needs to write to stderr, not stdout (POSIX) - start_job, since it's only used by builtins (fg/bg), needs to write its output to stdout, not stderr (POSIX) sig.c - add an `orig_flags' member to struct terminating_signal so the original signal handling flags (SA_RESTART, etc.) can be preserved on POSIX systems - make sure to preserve the signal flags state in initialize_terminating_signals and reset them for child processes in reset_terminating_signals builtins/fc.def - fixed an off-by-one error that caused `fc -l' to list one too many history entries - in posix mode, `fc' should not list any indication as to whether or not history lines have been modified (POSIX) - when in posix mode, the default editor for `fc' should be `ed' (POSIX) doc/bashref.texi - updated the description of `trap' behavior when given a first argument that is a valid signal number - noted that `fc -l' won't indicate whether a history entry has been modified if the shell is in posix mode builtins/command.def - fixed bug: `command -v' is supposed to be silent if a command is not found builtins/hash.def - `hash' should print its `hash table empty' message to stderr lib/readline/misc.c - back out 7/7 change to _rl_maybe_save_line; it breaks emacs-mode ^P general.c - changed base_pathname so that it will return reasonable results for non-absolute pathnames -- this is what is intended by all of its callers arrayfunc.c - fix array_variable_part to return NULL if it finds an invisible variable in the hash table. Fixes seg fault caused by referring to unset local variable using array notation {locale,variables}.c - support LC_TIME as a special locale variable so HISTTIMEFORMAT tracks the current locale 8/2 --- variables.c - fixed small memory leak in makunbound() when a local array variable is unset. Fix from William Park lib/readline/display.c - fixed a problem when computing the number of invisible characters on the first line of a prompt whose length exceeds the screen width (should only happen when invisible characters occur after the line wrap). Bug reported by agriffis@gentoo.org builtins/command.def - `command -V' passes a new flag, CDESC_ABSPATH, which means to convert to an absolute path builtins/type.def - in posix mode, `type' and `command -v/-V' should not report non-executable files, even if the execution code will attempt to run them. Other posix shells do this doc/bashref.texi - add note to POSIX Mode section describing behavior of type and command when finding a non-executable file execute_cmd.c - force extended_glob to 1 before calling binary_test in execute_cond_node so that the right extended pattern matching gets performed 8/3 --- braces.c - make sure lhs[0] and rhs[0] are cast to `unsigned char' so chars with values > 128 are handled correctly builtins/printf.def - change bexpand() and printstr() to handle strings with a leading '\0' whose length is non-zero, since that's valid input for the `%b' format specifier subst.c - fix a couple of instances of find_variable that didn't check the result for an invisible variable variables.c - BASH_ARGC, BASH_ARGV, BASH_SOURCE, BASH_LINENO no longer created as invisible vars pcomplete.c - make sure COMP_WORDS is not invisible when bind_comp_words returns - ditto for COMPREPLY in gen_shell_function_matches 8/4 --- braces.c - fix problem where ${ was ignored but did not increment the open brace count. Bug reported by Tim Waugh variables.c - if make_local_variable finds a variable in the correct context in the right variable scope, make sure it's not invisible before returning it 8/5 --- builtins/trap.def - fixed usage message to show `action' as not optional, though it actually is when not in posix mode (for a single argument) 8/7 --- configure.in - kfreebsd-gnu has had its sbrk() problems fixed, and no longer needs to be configured --without-gnu-malloc lib/readline/vi_mode.c - in rl_vi_search, free any saved history line before starting the search, so failure leaves you at that line, not the last line in the history (assuming the current line is not the last history line). Fix from llattanzi@apple.com to replace fix of 7/7 8/9 --- support/Makefile.in - renamed `mostly-clean' target to `mostlyclean' 8/11 ---- lib/readline/vi_mode.c - make same change for EOL in multibyte character case of rl_vi_change_char 8/12 ---- subst.c - in verify_substring_values, fix off-by-one error checking bounds of `offset', esp. in array values (e.g., getting the highest element of an array) 8/16 ---- aclocal.m4 - change BASH_CHECK_DEV_FD to make sure that file descriptors > 2 are accessible via /dev/fd, unlike FreeBSD 5.x lib/sh/strftime.c - make sure `zone' is initialized with gettimeofday before it is used - work around HPUX lack of `altzone' and differing definitions of `timezone' lib/malloc/malloc.c - internal_memalign and memalign now take a `size_t' as their first argument, which seems to be the prevailing standard lib/malloc/{malloc.c,shmalloc.h} - change sh_memalign to take a `size_t' as its first argument builtins/echo.def - if posixly_correct and xpg_echo are both set, don't try to interpret any arguments at all, as POSIX/XOPEN requires (fix inspired by Paul Eggert) doc/bashref.texi - amend description of bash posix mode to include new echo behavior builtins/fg_bg.def - allow bg to take multiple job arguments, as posix seems to specify, placing them all in the background, returning the status of the last one as the status of `bg' lib/readline/vi_mode - fix _rl_vi_change_mbchar_case (multibyte-char version of `~' command) to have the right behavior at EOL -- handle case where vi mode backs up at the end of the line 8/18 ---- array.c - check for an empty array in array_rshift before shifting elements and adjusting max_index - check for null array in array_subrange jobs.c - fix raw_job_exit_status to not ignore exit status of the last process in the pipeline when `set -o pipefail' is enabled 8/19 ---- lib/readline/mbutil.c - make sure _rl_find_next_mbchar_internal has a valid multibyte character before it checks whether or not it's a zero-width wide character and adjusts point accordingly 8/24 ---- bashline.c - new function, bash_directory_expansion, duplicates the expansions performed on the directory name by rl_filename_completion_function - call bash_directory_expansion in command_word_completion_function if we decide we're doing tilde expansion (and any other canonicalization) on the directory name being completed 8/25 ---- configure.in - use new-style AC_CHECK_HEADER to check for sys/ptem.h (which requires sys/stream.h). The correct checks are in the code, but autoconf complains if sys/stream.h is not included, rather than simply checking for the header's presence 8/26 ---- builtins/hash.def - fix a bug that prevented `hash -d' from working right (as soon as hash removed a command from the table, the bug caused it to be added right back) 8/27 ---- doc/{bash.1,bashref.texi} - explicitly note that conditional primaries that operate on files operate on the targets of symbolic links rather than the links themselves 8/30 ---- lib/readline/display.c - fix multibyte calculation of `physchars' in prompt expansion, to handle double-width multibyte characters correctly - changes to rl_redisplay to handle prompts longer than the screenwidth that might contain double-width multibyte characters. Fixes from Tomohiro Kubota 9/6 --- subst.c - change word_list_split to avoid really bad behavior caused by calling list_append for each split word -- as the list gets long, you have to traverse it every time. Keep a pointer to the end of the list and and just tack onto it 9/8 --- lib/readline/complete.c - change fnprint to calculate the displayed width of a filename in the same way as fnwidth subst.c - in verify_substring_values, when expanding ${array[@]:offset}, make sure negative offsets count from one greater than the array's maximum index so things like ${x[@}: -1} work to give the last element (requires fixing array tests) builtins/common.c - new error function, sh_wrerror(), for builtins to call when a write error occurs builtins/common.h - extern declaration for sh_wrerror() builtins/cd.def - change builtin_error call to use sh_wrerror() builtins/echo.def - report write errors with sh_wrerror() instead of just returning failure builtins/printf.def - change printstr to return failure (-1) or success (0) indication rather than void - report write errors when printstr() fails, return failure - if any of the PF/printf calls fail, report write error and return failure execute_cmd.c - change execute_in_subshell so the subshell command inherits the command timing flags from the enclosing COMMAND * 9/11 ---- [prayers for the victims of 9/11/2001] lib/sh/strnlen.c - new file, implementation of GNU libc extension function strnlen lib/sh/Makefile.in, {config.h,configure,Makefile}.in, MANIFEST - changes for strnlen configure.in - version changed to 3.1-devel doc/bash.1, lib/readline/doc/rluser.texi - added description of `-o plusdirs' to complete/compgen (thanks, Arnold) parse.y - new parser_state flag, PST_ASSIGNOK, if set indicates we're parsing arguments to a builtin that accepts assignment statement arguments - turn on PST_ASSIGNOK in read_token_word when appropriate - turn off PST_ASSIGNOK in read_token when appropriate - don't attempt to parse a compound assignment specially unless we're in a position where an assignment statement is acceptable, or PST_ASSIGNOK is set 9/13 ---- variables.c - make BASH_ARGC, BASH_ARGV, BASH_LINENO, and BASH_SOURCE non-unsettable, since the shell uses those values internally expr.c - make exponentiation right-associative, as is apparently correct 9/16 ---- arrayfunc.c - make sure convert_var_to_array marks the environment as needing recreation if the converted variable was exported 9/17 ---- braces.c - mark ${ as introducing an additional level of braces only if it's not in a quoted string -- quoted strings are handled before brace matching is done parse.y - fixed an obscure problem in history_delimiting_chars where the `in' in a case statement could have a semicolon added after it, if the `case word' was on a previous line support/config.guess - support for newest versions of tandem non-stop kernel lib/readline/display.c - in compute_lcd_of_matches, explicitly cast `text' to `char *' before passing it to rl_filename_dequoting_function lib/readline/terminal.c - bind the key sequence sent by the keypad `delete' key to delete-char (same as ^D in emacs mode) builtins/ulimit.def - in print_all_limits, don't print anything if get_limit returns -1/EINVAL, indicating that the kernel doesn't support that particular limit - add -i (max number of pending signals), -q (max size of posix msg queues), -x (max number of file locks) for systems (Linux) that support them doc/{bash.1,bashref.texi} - fix description of correspondence between FUNCNAME, BASH_LINENO, and BASH_SOURCE indices in description of BASH_LINENO 9/18 ---- lib/sh/shquote.c - don't quote CTLESC and CTLNUL with CTLESC in sh_backslash_quote, as long as the resultant string never gets sent to the word expansion functions without going through the shell parser externs.h - add extern declarations for strnlen and strpbkrk from lib/sh subst.[ch] - changes to handle case where IFS consists of multibyte characters. Changed: string_extract_verbatim, split_at_delims, string_list_dollar_star, string_list_dollar_at, list_string, get_word_from_string, setifs 9/19 ---- mailcheck.c - change file_mod_date_changed to reset the cached mail file data if the file size drops to zero lib/readline/complete.c - change append_to_match so that a non-zero value for rl_completion_suppress_append will cause no `/' to be appended to a directory name bashline.c - experimental change to suppress appending a slash for a completed filename that is found in PATH as well as a directory in the current directory under certain circumstances: a single instance found in $PATH when `.' is not in $PATH, and multiple instances found in the $PATH, even when `.' is in the $PATH 9/24 ---- command.h - new word flag: W_ASSIGNRHS, means word is rhs of assignment statement - new word flag: W_NOTILDE, means word is not to be tilde expanded - new word flag (internal): W_ITILDE, means the next character is a tilde that should be expanded general.c - new set of tilde suffixes for use when parsing the RHS of an assignment statement and =~ should not be subject to tilde expansion - if ASSIGN_P argument to bash_tilde_expand is 2, use tilde prefixes for parsing RHS of assignment statement general.[ch] - new function bash_tilde_find_word, drop-in replacement for tilde_find_word subst.c - call bash_tilde_expand with secord argument of 2 when expanding rhs of an assignment statement, so tildes after second and subsequent `=' in an assignment are not expanded - new function, expand_string_assignment, to expand the rhs of an assignment statement - add `~' to EXP_CHAR, the characters that will cause the word expansion functions to be called - move tilde expansion into expand_word_internal instead of many different calls to bash_tilde_expand scattered across different functions. NOTE: This means that double quotes surrounding a {paramOPword} expansion will cause tilde expansion to NOT be performed on `word'. I think this is right, what POSIX specifies, and consistent with the behavior of other characters in the rhs execute_cmd.c - take out calls to bash_tilde_expand before calling word expansion functions 9/26 ---- execute_cmd.c - make sure to call UNBLOCK_CHILD before returning on a pipe creation failure in execute_pipeline 9/27 ---- variables.c - change get_bash_command to deal with the_printed_command_except_trap being NULL execute_cmd.c - fix execute_simple_command to deal with the_printed_command being NULL when assigning to the_printed_command_except_trap -- fixes seg fault in savestring() parse.y - change the parser so that the closing `)' in a compound variable assignment delimits a token -- ksh93 does it this way doc/{bash.1,bashref.texi} - change description of tilde expansion to note that expansion is attempted only after the first =~ in an assignment statement builtins/declare.def - when assigning to an array variable with declare -a x=(...), make sure the last character in the rhs of the variable assignment is `)', not just that it appears somewhere 9/28 ---- command.h - add a `W_NOEXPAND' flag to inhibit all expansion except quote removal - add a `W_COMPASSIGN' flag to denote a word is a compound assignment statement parse.y - set W_COMPASSIGN on words that appear to be compound assignments subst.c - pass W_NOXPAND and W_COMPASSIGN through end of expand_word_internal subst.[ch] - new function, expand_assignment_string_to_string, calls expand_string_assignment and then string_list on the result variables.c - assign_in_env now calls expand_assignment_string_to_string 9/30 ---- builtins/common.c - change get_job_spec so the null job `%' once again means the current job 10/1 ---- subst.c - do_assignment_internal now takes a WORD_DESC * as its first argument, and uses its `word' member as the assignment string - change expand_word_list_internal to call do_word_assignment instead of do_assignment, passing it `word' instead of, e.g., `word->word' - change extract_array_assignment_list to just return the passed string minus a trailing `)' if the last character is a right paren - change do_assignment_internal to call extract_array_assignment_list subst.[ch] - change do_assignment and do_assignment_no_expand to take a `char *' instead of `const char *' first argument; change extern prototypes - new function, do_word_assignment, takes a WORD_DESC * and calls do_assignment_internal on it; add extern declaration with prototype general.h - new typedef, sh_wassign_func_t, like sh_assign_func_t but takes a WORD_DESC * as its first argument variables.[ch] - assign_in_env now takes a WORD_DESC * as its first argument 10/2 ---- command.h - new word flag, W_ASSNBLTIN, denotes that the word is a builtin command (in a command position) that takes assignment statements as arguments, like `declare' - new word flags, W_ASSIGNARG, denotes that word is an assignment statement given as argument to assignment builtin execute_cmd.c - set W_ASSNBLTIN flag in fix_assignment_words if necessary (if there are any arguments that are assignment statements) - set W_ASSIGNARG flag in fix_assignment_words if necessary subst.c - new function, do_compound_assignment, encapsulates the necessary code to perform a compound array assignment (including creation of local variables); called from do_assignment_internal - to fix the double-expansion problem with compound array assignments that are arguments to builtins like `declare', changed shell_expand_word_list to treat those arguments like assignment statements (with proper creation of local variables inside shell functions) and pass the attribute-setting portion of the statement onto the builtin. This is what ksh93 appears to do, from inspection of the `ksh93 -x' output execute_cmd.c - fix execute_simple_command: in case of pipeline or async command, when forking early, set `subshell_environment' so that it can contain both SUBSHELL_PIPE and SUBSHELL_ASYNC -- the two should not be mutually exclusive. Fixes bug reported by pierre.humblet@ieee.org - remove references to last_pid, old_command_subst_pid; use NO_PID as a sentinel value to decide whether or not a child process has been created and needs to be waited for. Submitted by pierre.humblet@ieee.org to fix recycling-pid problem on cygwin doc/{bash.1,bashref.texi} - fixed documentation of `@(pattern)' extended globbing operator -- it succeeds if the string matches one of the patterns, not exactly one. This is what ksh93 does, too lib/readline/complete.c - fixed rl_menu_complete so that a negative argument cycles backwards through the list 10/3 ---- subst.c - use W_COMPASSIGN flag in do_assignment_internal instead of deciding lexically which assignments are compound array assignments 10/6 ---- support/shobj-conf - additions for System V.5 from Boyd Gerber subst.c - in command_substitute, if subshell_environment includes SUBSHELL_ASYNC, call make_child with the `async_p' argument set to non-zero. This keeps command substitutions for async commands or pipelines from trying to give the terminal back to the shell's pgrp. make sure to save and restore last_asynchronous_pid. Fix suggested by 10/7 ---- config.h.in - add a placeholder definition for WCONTINUED_BROKEN 10/9 ---- aclocal.m4 - add BASH_CHECK_WCONTINUED, checks for glibc bug where WCONTINUED is defined but rejected as invalid by waitpid(2) configure.in - add call to BASH_CHECK_WCONTINUED, defines WCONTINUED_BROKEN redir.c - experimental change to add_undo_redirect to save manipulations to file descriptors >= SHELL_FD_BASE (10) on the list of redirections to be undone even if `exec' causes the list to be discarded doc/{bash.1,bashref.texi} - note that redirections using file descriptors > 9 should be used carefully, because they might conflict with file descriptors the shell uses internally 10/11 ----- parse.y - fix pipeline_command production to handle case where `pipeline' as `argument' of `!' or `time' is null (e.g., a syntax error not handled by the grammar) 10/13 ----- lib/readline/readline.c - new internal variable, _rl_bind_stty_chars; if non-zero, bind the terminal special characters to readline equivalents at startup - change readline_default_bindings() and reset_default_bindings() to understand _rl_bind_stty_chars lib/readline/rlprivate.h - new extern declaration for _rl_bind_stty_chars lib/readline/rltty.c - change rl_prep_terminal to add support for _rl_bind_stty_chars 10/15 ----- lib/readline/bind.c - new bindable variable, `bind-tty-special-chars', bound to value of _rl_bind_stty_chars doc/bash.1,lib/readline/doc/{readline.3,rluser.texi} - documented new readline variable `bind-tty-special-chars' builtins/pushd.def - make the first check for option `--' skip the rest of option checking 10/16 ----- lib/readline/shell.c - change sh_set_lines_and_columns to prefer setenv, which has predictable memory allocation behavior, to putenv, which does not 10/19 ----- variables.c - change push_exported_var so that a tempenv variable has to have the export attribute set (which they all do -- something to look at) and the `propagate' attribute set to be propagated down to the next scope execute_cmd.c - change execute_builtin so that if CMD_COMMAND_BUILTIN is set in the passed flags argument, call pop_scope with a value that says the builtin is not special, since `command' means that preceding variable assignments don't persist in the environment. Fixes problem with variable assignments preceding command preceding special builtin keeping those variable assignments around (when in posix mode) 10/20 ----- lib/sh/shquote.c - new function, sh_mkdoublequoted, brackets a given string with double quotes and returns a new string. Flags argument, if non- zero, means to quote embedded double quotes with backslashes externs.h - new extern declaration for sh_mkdoublequoted parse.y - use sh_mkdoublequoted after calling localeexpand() lib/sh/strtrans.c - change ansicstr to understand that (flags & 4) != 0 means to remove backslash from unrecognized escape sequences general.c - fix logic problem in assignment() that caused non-variable-starter characters to be allowed, resulting in things like `1=xxx' creating a variable `1' in the hash table 10/21 ----- bashline.c - don't call programmable_completions with an assignment statement argument 10/22 ----- lib/readline/rltty.c - in prepare_terminal_settings, turn echoing on (readline_echoing_p) if get_tty_settings fails because the input is not a terminal 10/24 ----- lib/readline/util.c - include rlmbutil.h for multibyte definitions - new function, _rl_walphabetic, wide char version of rl_alphabetic lib/readline/mbutil.c - new function, _rl_char_value(buf, ind), returns value of (possibly multibyte) character at buf[ind] lib/readline/rlmbutil.h - extern defines for _rl_walphabetic and _rl_char_value for when multibyte chars are not being used - new wrapper definitions for _rl_find_next_mbchar (MB_NEXTCHAR) and _rl_find_prev_mbchar (MB_PREVCHAR) that try to avoid unneeded function calls lib/readline/text.c - fix rl_foward_word to work with multibyte characters (or in a multibyte locale) using above utility functions - fix rl_backward_word to work with multibyte characters (or in a multibyte locale) using above utility functions 10/26 ----- parse.y - fix parse_matched_pair so that it doesn't swallow \ when parsing a $'...' construct (call shell_getc with different arg) 10/28 ----- lib/glob/glob.c - after some (compiled-in) threshold, glob_vector will stop using alloca to allocate `struct globval's and will switch to using malloc, with appropriate cleanup before returning subst.c - don't expand tildes after `=' in expand_word_internal, even if the W_TILDEEXP flag is set, unless it's the first tilde in a word marked W_ASSIGNMENT 10/31 ----- lib/readline/text.c - make sure rl_point doesn't go below 0 in rl_delete_horizontal_space (from SUSE, but not sent in) shell.c - make sure shell_is_restricted skips over a single leading `-' in the shell name (from SUSE, but not sent in) lib/readline/display.c - disable `fast redisplay' at the end of the line if in a locale that supports multibyte characters (from SUSE, but not sent in) lib/readline/histexpand.c - fix a problem with finding the delimiter of a `?' substring when compiled for multibyte characters (from SUSE, but not sent in) 11/1 ---- lib/readline/display.c - correct some assignments to _rl_last_c_pos: when in a multibyte locale, it's used as an absolute cursor position; when not using multibyte characters, it's a buffer offset. I should have caught this when the multibyte character support was donated 11/5 ---- general.c - change `assignment()' to accept `+=' assignment operator arrayfunc.[ch] - bind_array_variable and assign_array_element both take a new `flags' argument - assign_array_var_from_string, assign_array_from_string, and assign_array_var_from_word_list now all take a new `flags' argument - change assign_array_var_from_word_list to understand how to append to an array variable - change assign_array_var_from_string to understand how to append to an array variable. It does not unset the previous value if appending, allowing both old values to be changed and new ones to be added subst.h - new flag #defines to use for evaluating assignment statements {subst,variables}.c, builtins/{declare,read}.def - change callers of assign_array_element and bind_array_variable - change do_compound_assignment to understand assignment flags - change do_assignment_internal to set assignment flags and pass them to underlying functions pcomplete.c,builtins/{declare,read}.def - fix callers of assign_array_var_from_string, assign_array_var_from_word_list variables.[ch] - make_variable_value now takes a new `flags' argument - make_variable_value now understands how to append to a particular variable, using the old value - bind_variable_value now takes a new `flags' argument - change make_variable_value to understand ASS_APPEND flag - bind_variable now takes a new `flags' argument - bind_variable_internal now takes a new `flags' argument arrayfunc.c - change callers of make_variable_value to add flags arg builtins/declare.def - change callers of bind_variable_value to add flags arg {execute_cmd,mailcheck,pcomplete,shell,subst,variables}.c,parse.y builtins/{cd,command,declare,getopts,read,set,setattr}.def - change callers of bind_variable to add flags arg variables.c - change callers of bind_variable_internal - change bind_variable_internal to pass assignment flags on to make_variable_value - change assign_in_env to treat `var+=value' like `var=value' arrayfunc.c - break code that actually constructs the new value and assigns it to a particular array index out into a new functions: bind_array_var_internal. This fakes out make_variable_value by passing a dummy SHELL_VAR * so it can do proper appending and other += processing - changes to assign_array_var_from_string to accept and process as if they were `standalone' assignment statements array assignment words of the form [ind]+=val 11/7 ---- builtins/declare.def - added support for `declare [flags] var+=value'. `Flags' are applied before the assignment is performed, which has implications for things like `-i' -- if -i is supplied, arithmetic evaluation and increment will be performed builtins/setattr.def - add support for `+=' assignment for rest of `assignment builtins': export, readonly 11/12 ----- lib/readline/display.c - make sure prompt_physical_chars and prompt_invis_chars_first_line are reset to 0 if the prompt string passed to rl_expand_prompt is NULL or empty 11/14 ----- {configure,config.h}.in - check for `raise', define HAVE_RAISE if available lib/intl/dcigettext.c - make sure `raise' is defined if HAVE_RAISE is not before eval-plurah.h is included lib/malloc/trace.c - put extern declaration for imalloc_fopen inside the MALLOC_TRACE #ifdef 11/16 ----- lib/intl/Makefile.in - make sure SHELL is defined to cpp lib/intl/dcigettext.c - make sure we use getcwd() even if HAVE_GETCWD is not defined after including config.h; if SHELL is defined, #define HAVE_GETCWD 11/18 ----- trap.[ch] - new function, int signal_in_progress(int sig), returns TRUE if the trap handler for signal SIG is currently executing 11/19 ----- redir.c - slightly change do_redirection_internal to set the close-on-exec flag for file descriptors > 2 used to save file descriptors < 2 using explicit redirections (e.g., `exec 3>&1'). This keeps file descriptors pointing to pipes from being left open but doesn't change the shell's file descriptor semantics 11/20 ----- doc/{bash.1,bashref.texi} - correct some minor typos, forwarded from doko@debian.org 11/22 ----- doc/bash.1,lib/readline/doc/{readline.3,rluser.texi} - documented detail that yank-last-arg and yank-nth-arg use the history arg expansion code (and, as a result, are subject to restrictions of the history-comment character) 11/23 ----- execute_cmd.c - changes so that BASH_COMMAND preserves its value into a DEBUG trap: for commands, arithmetic for command expressions, select commands, case commands, (( commands, [[ commands, simple commands 11/24 ----- doc/{bash.1,bashref.texi} - changed description of `set' builtin slightly so that it is clear that only variables are displayed in posix mode and that read-only variables can't be reset by simply sourcing the output of `set' lib/sh/strftime.c - don't try to redefine `inline' if it's already defined 11/26 ----- execute_cmd.c - fix execute_function to check funcname_a after function execution, since FUNCNAME can be changed or unset within a function 11/27 ----- builtins/evalfile.c - make same changes as 11/26, this time to _evalfile execute_cmd.c - change execute_function to run the return trap after a function completes execution even if the shell is compiled without DEBUGGER defined trap.c - change reset_or_restore_signal_handlers so that the RETURN trap is not inherited by command substitution when DEBUGGER is not defined 11/30 ----- lib/readline/misc.c - fix memory leaks in _rl_free_history_entry and rl_maybe_replace_line caused by not freeing `timestamp' member of history entry - make sure timestamp is initialized to NULL in rl_maybe_save_line 12/1 ---- execute_cmd.c - fix execute_function so a function calling `return' will run the RETURN trap, if one's defined doc/{bash.1,bashref.texi} - fix description of RETURN trap in various places to indicate that it's only inherited by shell functions if function tracing is on globally or has been enabled for that function - fix documentation to indicate that the DEBUG and RETURN traps are inherited under the same conditions execute_cmd.c - a function does not inherit the RETURN trap if a DEBUG trap is currently running 12/2 ---- lib/glob/xmbsrtowcs.c - change xmbsrtowcs to handle the one case where malloc can fail (though it should not matter) -- don't try to free a null pointer 12/9 ---- subst.c - fix get_var_and_type to handle var[@], where `var' is a scalar variable, identically to var -- all calling expansions can now handle var[@] like var. Bug reported by agriffis@gentoo.org 12/10 ----- lib/readline/bind.c - make new-style "\M-x" keybindings obey `convert-meta' settings (bug reported by twaugh@redhat.com) 12/14 ----- builtins/set.def - added description of `-' option to help text builtins/shopt.def - fix bug that caused `gnu_errfmt' to not be compiled in unless READLINE is defined 12/16 ----- subst.c - fixed a typo in string_extract_verbatim in first call to MBLEN (used `slen - 1' instead of `slen - i') 12/17 ----- subst.c - avoid some calls to strlen if the value is only being used for ADVANCE_CHAR and MB_CUR_MAX == 1 (since ADVANCE_CHAR doesn't need it unless multibyte characters are possible) - change string_extract_verbatim so it takes the length of the string as a parameter, so we don't have to recompute the length of the same string over and over again when doing word splitting (that kills if it's a long string) 12/18 ----- subst.c - in string_list_dollar_star, make sure to null-terminate the separator if the character is longer than one byte 12/22 ----- doc/{bash.1,bashref.texi} - changed text in quoting section explaining that double quotes do not prevent history expansion from taking place, and that backslashes escaping ! are not removed 12/28 ----- shell.c - set gnu_error_format to 1 if running under emacs. This should allow the emacs `next-error' stuff to work, at least for interactive shells parse.y - change yy_stream_get to set interrupt_immediately before calling getc_with_restart when the shell is interactive. This avoids the synchronization problem caused by the call to QUIT in read_a_line, which results in the first character after a SIGINT/^C to be dropped 12/30 ----- builtins/mkbuiltins.c - changes to write long documentation to arrays as a single string by default, rather than an array of strings -- enabled by default - new option, -S, to restore old behavior of writing multiple strings for long documentation - changes to avoid filenames written when the separate-filenames option (-H) has been supplied being run through gettext configure.in - new cofiguration option, --enable-single-help-strings (on by default), causes help text to be stored as a single string (or smaller set than one string per line) builtins/Makefile.in - pass `-S' to mkbuiltins if single-help-strings is turned off doc/bashref.texi - documented new `single-help-strings' configure option 1/3/2005 -------- jobs.c - make wait_for return a non-zero status if the job or processed waited for is suspended. Returns 128 + stop signal. This fixes the problem with `echo one && sleep 5 && echo two' displaying `two' after the sleep is suspended 1/5 --- print_cmd.c - change indirection_level_string so the code duplicates the first character of $PS4 to indicate the indirection level, rather than the first byte 1/8 --- variables.c - new special variable hook function for COMP_WORDBREAKS; sets rl_completer_word_break_characters back to NULL when the variable is unset - change bind_variable_value to understand dynamic variables with assign_function set, and handle them correctly. If the variable is being appended to, use make_variable_value to create the new value - change bind_variable_internal to understand dynamic variables with assign_function set, and handle them the same way - RANDOM and LINENO now get the integer attribute, so appending works as expected - ditto for HISTCMD, MAILCHECK, OPTIND lib/readline/display.c - change _rl_make_prompt_for_search to set prompt_physical_chars appropriately - rl_save_prompt and rl_restore_prompt save and restore prompt_prefix_length - change redraw_prompt to use rl_save_prompt and rl_restore_prompt - change rl_restore_prompt to set the `save' variables back to NULL/0 so code can check whether or not the prompt has been saved - change rl_message and rl_clear_message to save and restore the prompt if the caller has not already done it (using a simple semaphore-like variable) - change rl_message to call expand_prompt, so that local_prompt and local_prompt prefix are set before calling the redisplay functions, in case the prompt is longer than a screenwidth (fixes bug reported to debian by epl@unimelb.edu.au) lib/readline/doc/rltech.texi - make sure to note that rl_save_prompt should be called before rl_message, and rl_restore_prompt before rl_clear_message pcomplete.c - make sure to save and restore the parser state around the call to execute_shell_function in gen_shell_function_matches. Fixes bug reported by a050106.1.keeLae3x@captaincrumb.com (cute) lib/readline/readline.c - fix _rl_dispatch_subseq in the case where we're recursing back up the chain (r == -2) and we encounter a key shadowed by a keymap, but originally bound to self-insert. Calling rl_dispatch with ANYOTHERKEY as the first argument will call rl_insert, but with ANYOTHERKEY (256) as the char to insert. Use the shadow keymap and set things up to dispatch to rl_insert with the shadowed key as the argument. Fixes the bug reported by Thomas Glanzmann (sithglan@stud.uni-erlangen.de) 1/13 ---- command.h - new word flag: W_HASQUOTEDNULL make_cmd.c - new function to allocate a WORD_DESC * without doing anything with a containing string: alloc_word_desc make_cmd.h - extern declaration for alloc_word_desc dispose_cmd.c - new function to just free a WORD_DESC * without freeing the contained string: dispose_word_desc dispose_cmd.h - extern declaration for dispose_word_desc subst.c - change some places to use alloc_word_desc - make same changes to word_list_quote_removal as were made to word_list_split - set W_HASQUOTEDNULL when a word is created with w->word[0] == CTLNUL and w->word[1] == '\0' subst.c - parameter_brace_expand_word now returns a WORD_DESC * -- changed callers to understand - parameter_brace_expand_indir now returns a WORD_DESC * -- changed callers to understand - parameter_brace_expand_rhs now returns a WORD_DESC * -- changed callers to understand - remove W_HASQUOTEDNULL from a word's flags when remove_quoted_nulls is called on the word's enclosed string 1/15 ---- subst.c - param_expand now returns a WORD_DESC * -- changed callers to understand - parameter_brace_expand now returns a WORD_DESC * -- changed callers to understand - in expand_word_internal, only call remove_quoted_nulls after a word is returned with W_HASQUOTEDNULL - changes to pass W_HASQUOTEDNULL flag out of expand_word_internal; changed callers to call remove_quoted_nulls only if return value has W_HASQUOTEDNULL set. This is a mostly-complete fix for the long-standing CTLNUL confusion between a quoted null expansion and the expansion of a variable with a literal '\177' in its value - change string_list_dollar_at to compute the separator character the same way as string_list_dollar_star: using the already-computed values generated in setifs() - when expanding unquoted $*, if $IFS is empty, check whether or not we're eventually going to split the results (e.g., on the rhs of an assignment statement) and concatenate the positional parameters as if the expansion were within double quotes if we're not going to split tests/iquote.tests - test cases based on old bug reports about the quoted-null vs. 0177 problem the recent code fixes 1/16 ---- dispose_cmd.c - set w->word to 0 before putting a WORD_DESC * back in the cache in dispose_word_desc; changed callers to delete those assignments variables.c - change assign_random and get_random_value so that the random number generator only gets re-seeded once in a subshell environment, and assigning a value to RANDOM counts as seeding the generator. This makes the sequences a little more predictable 1/20 ---- lib/readline/history.c - fix replace_history_entry, remove_history to return NULL if passed index is < 0 1/22 ---- lib/sh/netconn.c - fix isnetconn() to understand that getpeername can return ENOTCONN to indicate that an fd is not a socket configure.in - set BUILD_DIR to contain backslashes to escape any spaces in the directory name -- this is what make will accept in targets and prerequisites, so it's better than trying to use double quotes - set SIZE to the appropriate value if some cross-compiling tool chain is being used; `size' by default (can be overridden by SIZE environment variable) Makefile.in - use $(SIZE) instead of size; set SIZE from configure 1/31 ---- arrayfunc.c - in array_value_internal, return NULL right away if the variable's value is NULL, instead of passing a null string to add_string_to_list 2/1 --- jobs.h - new struct to hold stats and counters for child processes and jobs - change some uses of global and static variables to use members of new struct (struct jobstats) 2/2 --- jobs.[ch] - change PRUNNING to PALIVE - new define INVALID_JOB - new macro get_job_by_jid(ind), currently expands to jobs[ind] - new define J_JOBSTATE, operates on a JOB * like JOBSTATE operates on a job index - new function, reset_job_indices, called from delete_job if js.j_lastj or js.j_firstj are removed - change various functions to keep counters and stats in struct jobstats pcomplete.c, builtins/common.c, builtins/{exit,fg_bg,jobs,kill,wait}.def - change global variables (e.g., job_slots) to struct members (e.g., js.j_jobslots) - use INVALID_JOB define where appropriate - use get_job_by_jid and J_JOBSTATE where appropriate trap.c - change reset_or_restore_signal_handler to not free the exit trap string if the function pointer is reset_signal, which is used when the trap strings shouldn't be freed, like in command substitution 2/4 --- jobs.c - new function, realloc_jobs_list, copies jobs array to newly-allocated memory shrinking (or growing) size to have next multiple of JOB_SLOTS greater than js.j_njobs - change compact_jobs_list to just call reap_dead_jobs and then realloc_jobs_list, simplifying it considerably - discard_pipeline now returns `int': the number of processes freed - slightly changed the logic deciding whether or not to call compact_jobs_list: now non-interactive shells will compact the list if it reaches MAX_JOBS_IN_ARRAY in size parse.y - move test for backslash-newline after pop_string in shell_getc so that things like ((echo 5) \ (echo 6)) work right 2/8 --- jobs.h - new structs for holding status of exited background processes, as POSIX specifies - new job flag: J_ASYNC jobs.c - new functions to manipulate struct holding status of exited background processes - new members in struct jobstats to hold pointer to last created job and last created asynchronous job - initialize js.c_childmax in initialize_job_control - if the `async' arg to stop_pipeline is non-null, set the J_ASYNC flag in the job struct - set js.j_last_made_job and js.j_last_asynchronous_job in stop_pipeline - new function: find_last_proc, returns the PROCESS * to the last proc in a job's pipeline - changed find_last_pid to call find_last_proc - change delete_job to call bgp_add on the last proc of the job being deleted - change delete_all_jobs and wait_for_background_pids to call bgp_clear 2/9 --- jobs.c - change wait_for_single_pid to look for pid in bgpids.list (using bgp_search()) if find_pipeline returns NULL 2/10 ---- support/shobj-conf - change the solaris-gcc stanza so that it auto-selects the appropriate options for ld depending on which `ld' gcc says it's going to run 2/11 ---- jobs.h - add support for PS_RECYCLED as a process state, add PRECYCLED macro to test it. Change PALIVE and PRUNNING macros to not count processes in PS_RECYCLED state execute_cmd.c - restore use of last_pid as sentinel value; use NO_PID as sentinel only if RECYCLES_PIDS is defined jobs.c - change find_job to return a pointer to the PROCESS the desired pid belongs to, analogous to find_pipeline returning pointer to JOB - change find_job callers to add extra argument - change running_only arguments to find_pipeline and find_job to alive_only, since we don't want recycled pids returned here and it better describes the result - new function find_process, calls find_pipeline and searches the returned pipeline for the PROCESS * describing the desired pid - in make_child, if fork() returns the same pid as the value of last_asynchronous_pid when RECYCLES_PIDS is defined, avoid pid aliasing by resetting last_asynchronous_pid to 1 - use PRUNNING instead of child->running, since we, for the most part, don't want to consider recycled pids (e.g., in make_child()) - call find_process instead of find_pipeline in waitchld() - use PEXITED(p) instead of testing p->running == PS_DONE - in make_child, call bgp_delete to remove a just-created pid from the last of saved pid statuses - in add_process, check whether or not pid being added is already in the_pipeline or the jobs list (using find_process) and mark it as recycled if so - This set of fixes mostly came from Pierre Humblet to fix pid aliasing and reuse problems on cygwin variables.c - set $_ from the environment if we get it there, set to $0 by default if not in env doc/{bashref.texi,bash.1} - a couple of clarifying changes to the description of $_ based on comments from Glenn Morris 2/15 ---- shell.c - use strstr instead of strmatch when checking whether $EMACS contains `term' -- simpler and faster 2/18 ---- builtins/cd.def - implement posix requirement that `pwd -P' set $PWD to a directory name containing no symlinks - add new function, setpwd(), just sets (and changes exported value) of PWD doc/bashref.texi - add note to posix mode section about pwd -P setting $PWD doc{bash.1,bashref.texi} - added note that BASH_ARGC and BASH_ARGV are only set in extended debug mode - expand description of extdebug option to include everything changed by extended debug mode 2/19 ---- pathexp.h - new flag macro, FNMATCH_IGNCASE, evaluates to FNM_CASEFOLD if the match_ignore_case variable is non-zero execute_cmd.c - new variable, match_ignore_case - change call to strmatch() in execute_case_command so it includes FNMATCH_IGNCASE test.c - change call to strmatch() in patcomp() so that pattern matching calls for [[ ... ]] obey the match_ignore_case variable lib/sh/shmatch.c - if match_ignore_case is set, enable REG_ICASE in the regexp match flags builtins/shopt.def - new settable option, `nocasematch', controls the match_ignore_case variable. Currently alters pattern matching for case and [[ ... ]] commands (==, !=, and =~ operators) doc/{bashref.texi,bash.1} - updated descriptions of [[ and case to include reference to nocasematch option 2/22 ---- builtins/mkbuiltins.c - add `times' to the list of posix special builtins 2/23 ---- builtins/cd.def - posix mode no longer turns on effect of -P option on $PWD if a directory is chosen from CDPATH doc/bashref.texi - clarified that in posix mode, reserved words are not alias expanded only in a reserved word context - removed item about cd, $CDPATH, and -P from posix mode section 2/24 ---- builtins/reserved.def - minor cleanups to the description of `if' 3/2 --- subst.c - change list_string and get_word_from_string to explicitly treat an IFS character that is not space, tab, or newline *and any adjacent IFS white space* as a single delimiter, as SUSv3/XPG6 says builtins/read.def - check whether or not the number of fields is exactly the same as the number of variables instead of just assigning the rest of the line (minus any trailing IFS white space) to the last variable. This parses a field and checks whether or not it consumes all of the input (including any trailing field delimiters), falling back to the previous behavior if it does not. This is what POSIX.2 specifies, I believe (and the consensus of the austin-group list). This requires a few tests in read.tests to be changed: backslashes escaping IFS whitespace characters at the end of input cause the whitespace characters to be preserved in the value assigned to the variable, and the trailing non-whitespace field delimiter issue 3/7 --- configure.in - add -D_POSIX_SOURCE to the LOCAL_CFLAGS for Interix 3/8 --- bashline.c - make bash_directory_expansion a void function, since it doesn't have any return value 3/9 --- builtins/read.def - when testing for a pipe, use `fd' instead of hard-coding 0, since we can read from other file descriptors now lib/sh/zread.c - in zsyncfd, only set lind and lused to 0 if the lseek succeeds. If the lseek fails, we might steal input from other programs, but a failed lseek won't cause us to erroneously discard input 3/11 ---- builtins/evalstring.c - don't allow parse_and_execute to short-circuit and call exec() if the command's return value is being inverted 3/15 ---- builtins/printf.def - new macro PC to call putchar and increment number of chars printed - fixes bug in computation of value for %n format char - `tw' is now a global var so printstr can modify it using PC() - convert PF macro to use asprintf into a local buffer Preparation for printf -v var - add code to add the text printed to a `variable buffer' if -v option supplied. The buffer grows as needed - printf now takes a `-v var' option to put the output into the variable VAR rather than sending it to stdout. It does not: print partial output on error (e.g., format string error) handle NULs in the variable value, as usual 3/16 ---- parse.y - fix bug in prompt string decoding that caused a core dump when PS1 contained \W and PWD was unset (null pointer deref) builtins/printf.def - changed -v var behavior so it stores partial output into the named variable upon an error 3/24 ---- lib/readline/bind.c - bool_to_int now takes a `const char *' argument support/{printenv,recho,zecho}.c - include config.h - include "bashansi.h" for appropriate extern function declarations configure.in - on MacOS X 10.4, compensate for loader not allowing static library to override existing system dynamic library when compiling -dynamic (affects readline and history libraries); so use absolute pathname instead of -lreadline as library name lib/glob/{glob,sm_loop,smatch}.c - make sure to cast arguments to (char *) or (unsigned char *) as appropriate to avoid gcc4 warnings lib/glob/smatch.c - collsym (single-byte version) now takes a (CHAR *) first argument to match callers; cast argument to strncmp appropriately lib/sh/snprintf.c - fix ldfallback and dfallback to handle width and precision specs in the format passed to sprintf() - fix STAR_ARGS macro to deal with negative field widths and precisions 3/25 ---- builtins/printf.def - since a negative precision in a "x.x[fFgGeE]" format specifier should be allowed but treated as if the precision were missing, let it through lib/sh/snprintf.c - fix * code to deal with a negative precision by treating it as if the `.' and any digit string in the precision had not been specified - fix format parsing code to deal with a negative inline precision, e.g., "%4.-4f" by treating it as if the `'. and any digit string in the precision had not been specified - a `+' in a format specifier should only act as a flag if it comes before a `.' (otherwise it is ignored) lib/readline/vi_mode.c - new function, rl_vi_rubout, to rl_rubout as rl_vi_delete is to rl_delete; saves deleted text for possible reinsertion as with any vi-mode `text modification' command (fixes problem with `X' reported by beat.wieland@gmx.ch) lib/readline/vi_keymap.c - bind `X' in vi command mode to rl_vi_rubout lib/readline/funmap.c - add a bindable `vi-rubout' command, runs rl_vi_rubout lib/readline/text.c - rewrote internals of _rl_rubout_char to make structure cleaner lib/readline/{complete,text}.c - changed code to remove #ifdef HANDLE_MULTIBYTE where possible 3/28 ---- lib/readline/examples/rl.c - include instead of posixstat.h if READLINE_LIBRARY not defined subst.c - fix mbstrlen to treat invalid multibyte sequences as sequences of single-byte characters 4/8 --- configure.in - default SIZE to `:' if cross-compiling and an appropriate size for the target is not found 4/11 ---- subst.c - change match_upattern and match_wpattern to check whether or not the supplied pattern matches anywhere in the supplied string, prefixing and appending the pattern with `*' if necessary. If it doesn't we can short-circuit immediately rather than waste time doing up to N-1 unsuccessful calls to strmatch/wcsmatch (which kills for long strings, even if the pattern is short) 4/12 ---- configure.in - make sure the special case for MacOS X 10.4 only kicks in if the `--with-installed-readline' option isn't supplied lib/readline/{callback,readline,signals}.c - make sure rl_prep_term_function and rl_deprep_term_function aren't dereferenced if NULL (as the documentation says) builtins/mkbuiltins.c - don't bother with the special HAVE_BCOPY code; just use straight assignments builtins/ulimit.def - use _POSIX_PIPE_BUF in pipesize() if it's defined and PIPE_BUF is not 4/13 ---- execute_cmd.c - add cm_function_def to the list of control structures for which child processes are forked when pipes come in or out 4/14 ---- builtins/read.def - make sure the ^As added for internal quoting are not counted as characters read when -n is supplied 4/20 ---- redir.c - fix redir_open so that the repeat open on failure that AFS support adds restores the correct value of errno for any error message 4/26 ---- Makefile.in - make sure mksignames and mksyntax are invoked with the $(EXEEXT) extension 4/28 ---- lib/readline/readline.h - new state variable: RL_STATE_CALLBACK, means readline is using the callback interface lib/readline/callback.c - set RL_STATE_CALLBACK in rl_callback_handler_install, unset in rl_callback_handler_remove 4/29 ---- config-top.h - DONT_REPORT_SIGPIPE is now on by default, since it apparently interferes with scripts configure.in - arrange things so PGRP_PIPE is defined on Linux-2.4+ and version 3 kernels (ones that apparently schedule children to run before their parent) 4/30 ---- builtins/caller.def - add call to no_options, so it can handle `--' option doc/{bash.1,bashref.texi} - note explicitly that test, :, true, and false don't understand -- as meaning the end of options 5/7 --- support/shobj-conf - darwin 8 needs the same LDFLAGS setting as darwin 7 parse.y - in save_parser_state, make sure we cast the return value from xmalloc() to the right type - remove casts to (char *) in calls to yyerror() lib/readline/signals.c - make SIGQUIT and SIGALRM code conditional on their definition - use raise() to send a signal if we don't have kill() lib/readline/display.c - some MS-DOS and MINGW changes from the cygwin and mingw folks config.h.in - add HAVE_PWD_H for - add HAVE_FCNTL, HAVE_KILL for respective system calls - add HAVE_GETPW{ENT,NAM,UID} for passwd functions configure.in - add check for - add checks for fcntl, kill system calls - add checks for getpw{ent,nam,uid} C library functions - pass a flag indicating we're cross compiling through to CFLAGS_FOR_BUILD in Makefile.in lib/readline/complete.c - guard inclusion of with HAVE_PWD_H - don't provide a missing declaration for getpwent if we don't have it - guard calls to {get,end}pwent with HAVE_GETPWENT lib/readline/shell.c - guard inclusion of with HAVE_PWD_H - guard inclusion of with HAVE_FCNTL_H - don't provide a missing declaration for getpwuid if we don't have it - guard calls to getpwuid with HAVE_GETPWUID - don't bother with body of sh_unset_nodelay_mode if we don't have fcntl lib/tilde/tilde.c - guard inclusion of with HAVE_PWD_H - guard calls to getpw{nam,uid} with HAVE_GETPW{NAM,UID} - guard calls to {get,end}pwent with HAVE_GETPWENT Makefile.in,builtins/Makefile.in - @CROSS_COMPILE@ is substituted into CFLAGS_FOR_BUILD (equal to -DCROSS_COMPILING if bash is being cross-compiled) 5/9 --- aclocal.m4 - print version as `0.0' in RL_LIB_READLINE_VERSION if the `rl_gnu_readline_p' variable isn't 1 (accept no imitations) 5/11 ---- lib/readline/rlprivate.h - definition of a readline `search context', to be use for incremental search initially and other types of search later. Original from Bob Rossi as part of work on incremental searching problems when using callback interface lib/readline/isearch.c - functions to allocate and free search contexts - function to take a search context and a character just read and `dispatch' on it: change search parameters, add to search string, search further, etc. - isearch is now completely context-driven: a search context is allocated and passed to the rest of the functions 5/12 ---- lib/readline/isearch.c - an additional `isearch cleanup' function that can be called from the callback interface functions when the search is to be terminated - an additional `isearch callback' function that can be called from rl_callback_read_char when input is available - short-circuit from rl_search_history after initialization if the callback interface is being used lib/readline/callback.c - in rl_callback_read_char(), if RL_STATE_ISEARCH is set, call _rl_isearch_callback to read the character and dispatch on it. If RL_STATE_ISEARCH is unset when that call returns, and there is input pending, call rl_callback_read_char() again so we don't have to wait for new input to pick it up support/shobj-conf,configure.in - add support for dragonfly bsd, the same as freebsd 5/13-5/15 --------- lib/readline/callback.c - support for readline functions to `register' a function that will be called when more input is available, with a generic data structure to encapsulate the arguments and parameters. Primarily intended for functions that read a single additional character, like quoted-insert - support for callback code reading numeric arguments in a loop, using readline state and an auxiliary variable - support for callback code performing non-incremental searches using the same search context struct as the isearch code lib/readline/{callback,display}.c - if a callback function sets `_rl_redisplay_wanted', the redisplay function will be called as soon as it returns lib/readline/input.c - changes to _rl_read_mbchar to handle reading the null multibyte character and translating it into '\0' lib/readline/misc.c - break rl_digit_loop() into component functions that can be called individually from the callback code more easily - share some of the functions with rl_digit_loop1() in vi_mode.c lib/readline/readline.h - change the version #defines to reflect readline 5.1 lib/readline/search.c - break code into smaller functions that can be composed to work with the callback code more easily lib/readline/text.c - in rl_quoted_insert(), don't mess around with the tty signals if running in `callback mode' lib/readline/vi_mode.c - changed set-mark, goto-mark, change-char, and char-search to work when called by callback functions 5/17 ---- lib/readline/rlprivate.h - new struct declaration for a `reading key sequence' context lib/readline/readline.c - new variable, _rl_dispatching_keymap, keeps track of which keymap we are currently searching - functions to allocate and deallocate contexts for reading multi-char key sequences 5/18 ---- lib/readline/rlprivate.h - new struct defining a context for multiple-key key sequences (the base case is escape-prefixed commands) lib/readline/readline.c - change structure of _rl_dispatch_subseq to allow for callback code to use it - rudimentary support for supporting the existing recursion using a stack of contexts, each with a reference to the previous - fix so that ^G works when in callback mode lib/readline/callback.c - call the appropriate multiple-key sequence callback if the state is set 5/19 ---- lib/readline/readline.c - broke code from _readline_internal_char after call to rl_dispatch out into separate function: _rl_internal_char_cleanup, callable by other parts of the code - change _rl_internal_char_cleanup to unset _rl_want_redisplay after it calls (*rl_redisplay_func) lib/readline/callback.c - call _rl_internal_char_cleanup from rl_callback_read_char when appropriate 5/24 ---- lib/readline/callback.c - use _rl_dispatch_callback and a chain of _rl_keyseq_contexts to simulate the recursion used to decode multicharacter key sequences (even things like ESC- as meta-prefix) - call setjmp in rl_callback_read_char to give things like rl_abort a place to jump, since the saved location in readline() will not be valid - keep calling _rl_dispatch_callback from rl_callback_read_char while we are still decoding a multi-key key sequence - keep calling readline_internal_char from rl_callback_read_char while we are reading characters from a macro lib/readline/macro.c - use a slightly different strategy upon encountering the end of a macro when using the callback interface: when the last character of a macro is read, and we are reading a command, pop the macro off the stack immediately so the loop in rl_callback_read_char terminates when it should lib/readline/readline.c - if longjmp() is called and we end up at the saved location while using the callback interface, just return -- don't go back into a blocking read - new function to dispose a chain of rl_keyseq_cxts - only read new input in _rl_dispatch_callback if the KSEQ_DISPATCHED flag is not set in the current keyseq context -- if it is, we are traversing the chain back up and should use what we already saved - use -3 as a magic value from _rl_dispatch_subseq to indicate that we're allocating a new context and moving downward in the chain (a special return value for the benefit of _rl_dispatch_callback) lib/readline/rlprivate.h - new extern declaration for _rl_keyseq_chain_dispose 6/1 --- builtins/read.def - fixed a bug that occurred when reading a set number of chars and the nth char is a backslash (read one too many). Bug reported by Chris Morgan execute_cmd.c - fix execute_builtin so the `unset' builtin also operates on the temporary environment in POSIX mode (as well as source and eval), so that unsetting variables in the temporary environment doesn't leave them set when unset completes. Report by Eric Blake array.c - fix from William Park for array_rshift when shifting right on an empty array -- corrects calculation of array->max_index builtins/exec.def - if an exec fails and the execfail option is set, don't call restart_job_control unless the shell is interactive or job_control is set jobs.c - add a run-time check for WCONTINUED being defined in header files but rejected with EINVAL by waitpid(). Fix from Maciej Rozycki 6/20 ---- bashhist.c - make sure calls to sv_histchars are protected by #ifdef BANG_HISTORY - ditto for calls to history_expand_line_internal 6/23 ---- doc/bashref.texi - remove extra blank lines in @menu constructs variables.c - assign export_env to environ (extern char **) every time it changes (mostly in add_to_export_env define), so maybe getenv will work on systems that don't allow it to be replaced 6/29 ---- bashline.c - in bash_directory_completion_hook, be careful about not turning `/' into `//' and `//' into `///' for benefit of those systems that treat `//' as some sort of `network root'. Fix from Eric Blake lib/readline/complete.c - in to_print, do the right thing after stripping the trailing slash from full_pathname: // doesn't turn into /, and /// doesn't become //. Fix from Eric Blake 6/30 ---- lib/malloc/trace.c - include if it's available for a definition of size_t jobs.c - in wait_for, if a child process is marked as running but waitpid() returns -1/ECHILD (e.g., when the bash process is being traced by strace), make sure to increment c_reaped when marking the child as dead - in without_job_control, make sure to close the pgrp pipe after calling start_pipeline 7/1 --- Makefile.in - only remove pathnames.h when the other files created by running configure are removed (e.g., Makefile). Fix from William Park lib/sh/shquote.c - since backslash-newline disappears when within double quotes, don't add a backslash in front of a newline in sh_double_quote. Problem reported by William Park jobs.c - in notify_of_job_status, don't print status messages about terminated background processes unless job control is active bashhist.c - new variable, hist_last_line_pushed, set to 0 in really_add_history (used by `history -s' code) bashhist.h - new extern declaration for history -s builtins/history.def - don't remove last history entry in push_history if it was added by a call to push_history -- use hist_last_line_pushed as a sentinel and set it after adding history entry. This allows multiple calls to history -s to work right: adding all lines to the history rather than deleting all but the last. Bug reported by Matthias Schniedermeyer - pay attention to hist_last_line_pushed in expand_and_print_history() so we don't delete an entry pushed by history -s 7/4 --- print_cmd.c - fix print_arith_for_command to not print so many blanks between expressions in ((...)) command.h - new word flag: W_DQUOTE. Means word should be treated as if double quoted make_cmd.c - add W_DQUOTE to word flags in make_arith_for_expr parse.y - add W_DQUOTE to word flags for (( ... )) arithmetic commands subst.c - don't perform tilde expansion on a word with W_DQUOTE flag set - don't perform process substitution on a word with W_DQUOTE flag set arrayfunc.c - expand an array index within [...] the same way as an arithmetic expansion between (( ... )) lib/readline/input.c - use getch() instead of read() on mingw lib/readline/readline.c - add a few key bindings for the arrow keys on mingw lib/readline/rldefs.h - if on mingw, define NO_TTY_DRIVER lib/readline/rltty.c - compile in the stub functions for _rl_{disable,restore}_tty_signals if on mingw - compile in stub function for rl_restart_output on mingw - make sure enough functions and macros are defined to compile if NO_TTY_DRIVER is defined (lightly tested - builds on MacOS X, at least) 7/7 --- command.h - add a `flags' member to the PATTERN_LIST structure make_cmd.c - intialize the `flags' member of a PATTERN_LIST when it's created builtins/psize.c - protect extern declaration of errno with usual #ifdef errno configure.in, variables.c - changes for QNX 6.x 7/9 --- parse.y - fix parse_matched_pair to handle single and double quoted strings inside old-style command substitution (``) since they can each quote the ` and embedded $-expansions. Report by Eric Blake {configure,Makefile}.in - TILDE_LIB is now substituted into Makefile by configure configure.in - if configuring --with-installed-readline on cygwin, set TILDE_LIB to the empty string to avoid multiply-defined symbols. Cygwin doesn't allow undefined symbols in dynamic libraries. Report by Eric Blake 7/11 ---- input.c - in duplicate_buffered_stream, don't call free_buffered_stream if the two buffered streams share the same b_buffer object (e.g., if they had already been duplicated with a previous call). Fixes Debian bug reported by eero17@bigfoot.com 7/12 ---- shell.c - make set_shell_name more resistant to a NULL argument - in bind_args, use < instead of != when counting the arguments and making the arg list - in main(), make sure arg_index is not initialized to a value greater than argc 7/14 ---- lib/readline/display.c - in expand_prompt, don't set the location of the last invisible char if the sequence is zero length (\[\]) 7/15 ---- doc/{bash.1,bashref.texi} - document that the shell uses $TMPDIR when creating temporary files 7/20 ---- [bash-3.1-alpha1 frozen] 7/29 ---- builtins/evalstring.c - make sure that parse_and_execute saves and restores the value of loop_level, so loops in sourced scripts and eval'd strings don't mess up the shell's parser state bashline.c - change command_subst_completion_function to suppress appending any character to a unique completion, instead of a space, unless the last word in the quoted command substitution completes to a directory name. In that case we append the expected slash 8/1 --- builtins/printf.def - make sure variables are initialized if their values are tested later [bash-3.1-alpha1 updated and re-frozen] 8/2 --- variables.c - make sure to call stifle_history with an `int' instead of an intmax_t. Sometimes it makes a difference 8/3 --- [bash-3.1-alpha1 released] support/mksignames.c - add `SIGSTKFLT' (RHE3) - add `SIGXRES' (Solaris 9) 8/4 --- builtins/ulimit.def - fix typo to make `x' the right option for locks - add new options to short help synopsis variables.c - use get_variable_value instead of direct reference to value_cell in make_variable_value when appending to the current value, so references to array variables without subscripts will be equivalent to element 0 lib/readline/text.c - rewrote rl_change_case to correctly change the case of multibyte characters where appropriate 8/5 --- configure.in - remove call to obsolete macro AC_ACVERSION - remove special calls to AC_CYGWIN and AC_MINGW32; AC_CANONICAL_HOST takes care of those cases general.h - include `chartypes.h' for definition of ISALPHA - fix definitions of ABSPATH and RELPATH for cygwin - fix definition of ISDIRSEP for cygwin to allow backslash as a directory name separator 8/9 --- builtins/setattr.def - when setting a variable from the temporary environment in set_var_attribute (e.g., `LC_ALL=C export LC_ALL'), make sure to call stupidly_hack_special_variables after binding the variable in the current context builtins/printf.def - make sure to call stupidly_hack_special_variables if using `printf -v' to put formatted output in a shell variable 8/11 ---- support/shobj-conf - new variable: SHLIB_LIBPREF, prefix for shared library name (defaults to `lib' - new variable: SHLIB_DLLVERSION, used on Cygwin to set the library version number - new variable: SHLIB_DOT, separator character between library name and suffix and version information (defaults to `.') - new stanza for cygwin to generate windows-compatible dll 8/14 ---- variables.c - new special variable function for Cygwin, so the export environment is remade when HOME is changed. The environment is the only way to get information from the shell to cygwin dlls, for instanace, when bash is compiled to use an already-installed libreadline variables.h - new extern declaration for sv_home 8/15 ---- lib/readline/display.c - call init_line_structures from rl_redisplay if vis_lbreaks == 0 to avoid consequences of a poorly-timed SIGWINCH 8/16 ---- subst.c - fix logic for performing tilde expansion when in posix mode (don't rely on W_TILDEEXP flag always being set, because it won't be when expanding the RHS of assignment statement). Use W_TILDEEXP only when deciding to expand a word marked as W_ASSIGNMENT that doesn't precede a command name 8/17 ---- execute_cmd.c - in execute_function, when subshell == 1, don't short-cut by using the command contained in the group command -- if you do, any redirections attached to the group command (function) don't get executed general.h - new #define, FS_READABLE, indicates file is readable by current user findcmd.c - rewrote file_status to use S_xxx POSIX file mode bits and to add support for FS_READABLE (affects ./source and searching $PATH for scripts whose names are supplied as arguments on the command line) - change find_path_file to look for readable files -- source requires it - change find_in_path_element to do the right thing when FS_READABLE is supplied as a flag doc/bashref.texi - remove note about posix non-compliance in `.': we now require and look for readable files when searching $PATH 8/20 ---- subst.c - fix setifs to handle case where passed variable is non-zero but v->value == 0 (as in an unset local variable); treat IFS as unset in this case jobs.c - in kill_pid, if asked to killpg a process or pgrp whose pgrp is recorded as the same as the shell's, just call killpg and let the chips fall where they may -- there may be other processes in that pgrp that are not children of the shell, so killing each process in the pipeline will not do a complete job, and killpg'ing each such process will send too many signals in the majority of cases builtins/cd.def - in posix mode, pwd needs to check that the value it prints and `.' are the same file builtins/read.def - if reading input from stdin in a non-interactive shell and calling `read', call sync_buffered_stream to seek backward in the input stream if necessary (XXX - should we do this for all shell builtins?) 8/23 ---- builtins/cd.def - in posix mode, if canonicalization of the absolute pathname fails because the path length exceeds PATH_MAX, but the length of the passed (non-absolute) pathname does not, attempt the chdir, just as when not in posix mode builtins/type.def - don't have describe_command call sh_makepath if the full path found is already an absolute pathname (sh_makepath will stick $PWD onto the front of it) 8/24 ---- jobs.c - in posix mode, don't have start_job print out and indication of whether the job started by `bg' is the current or previous job - change start_job to return success if a job to be resumed in the background is already running. This means that bg won't fail when asked to bg a background job, as SUSv3/XPG6 requires - new function, init_job_stats, to zero out the global jobstats struct {jobs,nojobs}.c - change kill_pid to handle pids < -1 by killing process groups jobs.h - extern declaration for init_job_stats lib/readline/history.c - check whether or not the history list is null in remove_history builtins/history.def - delete_last_history is no longer static so fc builtin can use it builtins/fc.def - use free_history_entry in fc_replhist instead of freeing struct members individually - call delete_last_history from fc_replhist instead of using inline code - if editing (-l not specified), make sure the fc command that caused the editing is removed from the history list, as POSIX specifies builtins/kill.def - just call kill_pid with any pid argument and let it handle pids < -1 This is the only way to let kill_pid know whether a negative pid or a job spec was supplied as an argument to kill builtins/fg_bg.def - force fg_bg to return EXECUTION_SUCCESS explicitly if called by bg and start_job returns successfully - bg now returns success only if all the specified jobs were resumed successfully execute_cmd.c - call init_job_stats from initialize_subshell to zero out the global job stats structure 8/25 ---- bashline.c - change vi_edit_and_execute_command to just call vi when in posix mode, instead of checking $FCEDIT and $EDITOR lib/readline/search.c - if in vi_mode, call rl_free_undo_list in make_history_line_current to dispose of undo list accumulated while reading the search string (if this isn't done, since vi mode leaves the current history position at the entry which matched the search, the call to rl_revert_line in rl_internal_teardown will mangle the matched history entry using a bogus rl_undo_list) - call rl_free_undo_list after reading a non-incremental search string into rl_line_buffer -- that undo list should be discarded lib/readline/rlprivate.h - add UNDO_LIST * member to search context struct lib/readline/isearch.c - initialize UNDO_LIST *save_undo_list member of search context struct 8/27 ---- lib/readline/bind.c - change rl_parse_and_bind to strip whitespace from the end of a variable value assignment before calling rl_variable_bind doc/bash.1,lib/readline/doc/{rluser.texi,readline.3} - clarified the language concerning parsing values for boolean variables in assignment statements 8/28 ---- lib/sh/pathphys.c - fix small memory leak in sh_realpath reported by Eric Blake 8/31 ---- doc/bashref.texi - add additional notes to posix mode section 9/3 --- parse.y - if $'...' occurs within a ${...} parameter expansion within double quotes, don't single-quote the expanded result -- the double quotes will cause it to be expanded incorrectly 9/4 --- builtins/fc.def - if STRICT_POSIX is defined, the posix mode default for the editor to use is $FCEDIT, then ed shell.c - if STRICT_POSIX is defined, initialize `posixly_correct' to 1 config.h.in - add #undef STRICT_POSIX 9/5 --- configure.in - add new option argument, --enable-strict-posix-default, configures bash to be posix-conformant (including defaulting echo to posix conformance) by default builtins/echo.def - if STRICT_POSIX is defined, default echo to xpg-style doc/bashref.texi - describe the --enable-strict-posix-default option to configure 9/10 ---- builtins/mkbuiltins.c - change to not generate N_(""), because the translated empty string is special to GNU gettext 9/13 ---- lib/readline/complete.c - a negative value for rl_completion_query_items means to not ask lib/readline/doc/{{rltech,rluser}.texi,readline.3} - documented new semantics for rl_completion_query_items/ completion-query-items 9/14 ---- bashline.c - bind M-TAB in emacs mode to dynamic-complete-history even if the current binding is `tab-insert' (which is what it is by default), not just if it's unbound 9/15 ---- eval.c - call QUIT before calling dispose_command on current_command after the `exec_done' label. If we dispose current_command first, the longjmp might restore the value of current_command after we've disposed it, and the subsequent call to dispose_command from the DISCARD case will free memory twice 9/16 ---- lib/sh/strto[iu]max.c - make sure the function being declared is not a cpp define before defining it -- should fix problems on HP-UX 9/19 ---- Makefile.in - make sure the binaries for the tests are at the front of $PATH 9/22 ---- parse.y - new flag for parse_matched_pair: P_COMMAND, indicating that the text being parsed is a command (`...`, $(...)) - change calls to parse_matched_pair to include P_COMMAND where appropriate - if P_COMMAND flag is set and the text is unquoted, check for comments and don't try to parse embedded quoted strings if in a comment (still not exactly right yet) 9/24 ---- builtins/history.def - if running history -n, don't count these new lines as history lines for the current session if the `histappend' shell option is set. If we're just appending to the history file, the issue that caused history_lines_this_session to be recalculated doesn't apply -- the history file won't be missing any entries lib/readline/isearch.c - fix C-w handler for isearch string reader to handle multibyte chars lib/readline/rlmbutil.h - new defines for _rl_to_wupper and _rl_to_wlower lib/readline/text.c - use _rl_to_wupper and _rl_to_wlower as appropriate 9/26 ---- execute_cmd.c - in shell_execve, if the exec fails due to E2BIG or ENOMEM, just print the appropriate error message instead of checking out any interpreter specified with #! 9/30 ---- bashhist.c - make $HISTCMD available anytime remember_on_history is non-zero, which indicates that we're saving commands to the history, and let it evaluate to 1 if we're not 10/4 ---- lib/sh/snprintf.c - in floating(), make sure d != 0 before calling chkinfnan -- gcc on the version of Solaris 9 I have translates 0 to -inf on the call [bash-3.1-beta1 frozen] 10/6 ---- jobs.c - set the_pipeline to NULL right away in cleanup_the_pipeline, and dispose a copy of the pointer so we don't mess with the_pipeline while we're in the process of destroying it - block and unblock SIGCHLD around manipulating the_pipeline in cleanup_the_pipeline 10/7 ---- [bash-3.1-beta1 released] lib/readline/isearch.c - when switching directions, make sure we turn off the SF_REVERSE flag in the search context's flags word if we're going from reverse to forward i-search lib/readline/bind.c - new function, rl_variable_value, returns a string representing a bindable readline variable's value - new auxiliary function, _rl_get_string_variable_value, encapsulates everything needed to get a bindable string variable's value - rewrote rl_variable_dumper to use _rl_get_string_variable_value lib/readline/readline.h - new extern declaration for rl_variable_value lib/readline/doc/rltech.texi - documented rl_variable_value bashline.c - in command_word_completion_function, if readline sets rl_completion_found_quote, but doesn't set rl_completion_quote_character, we have an embedded quoted string or backslash-escaped character in the passed text. We need to dequote that before calling filename_completion_function. So far, this is in place only for absolute program names (those containing a `/') - in command_word_completion_function, use rl_variable_value to decide whether or not we should ignore case, and use strncasecmp instead of strncmp where appropriate 10/11 ----- builtins/fc.def - fixed a typo when using POSIX_FC_EDIT_COMMAND redir.h - new flag values for redirections: RX_INTERNAL and RX_USER (currently unused) redir.c - add_undo_redirect and add_undo_close_redirect now set RX_INTERNAL flag when making new redirects - in do_redirection_internal, only set file descriptors > 2 to CLEXEC if they're marked as RX_INTERNAL 10/12 ----- jobs.c - in wait_for_single_pid, if in posix mode, remove the waited-for pid from the list of background pids, forgetting it entirely. POSIX conformance tests test for this. lib/readline/{readline.h,vi_mode.c} - new state flag, RL_STATE_VICMDONCE, set after entering vi command mode the first time; reset on each call to readline() 10/13 ----- lib/readline/undo.c - in rl_revert_line, make sure that revert-line in vi mode leaves rl_point set to 0 no matter the state of the line buffer lib/readline/vi_mode.c - when entering vi_command mode for the first time, free any existing undo list so the previous insertions won't be undone by the `U' command. This is how POSIX.2 says `U' should work (and the test suite tests for it) lib/readline/bind.c - change rl_parse_and_bind so only `set' commands involving boolean readline variables have trailing whitespace stripped from the value string 10/16 ----- lib/glob/sm_loop.c - fix patscan() to correctly scan backslash-escaped characters 10/18 ----- lib/sh/{winsize.c,Makefile.in},{jobs,nojobs}.c,Makefile.in,externs.h - moved get_new_window_size from jobs.c/nojobs.c to new file, lib/sh/winsize.c, made function global {jobs,nojobs,sig}.c,{jobs,sig}.h - moved SIGWINCH handling code to sig.c rather than duplicate it in jobs.c and nojobs.c - call set_sigwinch_handler from sig.c code rather than job control signal initialization sig.[ch] - new variable, sigwinch_received, acts like interrupt_state for SIGWINCH, set by sigwinch_sighandler. sigwinch_sighandler no longer calls get_new_window_size parse.y - add call to get_new_window_size if sigwinch_received at top of shell_getc 10/19 ----- lib/malloc/malloc.c - to avoid orphaning memory on free if the right bucket is busy, use a new function xplit(mem, bucket) to split the block into two or more smaller ones and add those to the right bucket (appropriately marking it as busy) - audit bsplit(), bcoalesce(), and xsplit() for proper use of busy[], since they're dealing with two separate buckets 10/22 ----- subst.c - new flag for string_extract: EX_REQMATCH, means to return an error if a matching/closing character is not found before EOS - new static flag variables: extract_string_error and extract_string_fatal - change expand_word_internal to check for new error returns from string_extract and return errors if appropriate 10/23 ----- builtins/cd.def - make sure we free TDIR in change_to_directory after calling set_working_directory (which allocates new memory) and other places we short-circuit and return 10/24 ----- subst.c - modified fix from 10/22 to allow bare ` to pass through (for some backwards compatibility and more correctness) 10/27 ----- conftypes.h - make MacOS X use the RHAPSODY code that gets HOSTTYPE, et al. at build rather than configure time, to support universal binaries (fix from llattanzi@apple.com) 10/30 ----- builtins/evalstring.c - make sure we don't turn on CMD_NO_FORK in parse_and_execute if we're running a trap command on signal receipt or exit execute_cmd.c - in shell_execve, improve the error message a little bit if the interpreter name in a #! exec header ends with a ^M (as in a DOS- format file) 11/1 ---- lib/readline/vi_mode.c - fix vi-mode `r' command to leave the cursor in the right place [bash-3.1-rc1 frozen] 11/5 ---- execute_cmd.c - make sure a DEBUG trap doesn't overwrite a command string passed to make_child in execute_simple_command bashline.c - rearrange some code in bash_quote_filename so filenames with leading tildes containing spaces aren't tilde-expanded before being returned to the caller 11/6 ---- lib/readline/display.c - when deciding where to move the cursor in rl_redisplay and needing to move the cursor back after moving it vertically and compensate for invisible characters in the prompt string, make sure that _rl_last_c_pos is treated as an absolute cursor position in a multibyte locale and the wrap offset (number of invisible characters) is added explicitly when deciding how many characters to backspace 11/10 ----- lib/readline/terminal.c - _rl_set_screen_size now interprets a lines or columns argument < 0 as an indication not to change the current value 11/11 ----- lib/readline/terminal.c - new function, rl_reset_screen_size, calls _rl_get_screen_size to reset readline's idea of the terminal size - don't call _rl_get_screen_size in _rl_init_terminal_io if both _rl_screenheight and _rl_screenwidth are > 0 - don't initialize _rl_screenheight and _rl_screenwidth to 0 in _rl_init_terminal_io; let caller take care of it - set _rl_screenheight and _rl_screenwidth to 0 before calling _rl_init_terminal_io lib/readline/readline.h - new extern declaration for rl_reset_screen_size lib/readline/doc/rltech.texi - documented rl_reset_screen_size variables.c - if readline is being used, compile in a special var function for assignments to LINES and COLUMNS that calls rl_set_screen_size or rl_reset_screen_size as appropriate. Only do this in posix mode and only when STRICT_POSIX is defined at compile time - new semaphore variable, winsize_assignment, set while doing an assignment to LINES or COLUMNS - new variable, winsize_assigned, says LINES or COLUMNS was assigned to or found in the environment - if in the middle of an assignment to LINES or COLUMNS, make sh_set_lines_and_columns a no-op lib/sh/winsize.c - get_new_window_size now takes two int * arguments, to return the screen dimensions externs.h - change extern declaration for get_new_window_size {jobs,nojobs}.c, parse.y - change callers of get_new_window_size 11/12 ----- lib/readline/terminal.c - new variable, rl_prefer_env_winsize, gives LINES and COLUMNS precedence over values from the kernel when computing window size lib/readline/readline.h - extern declaration for rl_prefer_env_winsize lib/readline/doc/rltech.texi - document rl_prefer_env_winsize 11/13 ----- lib/readline/rltty.c - change rl_prep_terminal to make sure we set and reset the tty special characters in the vi insertion keymap if in vi mode. This matters if we get accept-line for the previous line while in vi command mode 11/14 ----- builtins/pushd.def - make sure any call to cd_builtin includes a leading `--' from the argument list (or constructs one) 11/16 ----- pcomplete.c - fix small memory leak in gen_wordlist_matches [bash-3.1-rc2 frozen] 11/21 ----- [bash-3.1-rc2 released] 11/23 ----- lib/readline/display.c - changes to rl_redisplay to compensate for update_line updating _rl_last_c_pos without taking invisible characters in the line into account. Important in multibyte locales where _rl_last_c_pos is an absolute cursor position - changes to _rl_move_cursor_relative to account for _rl_last_c_pos being an absolute cursor position in a multibyte character locale - rewrote _rl_move_cursor_relative to make it a little simpler 11/29 ----- lib/readline/display.c - changes to rl_redisplay and update_line for update_line to communicate upward that it took the number of invisible characters on the current line into account when modifying _rl_last_c_pos - in update_line, adjust _rl_last_c_pos by wrap_offset before calling _rl_move_cursor_relative, so we pass correct information about the true cursor position 12/1 ---- configure.in - changed release status to `release' [bash-3.1 frozen] 12/8 ---- [bash-3.1 released] 12/9 ---- doc/{bash.1,version.texi},lib/readline/doc/version.texi - remove `beta1' from man page footer and texinfo documents variables.c - make sure winsize_assignment is protected by #ifdef READLINE, so minimal shell will compile builtins/read.def - make sure error cases free memory and run any unwind-protects to avoid memory leaks 12/10 ----- execute_cmd.c - change execute_command_internal to set $PIPESTATUS for ((...)) and [[ ... ]] commands doc/{bash.1,bashref.texi,version.texi} - add documentation for ulimit -[iqx] and bump revision date 12/12 ----- parse.y - make sure parse_compound_assignment saves and restores the PST_ASSIGNOK parser state flag around its calls to read_token. Fixes bug reported by Mike Frysinger 12/13 ----- parse.y - change parse_compound_assignment to save and restore the value of last_read_token. Not sure why it was set unconditionally in the first place after parsing the complete compound assignment 12/14 ----- lib/readline/text.c - don't use return value of rl_kill_text (which always succeeds and returns the number of characters killed) in rl_delete as an indication of success or failure - ditto for return value of rl_delete_text lib/readline/readline.c - don't return the value of the called readline function as the return value from _rl_dispatch_subseq; -1 means something different to the callers (return 0 all the time to indicate that a readline function was found and dispatched). Fix from Andreas Schwab for bug in callback interface first reported by Mike Frysinger execute_cmd.c - fixed a typo in execute_case_command 12/15 ----- aclocal.m4 - add check for wctype() to BASH_CHECK_MULTIBYTE, define HAVE_WCTYPE config.h.in - add HAVE_WCTYPE #define config-bot.h - add HAVE_WCTYPE to the set of checks for HANDLE_MULTIBYTE. This should catch the deficient NetBSD multibyte support 12/16 ----- parse.y - use CTLESC instead of literal '\001' when decode_prompt_string prefixes RL_PROMPT_START_IGNORE and RL_PROMPT_END_IGNORE 12/20 ----- lib/readline/display.c - don't treat RL_PROMPT_START_IGNORE specially inside a sequence of ignored characters - keep track of the start of the current sequence of ignored characters; make sure that an empty sequence of such characters really is an empty sequence, not one that happens to end with '\001' (RL_PROMPT_START_IGNORE) 12/21 ----- subst.c - change expand_word_internal to process rest of `tilde-word' as a regular part of the word if tilde expansion leaves the tilde-word unchanged. This means that ~$USER expands to ~chet, which seems more intuitive, and is effectively what bash-3.0 did 12/23 ----- subst.c - when making a local array variable in do_compound_assignment, make sure that we don't use a variable of the same name from a previous context doc/bash.1 - documented expansions for word and patterns in case statement builtins/ulimit.def,doc/{bashref.texi,bash.1} - added new -e and -r (nice and rtprio) options to ulimit; documented them 12/26 ----- variables.c - use `hmax' instead of `num' in sv_histsize to avoid integer overflow problems with intmax_t builtins/read.def - add unwind-protect to restore rl_attempted_completion_function in case of a timeout {bashline,variables}.c - move initialization of HISTSIZE from initialization path to load_history, so it can be overridden by a value assigned in a startup file lib/readline/misc.c - add a missing `return r' so that rl_digit_loop returns a meaningful value lib/readline/{bind,callback,display,isearch,rltty,search,text,vi_mode}.c - minor cleanups to satisfy compiler warnings, mostly removing unused variables 12/27 ----- support/Makefile.in - add LIBS_FOR_BUILD support; defaults to ${LIBS} Makefile.in - add LIBS_FOR_BUILD with no default value; use when linking programs using CC_FOR_BUILD (e.g., bashversion) 12/28 ----- lib/readline/bind.c - fix rl_translate_keyseq bad translation of \M-\C-x sequences execute_cmd.c - in execute_arith_command, if the expression expands to more than one word, make sure we join the words into a single string and pass the entire thing to evalexp() expr.c - new functions: _is_arithop(c), returns true if C is a valid single- character arithmetic operator; _is_multiop(c), returns true if C is a token corresponding to a valid multi-character arithmetic operator - if we encounter a character that isn't a valid arithmetic operator, throw an error. Try to be intelligent about what type of error message to print subst.c - new function, expand_arith_string, calls expand_string_if_necessary; used where an arithmetic expression needs to be expanded subst.h - new extern declaration for expand_arith_string arrayfunc.c - in array_expand_index, call expand_arith_string to expand the subscript in a fashion consistent with other arithmetic expressions subst.c - fix parameter_brace_patsub so that we don't try to anchor the pattern at the beginning or end of the string if we're doing global replacement -- that combination doesn't doesn't make sense, and the changed behavior is compatible with ksh93 doc/{bash.1,bashref.texi} - changed description of pattern substitution to match the new semantics tests/new-exp.tests - change tests to remove all ${pat//#rep} and ${pat//%rep} expansions, since they don't mean the same thing anymore 12/29 ----- support/signames.c - new file, initialize_signames() function from old mksignames.c. This file builds the signal_names array support/mksignames.c - strip out initialize_signames(), move to signames.c. This file only writes signames.h - set up to only write a stub signames.h if CROSS_COMPILING is defined, with extern declaration for initialize_signames - if not cross compiling, #define initialize_signames to nothing Makefile.in - mksignames is now linked from mksignames.o and buildsignames.o - add rules to build signames.o, assuming we're building it as part of the shell (cross-compiling) trap.c - call initialize_signames from initialize_traps configure.in - set SIGNAMES_O to nothing (normal) or signames.o (cross-compiling), substitute into Makefile - don't set SIGNAMES_H if cross-compiling any more 12/30 ----- command.h - new word flag: W_NOPROCSUB, inhibits process substitution on a word subst.c - change expand_word_internal to suppress process substitution if the word has the W_NOPROCSUB flag shell.c - --wordexp turns on W_NOPROCSUB in addition to W_NOCOMSUB subst.c - change string_list_dollar_at and string_list_dollar_star so that MB_CUR_MAX is used to size an array only when using gcc, since gcc can handle non-constant array sizes using a mechanism like alloca. Other compilers, e.g. Sun's compiler, do not implement that extension 12/31 ----- builtins/mkbuiltins.c - when cross-compiling, don't include , since it's for the target rather than the host system. Instead, choose a reasonable set of default #defines based on a minimal POSIX system jobs.c - change find_process to handle a NULL return value from find_pipeline - return immediately from delete_job if jobs[index] is already NULL or if it has a null pipeline associated with it - in delete_job, if find_last_proc returns NULL, don't try to call bgp_delete 1/7/2006 -------- doc/bash.1 - patch from Tim Waugh to replace some literal single quotes with \(aq, the groff special character for it jobs.c - in realloc_jobs_list, make sure to zero out slots after j_lastj in the new list 1/9 --- support/mksignames.c - make sure to include to get right value of NSIG from (usually) 1/10 ---- parse.y - when calling parse_matched_pair on a $(...) command substitution, don't pass the P_DQUOTE flag so that single quotes don't get stripped from $'...' inside the command substitution. Bug report and fix from Mike Stroyan jobs.c - start maintaining true count of living children in js.c_living - call reset_current in realloc_jobs_list, since old values for current and previous job are most likely incorrect - don't allocate a new list in realloc_jobs_list if the old size and new size are the same; just compact the existing list - make sure realloc_jobs_list updates value of js.j_njobs - add some more itrace messages about non-null jobs after j_lastj in jobs array 1/11 ---- bashjmp.h - new value for second argument to longjmp: SIGEXIT. Reserved for future use 1/12 ---- jobs.c - add logic to make_child to figure out when pids wrap around - turn second argument to delete_job into flags word, added flag to prevent adding proc to bgpids list 1/13 ---- lib/readline/vi_mode.c - move code that moves forward a character out of rl_vi_append_mode into a separate function, _rl_vi_append_forward - change _rl_vi_append_mode to save `a' as the last command, so it can be redone properly - new function _rl_vi_backup, moves point back a character taking multibyte locales into account - change rl_vi_redo to handle redoing an `a' command specially -- it should be redone like `i' but after moving forward a character - change rl_vi_redo to use _rl_vi_backup to move point backward after redoing `i' or `a' jobs.c - new function, delete_old_job (pid), checks whether or not PID is in a job in the jobs list. If so, and the job is dead, it just removes the job from the list. If so, and the job is not dead, it zeros the pid in the appropriate PROCESS so pid aliasing doesn't occur - make_child calls delete_old_job to potentially remove an already-used instance of the pid just forked from the jobs list if pids have wrapped around. Finally fixes the bug reported by Tim Waugh trap.c - new define, GETORIGSIG(sig), gets the original handling for SIG and sets SIG_HARD_IGNORE if that handler is SIG_IGN - call GETORIGSIG from initialize_traps, get_original_signal, and set_signal jobs.c - in wait_for, if the original SIGINT handler is SIG_IGN, don't set the handler to wait_sigint_handler. This keeps scripts started in the background (and ignoring SIGINT) from dying due to SIGINT while they're waiting for a child to exit. Bug reported by Ingemar Nilsson lib/readline/vi_mode.c - don't save text to buffer unless undo pointer points to a record of type UNDO_INSERT; zero it out instead. This fixes bug reported by Craig Turner with redoing `ctd[ESC]' (empty insert after change to) shell.c - change set_shell_name so invocations like "-/bin/bash" are marked as login shells doc/bash.1 - add note about destroying functions with `unset -f' to the section on shell functions lib/readline/terminal.c - if readline hasn't been initialized (_rl_term_autowrap == -1, the value it's now initialized with), call _rl_init_terminal_io from _rl_set_screen_size before deciding whether or not to decrement _rl_screenwidth. Fixes bug from Mike Frysinger 1/14 ---- lib/readline/input.c - allow rl_set_keyboard_input_timeout to set the timeout to 0, for applications that want to use select() like a poll without any waiting lib/readline/doc/rltech.texi - documented valid values for timeout in rl_set_keyboard_input_timeout jobs.c - in stop_pipeline, don't have the parent shell call give_terminal_to if subshell_environment contains SUBSHELL_ASYNC (no background process should ever give the terminal to anything other than shell_pgrp) - in make_child, don't give the terminal away if subshell_environment contains SUBSHELL_ASYNC 1/15 ---- subst.c - in parameter_brace_expand, if extracting ${#varname}, only allow `}' to end the expansion, since none of the other expansions are valid. Fixes Debian bug reported by Jan Nordhorlz 1/17 ---- parse.y - in parse_matched_pair, protect all character tests with the MBTEST macro - in parse_dparen, take out extra make_word after call to alloc_word_desc (mem leak) 1/18 ---- parse.y - in parse_matched_pair, add P_ALLOWESC to flags passed to recursive parse_matched_pair call when encountering a single or double quote inside a ``-style command substitution execute_cmd.c - add call to QUIT at beginning of execute_command_internal; better responsiveness to SIGINT 1/21 ---- lib/readline/bind.c - change rl_invoking_keyseqs_in_map to honor the setting of convert-meta when listing key bindings, since if convert-meta is off, using '\M-' as the prefix for bindings in, for instance, emacs-escape-keymap, is wrong. This affects `bind -p' output - change rl_untranslate_keyseq to add '\e' instead of '\C-[' for ESC execute_cmd.c - add call to QUIT at end of execute_command 1/23 ---- lib/readline/display.c - changed two places in update_line where a check of whether the cursor is before the last invisible character in the prompt string to differentiate between the multibyte character case (where _rl_last_c_pos is a physical cursor position) and the single-byte case (where it is a buffer index). This prevents many unnecessary \r-redraw the line sequences. Reported by Dan Jacobson. 1/24 ---- quit.h - wrap QUIT macro in do...while(0) like other compound statement macros - CHECK_TERMSIG define (placeholder for now); future use will be to handle any received signals that should cause the shell to terminate (e.g., SIGHUP) {input,jobs,nojobs}.c - add calls to CHECK_TERMSIG where appropriate (reading input and waiting for children) - include quit.h if necessary 1/25 ---- parse.y - undo change that makes `)' in a compound assignment delimit a token. It messes up arithmetic expressions in assignments to `let', among other things sig.h,{jobs,nojobs,sig,trap}.c,builtins/trap.def - rename termination_unwind_protect to termsig_sighandler sig.c - split termsig_sighandler into two functions: termsig_sighandler, which runs as a signal handler and sets a flag noting that a terminating signal was received, and termsig_handler, which runs when it is `safe' to handle the signal and exit - new terminate_immediately variable, similar to interrupt_immediately - termsig_sighandler calls termsig_handler immediately if terminate_immediately is non-zero quit.h - change CHECK_TERMSIG macro to check terminating_signal and call termsig_handler if it's non-zero - add same check of terminating_signal and call to termsig_handler to QUIT macro {jobs,nojobs}.c - change call to termsig_sighandler to call termsig_handler directly, as was intended parse.y,builtins/read.def - set terminate_immediately to non-zero value when reading interactive input, as is done with interrupt_immediately 1/26 ---- doc/{bash.1,bashref.texi} - reworded the POSIX standard references to remove mention of POSIX.2 or 1003.2 -- it's all the 1003.1 standard now. Recommended by Arnold Robbins 1/27 ---- lib/readline/complete.c - move call to filename dequoting function into rl_filename_completion_function; call only if directory completion hook isn't set. This means that directory-completion-hook now needs to dequote the directory name. We don't want to dequote the directory name before calling the directory-completion-hook. Bug reported by Andrew Parker bashline.c - add necessary directory name dequoting to bash_directory_expansion and bash_directory_completion_hook lib/readline/doc/rltech.texi - add note to description of rl_directory_completion_hook that it needs to dequote the directory name even if no other expansions are performed 1/28 ---- braces.c - make sure that we skip over braces that don't start a valid matched brace expansion construct in brace_expand -- there might be a valid brace expansion after the unmatched `{' later in the string - brace_gobbler now checks that when looking for a `}' to end a brace expansion word, there is an unquoted `,' or `..' that's not inside another pair of braces. Fixes the a{b{c,d}e}f problem reported by Tim Waugh builtins/declare.def - when not in posix mode, and operating on shell functions, typeset and declare do not require their variable operands to be valid shell identifiers. The other `attribute' builtins work this way. Fixes inconsistency reported by Mike Frysinger {configure,config.h}.in - add test for setregid, define HAVE_SETREGID and HAVE_DECL_SETREGID as appropriate - add test for eaccess, define HAVE_EACCESS if found lib/sh/eaccess.c - new file, with sh_stat and sh_eaccess functions, moved from test.c - renamed old sh_eaccess as sh_stataccess, since it uses the stat(2) information to determine file accessibility - new function, sh_euidaccess, to call when uid != euid or gid != egid; temporarily swaps uid/euid and gid/egid around call to access - rewrote sh_eaccess to call eaccess, access, sh_euidaccess or sh_stataccess as appropriate. access(2) will take into account things like ACLs, read-only file systems, file flags, and so on. lib/sh/Makefile.in,Makefile.in - add necessary entries for eaccess.[co] test.c - change calls to test_stat to call sh_stat {test,general}.c - change calls to test_eaccess to call sh_eaccess externs.h - new extern declaration for sh_eaccess test.[ch] - remove test_stat and test_eaccess 1/29 ---- braces.c - make change from 1/28 dependant on CSH_BRACE_COMPAT not being defined (since old bash behavior is what csh does, defining CSH_BRACE_COMPAT will produce old bash behavior) 1/30 ---- bashline.c - last argument of bash_default_completion is now a flags word: DEFCOMP_CMDPOS (in command position) is only current value - attempt_shell_completion now computes flags before calling bash_default_completion - if no_empty_command_completion is set, bash does not attempt command word completion even if not at the beginning of the line, as long as the word to be completed is empty and start == end (catches beginning of line and all whitespace preceding point) 2/4 --- lib/readline/display.c - change _rl_make_prompt_for_search to use rl_prompt and append the search character to it, so the call to expand_prompt in rl_message will process the non-printing characters correctly. Bug reported by Mike Stroyan 2/5 --- lib/readline/display.c - fix off-by-one error when comparing against PROMPT_ENDING_INDEX, which caused a prompt with invisible characters to be redrawn one extra time in a multibyte locale. Change from <= to < fixes multibyte locale, but I added 1 to single-byte definition of PROMPT_ENDING_INDEX (worth checking) to compensate. Bug reported by Egmont Koblinger 2/8 --- lib/readline/terminal.c - call _emx_get_screensize with wr, wc like ioctl code for consistency - new function, _win_get_screensize, gets screen dimensions using standard Windows API for mingw32 (code from Denis Pilat) - call _win_get_screensize from _rl_get_screen_size on mingw32 lib/readline/rlconf.h - define SYS_INPUTRC (/etc/inputrc) as system-wide default inputrc filename support/shobj-conf - changes to make loadable builtins work on MacOS X 10.[34] builtins/pushd.def - changes to make it work as a loadable builtin compiled with gcc4 2/9 --- lib/readline/bind.c - add SYS_INPUTRC as last-ditch default (if DEFAULT_INPUTRC does not exist or can't be read) in rl_read_init_file lib/readline/doc/rluser.texi - add description of /etc/inputrc as ultimate default startup file 2/10 ---- lib/readline/bind.c - fix problem with rl_function_of_keyseq that returns a non-keymap bound to a portion of the passed key sequence without processing the entire thing. We can bind maps with existing non-map functions using the ANYOTHERKEY binding code. variables.c - shells running in posix mode do not set $HOME, as POSIX apparently requires 2/15 ---- braces.c - mkseq() now takes the increment as an argument; changed callers 2/16 ---- builtins/hash.def - print `hash table empty' message to stdout instead of stderr 2/17 ---- lib/readline/readline.c - when resetting rl_prompt in rl_set_prompt, make sure rl_display_prompt is set when the function returns 2/18 ---- lib/readline/display.c - further fixes to _rl_make_prompt_for_search from Eric Blake to deal with multiple calls to expand_prompt 2/21 ---- builtins/hash.def - don't print `hash table empty' message in posix mode 2/27 ---- lib/glob/sm_loop.c - change extmatch() to turn off FNM_PERIOD in flags passed to recursive calls to gmatch() when calling it with a substring after the start of the string it receives. Changed `+', `*', `?, `@', and `!' cases to do the right thing. Fixes bug reported by Benoit Vila braces.c - add QUIT; statements to mkseq to make large sequence generation interruptible 2/28 ---- lib/glob/glob.c - initialize nalloca in glob_vector 3/1 --- lib/glob/glob.c - in glob_vector, when freeing up the linked list after some error, make sure to set `tmplink' to 0 if `firstlink' is set to 0, else we get multiple-free errors 3/5 --- trap.c - inheritance of the DEBUG, RETURN, and ERR traps is now dependent only on the `functrace' and `errtrace' shell options, as the documentation says, rather than on whether or not the shell is in debugging mode. Reported by Philip Susi parse.y - in parse_matched_pair, don't recursively parse ${...} or other ${...} constructs inside `` - in parse_matched_pair, remove special code that recursively parses quoted strings inside `` constructs. For Bourne shell compatibility 3/6 --- builtins/pushd.def - let get_directory_stack take take an `int flags' argument and convert $HOME to ~ if flags&1 is non-zero builtins/common.h - change extern declaration for get_directory_stack variables.c - call get_directory_stack with an arg of 0 to inhibit converting $HOME to ~ in the result. Fixes cd ${DIRSTACK[1]} problem reported by Len Lattanzi (cd fails because the tildes won't be expanded after variable expansion) jobs.c - changed hangup_all_jobs slightly so stopped jobs marked J_NOHUP won't get a SIGCONT general.c - changed check_binary_file() to check for a NUL byte instead of a non-printable character. Might at some point want to check entire (possibly multibyte) characters instead of just bytes. Hint from ksh via David Korn 3/7 --- builtins/reserved.def - changed runs of spaces to tabs in variables help text to make indentation better when displayed builtins/mkbuiltins.c - changes to avoid the annoying extra space that keeps gettext from being passed an empty string 3/9 --- lib/glob/glob.c - make sure globbing is interrupted if the shell receives a terminating signal 3/14 ---- lib/readline/search.c - call rl_message with format argument of "%" in _rl_nsearch_init to avoid `%' characters in the prompt string from being interpreted as format specifiers to vsnprintf/vsprintf 3/19 ---- parse.y, eval.c, input.h - change execute_prompt_command to execute_variable_command; takes the variable name as a new second argument 3/25 ---- bashline.c - command_word_completion_function keeps track of when it's searching $PATH and doesn't return directory names as matches in that case. Problem reported by Pascal Terjan - command_word_completion_function returns what it's passed as a possible match if it's the name of a directory in the current directory (only non-absolute pathnames are so tested). 3/27 ---- subst.c - expand_arith_string takes a new argument: quoted. Either 0 (outside subst.c) or Q_DOUBLE_QUOTES (substitution functions); changed callers subst.h - changed extern declaration for expand_arith_string arrayfunc.c - changed call to expand_arith_string in array_expand_index 3/31 ---- lib/readline/histfile.c - change read_history_range to allow windows-like \r\n line endings execute_cmd.c - add new variable, line_number_for_err_trap, currently set but not used 4/2 --- lib/sh/strtrans.c - add code to echo -e and echo with xpg_echo enabled to require a leading 0 to specify octal constants 4/3 --- subst.c - slight change to wcsdup() replacement: use memcpy instead of wcscpy parse.y - before turning on W_COMPASSIGN, make sure the final character in the token is a `(' (avoids problems with things like a=(4*3)/2) 4/4 --- lib/sh/snprintf.c - in number() and lnumber(), turn off PF_ZEROPAD if explicit precision supplied in format - change number() and lnumber() to correctly implement zero-padding specified by a non-zero `.precision' part of the format subst.c - new flag for extract_delimited_string: EX_COMMAND. For $(...), so we can do things like skip over delimiters in comments. Added to appropriate callers - changes to extract_delimited_string to skip over shell comments when extracting a command for $(...) (EX_COMMAND is contained in the flags argument) 4/5 --- subst.c - first argument to skip_single_quoted is now a const char * - new function, chk_arithsub, checks for valid arithmetic expressions by balancing parentheses. Fix based on a patch from Len Lattanzi 4/6 --- {configure,config.h}.in - add separate test for isnan in libc, instead of piggybacking on isinf-in-libc test lib/sh/snprintf.c - separate the isnan replacement function so it's guarded by its own HAVE_ISNAN_IN_LIBC define lib/sh/wcsdup.c - new file, contains replacement wcsdup library function from subst.c with change back to using wcscpy Makefile.in,lib/sh/Makefile.in - make sure wcsdup.c is compiled and linked in subst.c - wcsdup now found in libsh; removed static definition 4/10 ---- lib/readline/callback.c - loop over body of rl_callback_read_char as long as there is additional input rather than just calling readline_internal_char, which does not handle multi-character key sequences or escape-prefixed chars lib/readline/macro.c - make sure we turn off RL_STATE_MACROINPUT when the macro stack is empty if we are reading additional input with RL_STATE_MOREINPUT support/shobj-conf - Mac OS X no longer likes the `-bundle' option to gcc when creating a dynamic shared library 4/11 ---- lib/tilde/tilde.c - don't try to dereference user_entry if HAVE_GETPWENT isn't defined lib/readline/input.c - make sure chars_avail is not used without being assigned a value in rl_gather_tyi - use _kbhit() to check for available input on Windows consoles, in rl_gather_tyi and _rl_input_available 4/21 ---- lib/readline/display.c - calculate (in expand_prompt) and keep track of length of local_prompt in local_prompt_len; use where appropriate - when using o_pos to check whether or not we need to adjust _rl_last_c_pos after calling update_line, assume that it's correct (a buffer index in non-multibyte locales and a cursor position in multibyte locales) and adjust with wrap_offset as appropriate - in update_line, set cpos_adjusted to 1 after calling _rl_move_cursor_relative to move to the end of the displayed prompt string - in _rl_move_cursor_relative, check that the multibyte display position is after the last invisible character in the prompt string before offsetting it by the number of invisible characters in the prompt (woff) 4/26 ---- lib/readline/doc/{rluser.texi,readline.3} - make sure to note that key bindings don't allow any whitespace between the key name or sequence to be bound and the colon 4/28 ---- lib/readline/display.c - in update_line, make sure we compare _rl_last_c_pos as strictly less than PROMPT_ENDING_INDEX, since it's 0-based, to avoid multiple prompt redraws 5/4 --- parse.y - in decode_prompt_string, only prefix the expansion of \[ or \] with CTLESC if the corresponding readline escape character is CTLESC (coincidentally the same as \[) or CTLNUL. Bug report sent by Mike Frysinger prompted the discovery aclocal.m4 - slight change to test for /dev/fd to compensate for a linux failing; suggested by Mike Frysinger 5/9 --- arrayfunc.c - broke assign_array_var_from_string into two functions: expand_compound_array_assignment and assign_compound_array_list; assign_array_var_from_string just calls those functions now arrayfunc.h - new extern declarations for expand_compound_array_assignment and assign_compound_array_list subst.c - in do_compound_assignment, call expand_compound_array_assignment before creating the local variable so a previous inherited value can be used when expanding the rhs of the compound assignment statement 5/11 ---- doc/{bash.1,bashref.texi} - clarifed `trap' description to make it clear that trapped signals that are not set to SIG_IGN are reset when a subshell is created 5/18 ---- locale.c - change reset_locale_vars to call setlocale (LC_ALL, "") if LANG is unset or NULL - if LANG is unset or NULL, reset the export environment before calling setlocale in reset_locale_vars, and trust that it will change the environment setlocale() inspects 5/21 ---- lib/readline/history.c - new function, HIST_ENTRY *alloc_history_entry (char *string, char *ts); creates a new history entry with text STRING and timestamp TS (both of which may be NULL) - new function, HIST_ENTRY *copy_history_entry (HIST_ENTRY *hist), which copies the line and timestamp entries to new memory but just copies the data member, since that's an opaque pointer - new function, void replace_history_data (int which, histdata_t *old, histdata_t *new) which replaces the `data' member of specified history entries with NEW, as long as it is OLD. WHICH says which history entries to modify - add calls to replace_history_data in rl_free_undo_list and rl_do_undo lib/readline/undo.c - new function, alloc_undo_entry (enum undo_code what, int start, int end, char *text) takes care of allocating and populating a struct for an individual undo list entry - new function: _rl_copy_undo_entry(UNDO_LIST *entry) - new function: _rl_copy_undo_list(UNDO_LIST *head) lib/readline/rlprivate.h - new extern declarations for _rl_copy_undo_{entry,list} execute_cmd.c - change execute_cond_node so that quoting the rhs of the =~ operator forces string matching, like the == and != operators 5/23 ---- redir.c - add_undo_redirect now takes as an additional argument the type of redirection we're trying to undo - don't add a "preservation" redirection for fds > SHELL_FD_BASE if the redirection is closing the fd 5/24 ---- subst.c - make sure that parameter_brace_substring leaves this_command_name set to either NULL or its previous value after setting it so that arithmetic evaluation errors while expanding substring values contain meaningful information 6/9 --- execute_cmd.c - make sure that SUBSHELL_ASYNC and SUBSHELL_PIPE are set as flag bits in subshell_environment, rather than setting only a single value - change execute_subshell_builtin_or_function to give the `return' builtin a place to longjmp to when executed in a subshell or pipeline (mostly as the last command in a pipeline). Bug reported by Oleg Verych - in execute_simple_command, make sure to call execute_disk_command with the_printed_command_except_trap to keep DEBUG trap command strings from overwriting the command strings associated with jobs and printed in job control messages. Bug reported by Daniel Kahn Gillmor [bash-3.2-alpha frozen] 6/22 ---- syntax.h - add new CBLANK (for [:blank:] class) flag value for syntax table and shellblank(c) character test macro mksyntax.c - add support for setting CBLANK flag in the syntax table depending on whether or not isblank(x) returns true for character x locale.c - change locale_setblanks to set or unset CBLANK flag for each character when locale changes parse.y - change call to whitespace(c) in lexical analyzer (read_token()) to call shellblank(c) instead, so locale-specific blank characters are treated as white space. Fixes bug reported by Serge van deb Boom print_cmd.c - when printing redirections, add a space between <, >, and <> and the following word, to avoid conflicts with process substitution. Bug reported by Ittay Dror 6/26 ---- configure.in - set CROSS_COMPILE to the empty string by default, so we don't inherit a random value from the environment. Bug reported by Lee Revell 6/29 ---- lib/glob/xmbsrtowcs.c - make sure destp is non-null before assigning a 0 to *destp in xdupmbstowcs. Fix from Louiwa Salem execute_cmd.c - fix execute_in_subshell to make sure asynchronous isn't set to 0 before subshell_environment is set appropriately and setup_async_signals is run. Based on report by Louiwa Salem lib/readline/bind.c - in rl_generic_bind(), make sure that the keys array is freed before an error return. Fix from Louiwa Salem 7/1 --- builtins/read.def - make sure all editing code is protected with #ifdef READLINE, esp. unwind-protect that restores the default completion function lib/readline/display.c - make sure to set local_prompt_len in rl_message() [in bash-3.2-alpha] 7/5 --- builtins/printf.def - add more of echo's write error handling to printf. Suggested by martin.wilck@fujitsu-siemens.com 7/7 --- lib/readline/display.c - save and restore local_prompt_len in rl_{save,restore}_prompt [in bash-3.2-alpha] 7/8 --- [bash-3.2-alpha released] 7/9 --- lib/readline/display.c - make sure that _rl_move_cursor_relative sets cpos_adjusted when it offsets `dpos' by wrap_offset in a multi-byte locale. Bug reported by Andreas Schwab and Egmont Koblinger subst.c - make sure that the call to mbstowcs in string_extract_verbatim is passed a string with enough space for the closing NUL. Reported by Andreas Schwab 7/18 ---- lib/readline/{display,terminal}.c - remove #ifdefs for HACK_TERMCAP_MOTION so we can use _rl_term_forward_char in the redisplay code unconditionally lib/readline/rlprivate.h - new extern declaration for _rl_term_forward_char lib/readline/display.c - in _rl_move_cursor_relative, use `dpos' instead of `new' when deciding whether or not a CR is faster than moving the cursor from its current position - in _rl_move_cursor_relative, we can use _rl_term_forward_char to move the cursor forward in a multibyte locale, if it's available. Since that function doesn't have a handle on where the cursor is in the display buffer, it has to output a cr and print all the data. Fixes rest of problem reported by Egmont Koblinger - change variable denoting the position of the cursor in the line buffer from c_pos (variable local to rl_redisplay) to cpos_buffer_position (variable local to file) for future use by other functions 7/25 ---- lib/malloc/{stats,table}.h - include for prototypes for memset, strlen lib/termcap/{termcap,tparam}.c - include and provide macro replacement for bcopy if necessary 7/27 ---- lib/readline/histexpand.c - add support for `<<<' here-string redirection operator to history_tokenize_word. Bug reported by agriffis@gentoo.org externs.h - don't add prototype for strerror() if HAVE_STRERROR defined 7/29 ---- subst.c - in list_string, use `string' instead of `s' -- s is not initialized 8/9 --- subst.c - fix parameter_brace_expand to set W_HASQUOTEDNULL in the WORD_DESC it returns if the result of parameter_brace_substring is a quoted null ("\177"). Fixes bug reported by Igor Peshansky 8/16 ---- lib/readline/readline.h - new #define, READERR, intended to be used to denote read/input errors lib/readline/input.c - in rl_getc, if read() returns an error other than EINTR (after the EWOULDBLOCK/EAGAIN cases are handled), return READERR rather than converting return value to EOF if readline is reading a top-level command (RL_STATE_READCMD) lib/readline/readline.c - if rl_read_key returns READERR to readline_internal_char[loop], abort as if it had read EOF on an empty line, without any conversion to newline, which would cause a partial line to be executed. This fixes the bug reported by Mathieu Bonnet aclocal.m4 - when testing for validity of /dev/fd/3, use /dev/null instead of standard input, since the standard input fails with linux and `su'. Bug reported by Greg Shafer 8/17 ---- Makefile.in - switch the TAGS and tags targets so TAGS is the output of `etags' and tags is the output of `ctags'. Suggested by Masatake YAMATO 8/25 ---- execute_cmd.c - change code to match documentation: set BASH_COMMAND (which takes its value from the_printed_command_except_trap) only when not running a trap. Rocky says the debugger is ok with this, and this is what his original diffs did 8/29 ---- variables.c - change set_if_not to create shell_variables if it is NULL, since -o invocation options can cause variables to be set before the environment is scanned [bash-3.2-beta frozen] 9/5 --- [bash-3.2-beta released] 9/8 --- variables.c - change dispose_used_env_vars to call maybe_make_export_env immediately if we're disposing a temporary environment, since `environ' points to the export environment and getenv() will use that on systems that don't allow getenv() to be replaced. This could cause the temporary environment to affect the shell. Bug reported by Vasco Pedro builtins/echo.def,doc/{bash.1,bashref.texi} - clarify that `echo -e' and echo when the `xpg_echo' shell option is enabled require the \0 to precede any octal constant to be expanded. Reported by Vasco Pedro 9/12 ---- builtins/printf.def - make sure `%q' format specifier outputs '' for empty string arguments Bug reported by Egmont Koblinger make_cmd.c - change make_here_document to echo lines in here-doc if set -v has been executed. Reported by Eduardo Ochs aclocal.m4 - change BASH_CHECK_MULTIBYTE: o replace check for wctomb with check for wcrtomb o add checks for wcscoll, iswctype, iswupper, iswlower, towupper, towlower o add call to AC_FUNC_MBRTOWC to check for mbrtowc and mbstate_t define HAVE_MBSTATE_T manually o add checks for wchar_t, wctype_t, wint_t config.h.in - add defines for wcscoll, iswctype, iswupper, iswlower, towupper, towlower functions - replace define for wctomb with one for wcrtomb - add defines for wchar_t, wint_t, wctype_t types config-bot.h, lib/readline/rlmbutil.h - add check for HAVE_LOCALE_H before defining HANDLE_MULTIBYTE - add checks for: ISWCTYPE, ISWLOWER, ISWUPPER, TOWLOWER, TOWUPPER - add checks for: WCTYPE_T, WCHAR_T, WCTYPE_T 9/13 ---- lib/readline/display.c - when displaying prompts longer than the screenwidth in rl_redisplay, and looking for the index of the last character whose buffer index is <= the screen width to set up the inv_lbreaks array, make sure to catch the case where the index == the screen width (an off-by-one error occurs otherwise with prompts one character longer than the screen width). Bug reported by Alexey Toptygin configure.in - change DEBUGGER_START_FILE to start with ${ac_default_prefix}/share, like bashdb installs itself. Reported by Nick Brown 9/14 ---- lib/readline/display.c - make multibyte code that computes the buffer indices of line breaks for a multi-line prompt dependent on MB_CUR_MAX, so we don't take the function call hit unless we're in a locale that can have multibyte characters 9/19 ---- subst.c - make dequote_list extern so other parts of the shell can use it subst.h - extern declaration for dequote_list builtins/read.def - call dequote_list before assigning words read to array variable if we saw an escape character. Old code left spurious CTLESCs in the string after processing backslashes. Bug reported by Daniel Dawson 9/21 ---- [bash-3.2 frozen] 10/9 ---- support/shobj-coonf - change -fpic to -fPIC for FreeBSD systems (needed for SPARC at least) 10/11 ----- [bash-3.2 released] 10/12 ----- parse.y - change parse_matched_pair to make sure `` command substitution does not check for shell comments while parsing. Bug reported against bash-3.2 by Greg Schaefer 10/14 ----- parse.y - add new parser_state flag: PST_REGEXP; means we are parsing a regular expression following the =~ conditional operator - cond_node sets PST_REGEXP after reading the `=~' operator - change read_token to call read_token_word immediately if the PST_REGEXP bit is set in parser_state - change read_token_word to skip over `(' and `|' if PST_REGEXP is set, since those characters are legitimate regexp chars (but still parse matched pairs of parens) 10/16 ----- builtins/ulimit.def - add -e and -r to $SHORT_DOC usage string po/ru.po - fix encoding; Russian text in the file is actually encoded in KOI8-R 10/23 ----- shell.c - make sure that the call to move_to_high_fd in open_shell_script passes 1 for the `check_new' parameter so open high file descriptors don't get closed and reused. Bug reported by Mike Stroyan doc/bashref.texi - fixes for typos and misspellings sent in by Brian Gough 10/24 ----- support/shobj-conf - make netbsd shared library creation like openbsd's until I hear differently (called using `gcc -shared') 10/26 ----- subst.c - fix bug in parameter_brace_patsub so if the first character of the expanded pattern is a `/', it is not taken as a global replacement specifier. Bug reported on forums.nekochan.net 10/27 ----- builtins/printf.def - if we need an extern declaration for asprintf, make sure we include stdarg.h or varargs.h, whichever is appropriate - if we do not have asprintf, add an extern declaration using stdarg format. This fixes the bugs with %G on IRIX reported by Matthew Woehlke and Stuart Shelton lib/sh/snprintf.c - add note to not call log_10 with 0 argument -- we don't want to do what real log10 does (-infinity/raise divide-by-zero exception) - make sure numtoa (used by dtoa) takes the precision into account when computing the fractional part with an argument of `0.0' - make sure `g' and `G' formats don't print radix char if there are no characters to be printed after it (change to floating()) - change callers of log_10 (exponent, 'g' and 'G' cases in vsnprintf_internal) to not call it with 0 for argument. This fixes the hang reported on IRIX by Matthew Woehlke and Stuart Shelton 10/28 ----- builtins/{caller,pushd}.def - changed longdoc strings in loadable builtin section to be single strings, as put in the build directory builtins.c file, to aid translators 11/1 ---- execute_cmd.c - reset subshell_environment to 0 after make_child() call in execute_null_command. Fix provided by Roy Marples 11/7 ---- lib/tilde/tilde.c lib/readline/{util,undo,callback,input,isearch,kill}.c - make sure that memory allocated with xmalloc is freed with xfree 11/9 ---- lib/readline/display.c - make sure that _rl_redisplay_after_sigwinch clears the last displayed line instead of the current line (instead of assuming that the cursor is on the last line). Fixes bug reported by Egmont Koblinger 11/10 ----- lib/readline/display.c - make sure that _rl_col_width is never called with MB_CUR_MAX == 1, since it doesn't count invisible characters and they are not compensated for. Added a warning in _rl_col_width if called when MB_CUR_MAX == 1. Bug reported and solution suggested by Eric Blake 11/11 ----- lib/readline/display.c - make sure _rl_wrapped_line is initialized to inv_lbsize int chars. inv_lbsize and vis_lbsize are the same at that point, but it makes the intent clearer. Fix from jan.kratochvil@redhat.com. - in rl_redisplay, make sure we call memset on _rl_wrapped_line with its full initialized size: inv_lbsize*sizeof(int). Fix from jan.kratochvil@redhat.com. - wrap the invisible and visible line variables and _rl_wrapped_line into line_state structures, which can be swapped more efficiently. Have to watch the wrapped_line field, since there's now one for each struct. Changes from jan.kratochvil@redhat.com. lib/readline/complete.c - in stat_char, check for `//server' on cygwin and return `/', since it will always behave as a directory. Fix from Eric Blake lib/readline/histfile.c - Cygwin's mmap() works in recent versions, so don't #undef HAVE_MMAP. Recommendation from Eric Blake lib/readline/rlwinsize.h - make sure tcflow() is defined on SCO Unix. Fix from William Bader aclocal.m4 - add check for localeconv to AM_INTL_SUBDIR macro config.h.in - add HAVE_LOCALECONV lib/sh/snprintf.c - add check for HAVE_LOCALECONV for GETLOCALEDATA macro general.[ch] - first argument to legal_number is now `const char *' 11/14 ----- lib/readline/{readline,rlprivate}.h - move rl_display_prompt declaration from rlprivate.h to readline.h lib/readline/util.h - new function: rl_free(void *mem), for use by users of readline dlls on Windows lib/readline/readline.h - new extern declaration for rl_free lib/readline/doc/rltech.texi - document rl_free and rl_display_prompt for use by application writers 11/15 ----- aclocal.m4 - change tests for /dev/fd and /dev/stdin to use constructs of the form (exec test ... ) instead of test ... to avoid bash's /dev/fd and /dev/stdin emulation 11/16 ----- jobs.c - in delete_job, reset_current was being called before the job slot was cleared -- moved after job_slots[job] was set to NULL. Fixes bug reported by Dan Jacobson 11/19 ----- findcmd.c - when the checkhash option is set, fix the check for the hashed pathname being an existing executable file. Old code required a hash table deletion and re-addition. Bug reported by Linda Walsh 11/21 ----- subst.c - in pos_params, handle case of `start' == 0 by making the list of positional parameters begin with $0 - in parameter_brace_substring, increment `len' if start == 0, sicne we will be adding $0 to the beginning of the list when we process it doc/{bash.1,bashref.texi} - document new behavior of `0' offset when using substring expansion with the positional parameters support/shobj-conf - changes to shared object creation for loadable builtins on Mac OS X 10.4 to use libtool instead of ld by specifying -dynamiclib argument and changing options to be appropriate for libtool. This winds up creating a dynamic shared library instead of an executable 11/24 ----- {jobs,nojobs}.c - don't set last_asynchronous_pid to the child's pid in the child for asynchronous jobs (for compatibility -- all other posix shells seem to do it this way). This means that (echo $! )& echo $! should display two different pids. Fix from discussion on the austin-group-l list builtins/mkbuiltins.c - change builtins.c file generation so short doc strings are marked for gettext and available for subsequent translation. Suggestion by Benno Schulenberg builtins/{bind,cd,hash,inlib,printf,pushd,test,times,ulimit}.def lib/malloc/malloc.c {shell,subst}.c - fix a few strings that were not marked as translatable. Fix from Benno Schulenberg lib/readline/misc.c - new function, _rl_revert_all_lines(void). Goes through history, reverting all entries to their initial state by undoing any undo lists. lib/readline/rlprivate.h - extern declaration for _rl_revert_all_lines rldefs.h - add #undef HAVE_STRCOLL if STRCOLL_BROKEN is defined, prep to move from config.h.in. Problem reported by Valerly Ushakov 11/25 ----- lib/readline/readline.c - call _rl_revert_all_lines from readline_internal_teardown if the variable _rl_revert_all_at_newline is non-zero - declare _rl_revert_all_lines initially 0 11/27 ----- doc/{bash.1,bashref.texi} - make sure to be explicit that `typeset +r' cannot remove the readonly attribute from a variable 11/28 ----- lib/sh/zmapfd.c - new file, implements zmapfd(), which takes a file and returns its contents in a string externs.h - extern declaration for zmapfd 11/29 ----- builtins/evalfile.c - in _evalfile, use zmapfd to read the contents of the file into a string, rather than using the size reported by stat and reading that many characters, if the file is not a regular file (for things like named pipes, stat reports the size as 0) 12/3 ---- lib/sh/snprintf.c - make sure number() sets the FL_UNSIGNED flag for %x and %X, so fmtulong treats them as unsigned numbers. Fixes bug reported by James Botte 12/13 ----- lib/readline/util.c - new function, _rl_ttymsg, for internal warning messages -- does redisplay after printing message - new function, _rl_errmsg, for internal warning/error messages -- does not do redisplay after printing message lib/readline/rlprivate.h - new extern declaration for _rl_ttymsg, _rl_errmsg lib/readline/{bind,callback,complete,display,rltty}.c - use _rl_ttymsg/_rl_errmsg instead of direct writes to stderr lib/sh/tmpfile.c - in get_tmpdir(), make sure that $TMPDIR names a writable directory; otherwise skip it. This catches names longer than PATH_MAX, but in case it doesn't test that the length does not exceed PATH_MAX. Fixes heap overrun bug reported by Eric Blake 12/16 ----- builtin/{set,declare,shopt,trap,wait,bind,complete,enable,fc,history,read,setattr}.def doc/{bash.1,bashref.texi} - improvements and clarifications to the help text associated with several builtins, in some cases bringing them into line with the man page text. From Benno Schulenberg doc/{bash.1,bashref.texi} - add `E' and `T' to the synopsis of the set builtin. From Benno Schulenberg builtins/{break,exit,fg_bg,hash,jobs,type,ulimit}.def builtins/{common,evalfile}.c {error,expr,jobs,mksyntax,nojobs,shell,subst,version,siglist}.c - add gettextizing marks to untranslated strings From Benno Schulenberg 12/19 ----- builtins/common.c - change display_signal_list (used by `trap -l' and `kill -l') to use five columns instead of 4 to display signal names builtins/help.def - use the true terminal width instead of assuming 80 when displaying help topics, leaving two characters of whitespace between horizontal descriptions instead of 1 - change to print in columns with entries sorted down rather than across (that is, like `ls' rather than `ls -x'). Change inspired by Benno Schulenberg jobs.h - give values to the JOB_STATE enumerations so they can be used as bitmasks, too 12/22 ----- doc/{bash.1,bashref.texi} - change description of `set' to make it clearer that you can use `+' to turn off options - clarify in the description of word splitting that sequences of IFS whitespace at the beginning or end of the string are ignored 12/26 ----- doc/bashref.texi - move `shopt' builtin to its own section; change internal references from `Bash Builtins' to the new shopt builtin - new section for builtins that modify shell behavior in `Shell Builtin Commands'; move set and shopt to new section. Changes inspired by Benno Schulenberg {redir,subst}.c - add MT_USETMPDIR flag to calls to sh_mktmpfd and sh_mktmpname. Bug reported by Eric Blake {configure,Makefile}.in - changes so that the pathname for DEBUGGER_START_FILE is substituted into pathnames.h at make time (allowing more flexibility in setting `prefix' or `datadir') instead of at configure time. Suggested by Nick Brown shell.c - declaration for have_devfd; initialized from HAVE_DEV_FD - declaration for check_jobs_at_exit; initialized to 0 - declaration for autocd; initialized to 0 variables.c - new dynamic variable, BASHPID, always set from return value from getpid() (changes even when $$ doesn't change). Idea from Bruce Korb builtins/exit.def - if check_jobs_at_exit is non-zero, list jobs if there are any stopped or running background jobs; don't exit shell if any running jobs execute_cmd.c - in execute_simple_command, if the first word of a simple command is a directory name (after looking for builtins, so `.' isn't caught) that isn't found in $PATH, and `autocd' is non-zero, prefix a "cd" to the command words builtins/shopt.def - new `checkjobs' option, changes value of check_jobs_at_exit - new `autocd' option, changes value of autocd pcomplete.c - add COMP_TYPE, set to rl_completion_type, to list of variables set by bind_compfunc_variables and unset by unbind_compfunc_variables doc/{bash.1,bashref.texi} - document BASHPID - document new shopt `checkjobs' option - document new shopt `autocd' option - document COMP_TYPE completion variable 12/29 ----- aclocal.m4 - in BASH_SYS_SIGLIST, check HAVE_DECL_SYS_SIGLIST instead of the obsolete and no-longer-supported SYS_SIGLIST_DECLARED 12/30 ----- lib/readline/vi_mode.c - add ` (backquote) to the list of vi motion characters - in rl_vi_delete_to, rl_vi_change_to, and rl_vi_yank_to, don't delete character under the cursor if the motion command moves the cursor backward, so add F and T to the commands that don't cause the mark to be adjusted - add ` to the characters that don't cause the mark to be adjusted when used as a motion command, since it's defined to behave that way - when a motion character that may adjust the mark moves point backward, don't adjust the mark so the character under the cursor isn't deleted lib/readline/complete.c - add variable rl_sort_completion_matches; allows application to inhibit match list sorting - add variable rl_completion_invoking_key; allows applications to discover the key that invoked rl_complete or rl_menu_complete lib/readline/readline.h - extern declarations for rl_completion_invoking_key and rl_sort_completion_matches lib/readline/doc/rltech.texi - documented rl_completion_invoking_key and rl_sort_completion_matches pcomplete.c - export variable COMP_KEY to completion functions; initialized from rl_completion_invoking_key; unset along with rest of completion variables doc/{bash.1,bashref.texi},lib/readline/doc/rluser.texi - document COMP_KEY [many files] - changes to make variables and function parameters `const' for better text sharing. Changes originally from Andreas Mohr 1/4/2007 -------- lib/intl/Makefile.in - use cmp before copying libgnuintl.h to libintl.h -- maybe save a few rebuilds lib/builtins/Makefile - fixes to build LIBINTL_H if necessary, dependency on this for mkbuiltins.o prevented `make -j 6' from working correctly 1/8 --- subst.c - new function, fifos_pending(), returns the count of FIFOs in fifo_list (process substitution) subst.h - extern declaration for fifos_pending() execute_cmd.c - in execute_simple_command, if CMD_NO_FORK is set before we call execute_disk_command, make sure there are no FIFOs in the expanded words (from process substitution) and turn off CMD_NO_FORK if there are, so they can get unlinked when the command finishes 1/10 ---- subst.c - read_comsub now takes a flags parameter and returns appropriate W_* flags in it - command_substitute now returns a WORD_DESC *, with the string it used to return as the `word' and `flags' filled in appropriately subst.h - changed extern declaration for command_substitute {pcomplete,subst}.c - changed callers of command_substitute appropriately subst.c - string_extract_verbatim now takes an additional int flags argument; changed callers 1/11 ---- support/texi2html - fix problem that caused index links to not be generated if the first index node had a name different than the node name doc/bashref.texi - encapsulated all indexes into a single `Indexes' appendix; works around bug fixed in texi2html 1/12 ---- subst.c - add call to sv_histtimefmt in initialize_variables so HISTTIMEFORMAT from the environment is honored. Fix from Ark Submedes (heh) lib/readline/histfile.c - make sure that the first character following the history comment character at the beginning of a line is a digit before interpreting it as a timestamp for the previous line doc/{bash.1,bashref.texi},lib/readline/doc/hsuser.texi - added detail to make it clear exactly how history timestamps are saved to and read from the history file subst.c - change quote_escapes to add CTLESC before spaces if IFS is null, just in case we have to split on literal spaces later on (e.g., in case of unquoted $@). Corresponding changes to dequote_escapes. Fixes a couple of problems reported by Brett Stahlman 1/14 ---- subst.c - make same change to read_comsub to add CTLESC before ' ' if $IFS is null, since we will split on literal spaces later 1/15 ---- array.c - new function, array_quote_escapes (ARRAY *a), calls quote_escapes on each element of the array in the same way array_quote calls quote_string - call array_quote_escapes if match is not quoted in array_patsub - array_slice is now used, so remove the #ifdef INCLUDE_UNUSED define - change structure of array_subrange to call array_slice to create a new array with the desired subset of elements, then call array_quote or array_quote_escapes as necessary, like array_patsub. Convert to a string by calling array_to_string on the sliced-out array array.h - new extern declaration for array_quote_escapes subst.c - since array_patsub now calls quote_escapes as necessary, callers don't need to call it after array_patsub returns. Fixes first bug reported by Brett Stahlman - since array_subrange now calls quote_escapes as necessary, callers don't need to call it after array_patsub returns. Same fix as for array_patsub 1/31 ---- configure.in - add -DSOLARIS to LOCAL_CFLAGS for solaris x config-bot.h - don't #undef HAVE_GETCWD if GETCWD_BROKEN and SOLARIS are both defined. Solaris's loopback mount implementation breaks some of the file system assumptions the replacement getcwd uses. builtins/common.c - if GETCWD_BROKEN is defined, call getcwd with PATH_MAX for the size argument, so it will allocate a buffer for the current working dir with that size, instead of one that's `big enough' config.h.in - add #undef PRI_MACROS_BROKEN for AIX 4.3.3 pathexp.h - new flag value for quote_string_for_globbing: QGLOB_REGEXP (quoting an ERE for matching as a string) pathexp.c - change quote_string_for_globbing to understand QGLOB_REGEXP execute_cmd.c - change execute_cond_node to pass 2 (regexp match), 1 (shell pattern match), or 0 (no matching) to cond_expand_word subst.c - change cond_expand_word to translate SPECIAL==2 into passing QGLOB_REGEXP to quote_string_for_globbing locale.c - by default, if all else fails, set shell's idea of locale to "" instead of its idea of `default_locale' -- the library functions behave better with that value 2/2 --- builtins/printf.def - if PRI_MACROS_BROKEN is defined, #undef PRIdMAX (AIX 4.3.3 broken) 2/3 --- Makefile.in,{builtins,doc}/Makefile.in,lib/*/Makefile.in - add assignment for datarootdir as per GNU coding standards Makefile.in,builtins/Makefile.in,lib/intl/Makefile.in,po/Makefile.in.in - use @localedir@ instead of $(datadir)/locale in assignment 2/13 ---- jobs.c - fix compact_jobs_list to not return js.j_lastj, since that is in use and should not be overwritten. Fix from Len Lattanzi 2/16 ---- lib/readline/text.c - change rl_forward_char to allow moving to the end of the line when using the arrow keys in vi insertion mode, rather than having the behavior identical between vi command and insertion modes. Change suggested by Hugh Sasse 2/19 ---- CWRU/audit-patch - patch from Steve Grubb of RedHat to make bash audit root's behavior by logging commands using his audit framework. Enabled if the shell's name is `aubash'. 3/8 --- jobs.c - use WSTATUS (p->status) instead of bare p->status. Fix from Jim Brown 3/9 --- lib/readline/{complete,input,isearch,misc,readline,text,vi_mode}.c - make sure cases where rl_read_key returns -1 (usually due to EIO because the controlling tty has gone away) are handled correctly. Prompted by report from Thomas Loeber 3/10 ---- sig.c - new function, top_level_cleanup, callable from contexts where some cleanup needs to be performed before a non-fatal call to jump_to_top_level sig.h - new extern declaration for top_level_cleanup builtins/common.c - add calls to top_level_cleanup before calls to jump_to_top_level in a builtin command context (no_args(), get_numeric_arg()). Fixes bug reported by Ian Watson lib/readline/display.c - in _rl_move_cursor_relative, use `new' when comparing against the last invisible character in the prompt, since they both denote buffer indices when in a multibyte locale, whereas `dpos' is a display position 3/13 ---- lib/readline/complete.c - set rl_completion_append_character to the default (' ') in set_completion_defaults(). Fixes bug reported by David Emerson 3/23 ---- builtins/evalfile.c - make sure read() returns a value >= 0 before using it as an index into string[] - use a variable of type `ssize_t' for return value from read() - only try to read the entire contents of a regular file in one shot if the file size is less than SSIZE_MAX. These fix problems reported by hooanon05@yahoo.co.jp. include/typemax.h - define SSIZE_MAX as 32767 if it's not defined lib/readline/display.c - in rl_redisplay() and update_line(), if redrawing the prompt because it contains invisible characters, make sure we redraw the character indicating a modified history line and take it into account when computing _rl_last_c_pos - in update_line, if deleting characters and redrawing the new text, make sure we adjust _rl_last_c_pos by wrap_offset in a multibyte locale if the text we're drawing starts before or at the last invisible character in the prompt string. Fixes bug reported on bug-readline by J Pelkey parse.y - when adding at CTLESC character to the current token, do not escape it with CTLESC if pass_next_character indicates that the CTLESC was escaped by a backslash. Fixes bug reported by Paul Bagshaw . 3/25 ---- lib/readline/text.c - in rl_forward_char, short-circuit the loop if in emacs mode and rl_point == rl_end. Fixes problem with multibyte locales reported by Len Lattanzi 3/29 ---- command.h - new flag for subshell_environment: SUBSHELL_PROCSUB, for process substitution subst.c - add SUBSHELL_PROCSUB to subshell_environment in process_substitute 3/30 ---- doc/Makefile.in - fix installation of bash.info to understand that it is in the build directory, not the source directory mailcheck.c - new function, init_mail_dates, calls remember_mail_dates only if there are no mailboxes in `mailfiles' - new function, init_mail_file, initializes a FILEINFO, using the last time mail was checked as the mtime and atime (or the time the shell was started if last_time_mail_checked is uninitialized) - call init_mail_file instead of update_mail_file in add_mail_file, called from remember_mail_dates (which is supposed to initialize the list of mail files) - new convenience functions, alloc_mail_file and dispose_mail_file to allocate and free FILEINFO structs mailcheck.h - extern declaration for init_mail_dates shell.c - call init_mail_dates instead of remember_mail_dates 4/4 --- builtins/read.def - changes to print $PS2 when a line is continued with a backslash in an interactive shell. This is as POSIX requires 4/5 --- subst.c - make sure quote_escapes is only ever called when the word to be escaped is not marked as double-quoted -- cleaner, and allows us to make certain assumptions 4/6 --- subst.c - change all EX_* defines to begin with SX_ - new flag, SX_NOCTLESC, obeyed by string_extract_verbatim, tells it to not obey CTLESC quoting - change quote_escapes to not quote CTLESC with CTLESC if one of the chars in $IFS is CTLESC, since the return value from quote_string will be passed to word splitting and filename generation - change read_comsub to do the same thing for unquoted command substitutions - change list_string to pass SX_NOCTLESC if CTLESC is one of the chars in $IFS, so it will split on CTLESC instead of using it as a quote character 4/7 --- subst.c - slight change to string_extract_verbatim to allow CTLESC to quote CTLNUL even if SX_NOCTLESC is set in the flags passed, to protect the CTLNULs from future calls to remove_quoted_nulls. Only matters when $IFS contains CTLESC - changes to cope with $IFS containing CTLNUL in the same way as the CTLESC changes builtins/read.def - changes to cope with $IFS containing CTLNUL in the same way as the CTLESC changes 4/16 ---- lib/sh/strftime.c - a couple of fixes to the `%z' code eval.c - add an fflush after printing the auto-logout message 4/24 ---- subst.c - add call to top_level_cleanup in exp_jump_to_top_level to get things like unwind-protects and the loop levels cleaned up {arrayfunc,expr,variables}.c - add calls to top_level_cleanup before jump_to_top_level() 4/27 ---- builtins/complete.def - make sure the `command' argument to the -C option is printed with single quotes, since multi-word commands will require them. Bug reported by martin@snowplow.org execute_cmd.c - change execute_builtin_or_function and execute_subshell_builtin_or_function to call fflush(stdout) after the builtin or function returns, to make sure that all output is flushed before the call returns. It matters on cygwin. Fix suggested by Eric Blake redir.c - in do_redirection_internal, if the file descriptor being acted upon is the same one used by the stdout stream, call fflush(stdout) to make sure all output is flushed before changing the underlying fd out from underneath stdio. Fix suggested by Eric Blake 4/30 ---- builtins/common.c - new function, sh_chkwrite(int), fflushes stdout and checks for error; printing an error message and returning a new exit status if there's an error on stdout. Takes exit status as argument; returns new exit status (EXECUTION_FAILURE if write error) builtins/common.h - new extern declaration for sh_chkwrite builtins/{alias,cd,complete,echo,fc,history,pushd,shopt,times,trap,type,ulimit,umask}.def - change to use sh_chkwrite to report write errors builtins/fc.def - if an error occurs while writing commands from the history to a file to be executed, report a write error and return failure without attempting to execute any commands 5/1 --- builtins/{bind,declare,set,setattr}.def - change to use sh_chkwrite to report write errors 5/2 --- lib/readline/input.c - fix off-by-one errors in _rl_get_char (pop_index) and rl_stuff_char (push_index) that caused the 511th character in the buffer to be discarded. Fixes bug reported by Tom Bjorkholm 5/8 --- subst.c - fix parameter_brace_remove_pattern to pass getpattern() newly-allocated memory. If word expansions (particularly brace expansions) are required, the expansion code will free the string passed to expand_word_internal, and we don't want to free unallocated memory (patstr++) or have duplicate frees (patstr). Fixes bug reported on Red Hat bugzilla 5/9 --- lib/readline/signals.c - fix bug in rl_set_signals that caught SIGINT twice and didn't catch SIGTERM. Bug reported by Ed Kwan 5/18 ---- jobs.c - change compact_jobs_list to return 1 if js.j_lastj == 0 and there is a job in jobs[0]; compact_jobs_list should never return an index already occupied - change reset_job_indices to avoid infinite looping when js.j_firstj == 0 or js.j_firstj == js.j_jobslots upon function entry. Fixes bug reported by osicka@post.cz 5/20 ---- execute_cmd.c - new variable, executing_builtin, keeps track of number of "levels" of builtins being executed; incremented by execute_builtin; saved and restored by execute_simple_command subst.c - new variable, assigning_in_environment, set and unset around calls to assign_in_env by the expansion code variables.c - use executing_builtin and assigning_in_environment to decide whether or not to look into temporary_env when calling find_variable_internal. Fixes problem reported by Kevin Quinn 5/22 ---- redir.c - change add_undo_redirect to differentiate between file descriptors greater than SHELL_FD_BASE (currently 10) used internally to save others and then being the targets of user redirection and fds that are just the target of user redirections. The former need to have an `exec undo' redirect added to undo it in case exec throws away redirections; the latter does not. We use the close-on-exec flag for this: if it's set, we assume that the file descriptor is being used internally to save another. Fixes problem reported by Ian Jackson shell.c - new function, init_interactive_script(), does interactive initialization for a script run with `bash -i script' -- does everything the same as init_interactive except set `interactive == 1', which causes the shell to read from the standard input, after calling init_noninteractive - call init_interactive_script if a script is run as `bash -i script'. Fixes problem reported by Joseph Michaud 5/24 ---- builtins/printf.def - change vbadd to only call FASTCOPY if the passed buffer length is > 1 - if the `-v' option is supplied and `vbuf' is already non-null from a previous `printf -v var' call, set vbuf[0]=0 explicitly instead of relying on vbadd to do it -- vbadd may not be called. - fix PRETURN macro to set vbuf[0] == 0 if vbuf is not freed. These should fix problem reported by Elmar Stellnberger lib/readline/display.c - fix update_line to deal with the case where col_lendiff > 0 (meaning the new string takes up more screen real estate than the old) but lendiff < 0 (meaning that it takes fewer bytes to do so). This can happen when a multibyte prompt string is replaced with a longer one containing only single-byte characters (e.g., when doing a reverse i-search). Fixes gentoo bug reported by Peter Volkov builtins/read.def - make sure we only print $PS2 if the standard input is a terminal - new function, read_mbchar, to read a multibyte character so we can make sure we read entire multibyte chars when `read -n' is used, rather than bytes. Only called when -n is supplied. Fixes problem reported by Stanislav Brabec 5/25 ---- externs.h - new #defines for third argument to named_function_string: FUNC_MULTILINE (don't suppress newlines) and FUNC_EXTERNAL (convert to external display form) subst.h - new extern declaration for remove_quoted_escapes subst.c - remove_quoted_escapes is now global print_cmd.c - in named_function_string, if FUNC_EXTERNAL is in the flags argument, call remove_quoted_escapes to convert from internal to external form. Fixes bug reported by Bo Andresen variables.c,builtins/{declare,setattr,type}.def - use FUNC_MULTILINE in calls to named_function_string as appropriate - add FUNC_EXTERNAL to calls to named_function_string as appropriate 5/27 ---- {make_cmd,variables}.c - changes to enable the shell to compile when debugger support is configured out (function_def hash table and access functions). Fixes bug reported by Horst Wente builtins/help.def - fix bug in `help' two-column printing to avoid referencing shell_builtins[num_shell_builtins] error.c - in get_name_for_error, use dollar_vars[0] if the name returned from looking in $BASH_SOURCE[0] is the empty string as well as if it's null 5/31 ---- arrayfunc.c - change array_value_internal to set *RTYPE to 1 if the reference is array[*] and 2 if the reference is array[@] subst.c - in parameter_brace_expand_word, set the flags returned by the word desc to include W_HASQUOTEDNULL if array_value returns QUOTED_NULL for an array reference like x[*] and the word is quoted. Fixes bug reported by Christophe Martin 6/1 --- jobs.c - several changes to preserve errno if tcgetpgrp/tcgetattr/tcsetattr fail, for subsequent error messages - change initialize_job_control to turn off job control if the terminal pgrp == -1 or is not equal to shell_pgrp (with an error message) - in initialize_job_control, if the shell has been forced interactive with -i, make sure stderr is hooked to a tty before using it as the controlling terminal. If it's not, try to open /dev/tty and assign it to shell_tty. Fixes problems reported by Derek Fawcus 6/13 ---- support/shobj-conf - changes to support shared object and shared library creation on AIX 5.x and later versions. From Niklas Edmundsson 6/17 ---- builtins/mkbuiltins.c - new array of builtins, posix_builtins, containing builtins listed as special to the command search order by POSIX - add POSIX_BUILTIN to the builtin flags if the builtin name is one that's special to the posix command search order builtins.h - new define, POSIX_BUILTIN, means that a builtin is special to the posix command search order 6/22 ---- lib/readline/display.c - new macro, WRAP_OFFSET, intended to replace W_OFFSET. Takes prompt strings longer than one physical line with invisible characters on the second line into account when calculating the number of invisible characters on the current screen line - use WRAP_OFFSET where appropriate (update_line, _rl_move_cursor_relative) - change update_line to deal with adjusting _rl_last_c_pos in a multibyte environment when the prompt has invisible chars on the second line and redisplay has output the invisible characters - change _rl_move_cursor_relative to adjust _rl_last_c_pos in a multibyte environment when the prompt has invisible chars on the second line and the redisplay draws the invisible character. Fixes redisplay bug reported by Andreas Schwab 7/11 ---- lib/readline/rltty.c - enable flush-output code for systems other than AIX 4.1. Problem reported by Jan Kratochvil 7/12 ---- lib/readline/display.c - set prompt_invis_chars_first_line from the portion of the prompt following the final newline, instead of from the prefix. Fixes bug reported on the Ubuntu bug list by dAniel hAhler 7/13 ---- variables.c - use native __QNX__ and __QNXNTO__ cpp defines instead of qnx and qnx6, respectively. Patch from Sean Boudreau lib/sh/getcwd.c - #undef HAVE_LSTAT on qnx, so it uses stat instead. Patch from Sean Boudreau 7/21 ---- builtins/common.c - change sh_invalidnum to be a little smarter about octal and hex numbers and change the message appropriately. Bug originally reported on coreutils list by Jürgen Niinre 7/26 ---- test.c - make sure the string passed to test_unop has only a single character following the `-'. Fixes bug reported by Michael A. Smith parse.y - better input validation: make sure a word looks like a conditional unary operator (-X) before calling test_unop 7/28 ---- trap.c - in trap_handler, if it's called directly from the signal handler (e.g., SIGINT sighandler, set by set_sigint_handler), but the trap disposition has been reset to the default between the assignment and receipt of the signal, check that the signal is trapped and issue a warning if the shell was compiled with debugging enabled. Fixes bug reported by Fergus Henderson 8/1 --- lib/readline/{util,histexpand}.c - fixes for small memory leaks from Michael Snyder 8/18 ---- Makefile.in - add dependency on builtins/builtext.h to nojobs.o list. Fixes `make -j 5' issue reported by Chris MacGregor examples/loadables/Makefile.in - add @LDFLAGS@ to SHOBJ_LDFLAGS assignment -- experimental. Suggested by Mike Frysinger examples/loadables/{basename,cut,dirname,finfo,head,ln,logname,mkdir,pathchk,print,printenv,push,realpath,rmdir,sleep,tee,truefalse,tty,uname,unlink,whoami}.c - fix up some includes. Fix from Mike Frysinger 8/21 ---- histexpand.c - fix another memory leak in history_find_word. Bug report originally from Michael Snyder ; test case suggested by Jim Blandy 8/26 ---- subst.c - change to do_assignment_internal to make an assignment to a variable with the `noassign' internal attribute not a variable assignment error. - fix do_assignment_internal so assignment to a `noassign' variable does not cause it to suddenly become visible if it's currently invisible 9/3 --- stringlib.c - change strsub to check whether or not temp is non-null before trying to null-terminate it. Also make sure temp is allocated even if the pattern and replacement strings are empty, and set to a copy of string (like ${foo//}) Bug report from Timo Lindfors 9/10 ---- {config.h,Makefile,configure}.in,aclocal.m4 - new tests for fpurge and __fpurge lib/sh/fpurge.c, externs.h - new file, fpurge(3) implementation with external decl in externs.h builtins/common.c - add call to fpurge(stdout) to sh_chkwrite {redir,execute_cmd}.c - add call to fpurge(stdout) after fflush(stdout) before changing stdout file descriptor and after a builtin or function executes 9/12 ---- expr.c - make sure noeval is set to 0 when a longjmp occurs, since it will not be reset otherwise, and it can be set to 1 while processing a {pre,post}-increment or {pre,post}-decrement token - set noeval to 0 at the beginning of evalexp, since it's never called recursively 9/14 ---- config-top.h - new builder-modifiable define: DONT_REPORT_BROKEN_PIPE_WRITE_ERRORS Turning it on will cause errors from EPIPE to not be reported by the normal shell write error message mechanism builtins/common.c - if DONT_REPORT_BROKEN_PIPE_WRITE_ERRORS is defined, don't print an error message from sh_wrerror if errno == EPIPE. Suggestion from Petr Sumbera 9/19 ---- {jobs,nojobs}.c,jobs.h - add code to retry fork() after EAGAIN, with a progressively longer sleep between attempts, up to FORKSLEEP_MAX (16) seconds. Suggested by Martin Koeppe 9/21 ---- version.c - change copyright year to 2007 9/25 ---- pathexp.c - change quote_string_for_globbing to add a backslash in front of a backslash appearing in the pathname string, since the globbing code will interpret backslashes as quoting characters internally. Bug reported by on the debian list (443685) 10/8 ---- lib/readline/display.c - in update_line, make sure _rl_last_c_pos is > 0 before setting cpos_adjusted (or we actually moved the cursor to column 0 in _rl_move_cursor_relative). Fixes redisplay bug with prompt with only invisible characters reported by dAniel hAhler 10/10 ----- lib/readline/display.c - in rl_redisplay, when calculating the new physical cursor position in a multibyte locale (`tx'), do not call rl_backspace if tx ends up < 0. Rest of fix for bug reported by dAniel hAhler 10/12 ----- lib/sh/getcwd.c - fix memory overwrite problem that's possible if buf is NULL and passed size is greater than the pathname length. Reported by Ian Campbell builtins/ulimit.def - change the multiplier for the -c and -f options (`blocks') to 512, the traditional value (and the one POSIX specifies). Bug reported by Pete Graner braces.c - pass process substitution through unchanged the same as command substitution. Prompted by suggestion from Stephane Chazelas lib/readline/input.c - in rl_unget_char, fix off-by-one error when resetting pop_index if it's < 0. Bug reported by Uwe Doering builtins/type.def - change exit status of `type' to not successful if any of the requested commands are not found. Reported by Stephane Chazleas pcomplete.c - change command_line_to_word_list to use rl_completer_word_break_characters instead of the shell metacharacters to split words, so programmable completion does the same thing readline does internally. Reported by Vasily Tarasov 10/16 ----- bashline.c - When completing a command name beginning with a tilde and containing escaped specical characters, dequote the filename before prefixing it to the matches, so the escapes are not quoted again. Reported by neil@s-z.org 10/17 ----- expr.c - in readtok(), don't reset lasttp if we've consumed the whitespace at the end of the expression string. Fixes error message problem reported by 11/1 ---- builtins/printf.def - change asciicode() to return intmax_t; add multibyte character support instead of assuming ASCII (depending on behavior of system multibyte support functions). Fixes bug reported by Rich Felker 11/5 ---- execute_cmd.c - if redirections attached to a compound command fail, make sure to set last_command_exit_value when returning EXECUTION_FAILURE. Fixes bug reported separately by Andreas Schwab and Paul Eggert 11/9 ---- builtins/read.def - make sure the return value from get_word_from_string is freed if non-null. Fixes memory leak bug reported by Lars Ellenberg 11/10 ----- variables.c - use getpid() as value of seeded_subshell to avoid problems with random number generator not getting re-seeded correctly when subshells are created. Fix from Tomas Janousek lib/readline/display.c - in update_line(), when outputting characters at the end of the line, e.g., when displaying the prompt string, adjust _rl_last_c_pos by wrap_offset if the text we're drawing begins before the last invisible character in the line. Similar to fix from 5/24. Fixes bug reported by Miroslav Lichvar 11/14 ----- subst.c - fix $[ expansion case to deal with extract_arithmetic_subst returning NULL (if the `]' is missing) and return the construct unchanged in that case. Fixes tab completion bug reported by Heikki Hokkanen (debian bug 451263) lib/readline/mbutil.c - fix _rl_find_next_mbchar_internal to deal with invalid multibyte character sequences when finding non-zero-length chars. Fixes bug reported by Morita Sho 11/15 ----- variables.c - add new function `seedrand' to seed the bash random number generator from more random data. Suggestion from Steve Grubb - replace the rng in brand() with a slightly better one from FreeBSD (filtered through Mac OS X 10.5). Replacement suggested by Steve Grubb 11/21 ----- configure.in - darwin 9 also requires linking against libreadline.a and libhistory.a because of Apple's questionable decision to ship a libreadline "replacement" that doesn't provide all functions doc/{bash.1,bashref.texi} - slight change to the text describing the effect of set -e when in a || or && list 12/5 ---- jobs.c - fix raw_job_exit_status to correct mixing of int/WAIT values (need to return a WAIT) - arrange so that children run as part of command substitutions also set the SIGINT handler to wait_sigint_handler, since they effectively don't do job control - in wait_for, if a child run as part of a command substitution exits due to SIGINT, resend the SIGINT to the waiting shell with kill(2). This makes sure the exit status propagates doc/{bash.1,bashref.texi} - tighten up the language describing when bash tries to see if its stdin is a socket, so it can run the startup files. Suggested by Vincent Lefevre eval.c - in the DISCARD case of a longjmp to top_level, make sure last_command_exit_value is set to EXECUTION_FAILURE if it's 0, but leave existing non-zero values alone subst.c - in command_substitute, don't reset pipeline_pgrp in the child process -- this means that second and subsequent children spawned by this comsub shell get put into the wrong process group, not the shell's. Fix for bug reported by Ingo Molnar 12/6 ---- support/shobj-conf - make sure the cases for darwin8.x (Mac OS X 10.4.x) are extended to darwin9.x (Mac OS X 10.5.x). Fixes problem originally reported against readline-5.2 by schneecrash@gmail.com 12/8 ---- subst.c - make sure to add the results of (successful) tilde expansion as a quoted string, to inhibit pathname expansion and word splitting. From recent Austin Group interpretation. include/shtty.h, lib/sh/shtty.c - add ttfd_onechar, ttfd_noecho, ttfd_eightbit, ttfd_nocanon, and ttfd_cbreak to set tty attributes associated with a particular file descriptor (which is presumed to point to a terminal). Support for fix for bug reported by b_bashbug@thebellsplace.com lib/readline/display.c - make sure we only use rl_invis_chars_first_line when the number of physical characters exceeds the screen width, since that's the only time expand_prompt sets it to a valid value 12/12 ----- builtins/set.def - change set_minus_o_option to return EX_USAGE if an invalid option name is supplied. All callers can handle it. - change set_builtin to return what set_minus_o_option returns if it's not EXECUTION_SUCCESS. This allows EX_USAGE errors to abort a shell running in posix mode 12/14 ----- builtins/read.def - generalize the calls to the tty attribute functions to maintain a local copy of the terminal attributes and use the fd supplied as the argument to the -u option (default 0). Fix for bug reported by b_bashbug@thebellsplace.com doc/bashref.texi, lib/readline/doc/{history,rlman,rluser,rluserman}.texi - Slight changes to conform to the latest FSF documentation standards. Patch from Karl Berry 12/20 ----- execute_cmd.c - after calling clear_unwind_protect_list, make sure we reset parse_and_execute_level to 0, since there's nothing left to restore it if top_level_cleanup tests it. Fixes bug reported by Len Lattanzi 12/31 ----- lib/sh/getcwd.c - new function, _path_checkino, checks whether the inode corresponding to the path constructed from the first two arguments is the same as the inode number passed as the third argument - if BROKEN_DIRENT_D_INO is defined, meaning the d_ino/d_fileno member of struct dirent doesn't contain valid values, use _path_checkino instead of directly comparing against d_fileno. Fixes Interix problem reported by Michael Haubenwallner 1/7/2008 -------- array.c - fix array_subrange to separate elements in returned string with first char of $IFS if QUOTED is non-zero, since this indicates the caller used ${array[@]:foo}. Fixes bug reported by Lea Wiemann 1/8 --- subst.c - new function returning a string containing the first character of $IFS: char *ifs_firstchar(int *) subst.h - extern declaration for ifs_firstchar() array.c - call ifs_firstchar() to get first character of $IFS when needed (array_subrange() and array_patsub()) 1/11 ---- lib/readline/display.c - use sentinel variable set at end of init_line_structures to decide whether to call it from rl_redisplay, since early SIGWINCH on Mac OS X that hits during this function can cause _rl_wrapped_line to be referenced before initialization. Fix for bug reported by Len Lattanzi subst.[ch] - skip_to_delim is now compiled into the shell all the time, not just when readline is linked in subst.c - use skip_to_delim to find the `/' denoting the end of a pattern in pattern substitution, since it knows more shell syntax than quoted_strchr and understands multibyte characters. Fixes bug reported by Dmitry V Golovashkin 1/15 ---- subst.c - add `flags' argument to skip_to_delim telling it whether or not to set no_longjmp_on_fatal_error; set this flag when calling from the readline completion code subst.h - update extern declaration for skip_to_delim 1/17 ---- subst.c - expand_prompt_string takes a third argument: the initial flags for the WORD subst.h - change extern declaration for expand_prompt_string to add third arg bashline.c - pass W_NOCOMSUB as third argment to expand_prompt_string when calling from bash_directory_completion_hook, since we don't want to do command substitution from the completion code parse.y - change call to expand_prompt_string 1/18 ---- doc/Makefile.in - added an `install_builtins' rule to install the builtins.1 man page, preprocessing it with sed to force `.so man1/bash.1', which some versions of man require. Suggestion from Peter Breitenlohner - new target `install_everything' that will install normal documentation and builtins man page - changed uninstall target to remove bash_builtins page from man directory lib/readline/vi_mode.c - new function, rl_vi_insert_mode, which calls rl_vi_start_inserting to make sure the value of `last command to repeat' is set correctly. Fix from Thomas Janousek - add support for redoing inserts made with the `I' command. Fix from Thomas Janousek - add support for redoing inserts made with the `A' command lib/readline/readline.h - new extern declaration for rl_vi_insert_mode lib/readline/{misc,readline,vi_mode,vi_keymap}.c - change calls to rl_vi_insertion_mode to rl_vi_insert_mode 1/19 ---- builtins/read.def - change timeout behavior when not reading from a tty device to save any partial input in the variable list, but still return failure. This also causes variables specified as arguments to read to be set to null when there is no input available. Fix inspired by Brian Craft 1/21 ---- builtins/fc.def - change computation of last_hist to use remember_on_history instead of a hard-coded `1'. This keeps fc -l -1 in PROMPT_COMMAND from looking too far back 1/25 ---- lib/readline/complete.c - fix fnwidth to use string[pos] instead of *string when testing the current character for a control character or rubout 2/2 --- general.c - change posix_initialize to turn off source/. searching $PWD when the file sourced is not found in $PATH. Fixes bug reported by Paolo Bonzini and Eric Blake 2/9 --- builtins/*.def - changes to text and formatting suggested by Jan Schampera 2/16 ---- bashline.c - change command_word_completion_function to use the word completion found by readline, which matters only when ignoring case is on and the completion found in the file system differs in case from the text the user typed (this is what readline does for normal filename completion). Fixes issue reported by Jian Wang . 2/18 ---- builtins/source.def - if the filename passed as an argument contains a `/', don't search $PATH. Not sure why it wasn't like this before 2/21 ---- lib/readline/terminal.c - change rl_crlf so that the MINT system on ATARI systems adds a carriage return before the \n 2/22 ---- doc/{bash.1,bashref.texi} - added text to the EXIT STATUS section noting that exit statuses fall between 0 and 255, inclusive support/mkversion.sh - output a #define for DEFAULT_COMPAT_LEVEL (${major}${minor}; e.g. 32) to version.h version.c - int variable, shell_compatibility_level, set to DEFAULT_COMPAT_LEVEL by default builtins/shopt.def - new shopt variable, compat31, sets shell_compatibility_level to 31 (or back to default if unset) execute_cmd.c - in execute_cond_node, restore bash-3.1 behavior of quoted rhs of regexp matches if shell_compatibility_level == 31 2/28 ---- lib/readline/rltty.c - set readline_echoing_p = 1 if tcgetattr fails and sets errno to EINVAL, as Linux does when the fd is a pipe. Reported by Mike Frysinger 3/6 --- {MANIFEST,Makefile.in},lib/sh/{casemod,uconvert,ufuncs}.c - new library sources from bash-4.0-devel tree lib/sh/spell.c - moved cdspell() here from builtins/cd.def, renamed dirspell() externs.h - new declarations for extern functions from new library files - new extern declaration for lib/sh/spell.c:dirspell() builtins/cd.def - call extern library function dirspell(); remove static cdspell() builtins/read.def - when read times out, make sure input_string is null-terminated before assigning any partial input read to the named variables 3/10 ---- lib/glob/xmbsrtowcs.c - cut the number of memory allocations in xdupmbstowcs by not keeping track of the indices if the caller hasn't asked for it 3/17 ---- builtins/fc.def - make sure the adjustment to i in fc_gethnum uses the same formula fc_builtin uses to calculate last_hist - make sure that every time fc_gethnum is called, the fc command last in the history list has not yet been deleted, since fc_gethnum assumes that it has not. Fix from John Haxby lib/readline/complete.c - new private library function, _rl_reset_completion_state(), used to reset any completion state internal to the library when a signal is received - call _rl_reset_completion_state() before returning from rl_complete_internal lib/readline/rlprivate.h - new extern declaration for _rl_reset_completion_state lib/readline/signals.c - call _rl_reset_completion_state from rl_signal_handler on SIGINT. This fixes one of the problems identified by Mika Fischer pcomplete.c - programmable_completions now saves pointer to the compspec it's working with in new global variable CURCS - new function, pcomp_set_readline_variables, that sets or unsets readline variables based on a passed flags value (COPT_FILENAMES, etc.) - new function, pcomp_set_compspec_options, to set or unset bits in the options word of a passed compspec (default CURCS) - only call bash_dequote_filename (via rl_filename_dequoting_function) from pcomp_filename_completion_function if the readline state word indicates word completion is in progress pcomplete.h - new extern declaration for curcs - new extern declaration for pcomp_set_readline_variables - new extern declaration for pcomp_set_compspec_options bashline.c - fix bash_dequote_filename to implement shell quoting conventions: 1. Inhibit backslash stripping within single quotes 2. Inhibit backslash stripping within double quotes only if the following character is one of the special ones - call pcomp_set_readline_variables from attempt_shell_completion instead of doing the equivalent inline 3/18 ---- bracecomp.c - make sure we sort array of matches in byte order (using strcmp). so the brace calculations work correctly even when the locale orders characters like aAbBcC...zZ. Fixes bug reported by Torsten Nahm 3/20 ---- lib/readline/{rltty,signals}.c - move block_sigint and release_sigint from rltty.c to signals.c; add _rl_ prefix to make them public to the library; change callers. From Jan Kratochvil lib/readline/rlprivate.h - new extern declarations for _rl_block_sigint and _rl_release_sigint lib/readline/display.c - add calls to _rl_block_sigint and _rl_release_sigint to rl_redisplay, since it maniupluates global data structures. Fix from Jan Kratochvil builtins/printf.def - change calls to asprintf and manually adding to vbuf to use calls to vsnprintf against vbuf directly -- if the number of characters to be written overflows the buffer, realloc the buffer and use vsnprintf again. This should reduce the memory used by printf. Idea from Yuya Katayama lib/readline/doc/rltech.texi - documented rest of readline's state flags, including RL_STATE_CALLBACK - documented rl_save_state and rl_restore_state 3/27 ---- lib/readline/{rlprivate.h,{display,readline,rltty,terminal,text}.c} - rename readline_echoing_p to _rl_echoing_p for namespace consistency lib/readline/{rlprivate.h,{callback,readline,util}.c} - rename readline_top_level to _rl_top_level for namespace consistency builtins/ulimit.def - new -b (socket buffer size) and -T (number of threads) options array.c - fix bug in calculation of the array element assignment string length: use length of `is' instead of `indstr'. Reported as ubuntu bug #202885 by John McCabe-Dansted builtins/setattr.def - new function, show_all_var_attributes, displays attributes and values for all shell variables (or shell functions) in a reusable format builtins/common.h - new extern declaration for show_all_var_attributes builtins/declare.def - change `declare -p' to print out all variable attributes and values, and `declare -fp' to print out all function attributes and definitions. Inspired by request from John Love-Jensen doc/{bash.1,bashref.texi} - document new -b and -T options to ulimit - tighten up language describing AND and OR lists - add description of new behavior of `declare -p' 3/28 ---- pcomplete.c - rename curcs -> pcomp_curcs - new global completion variable, pcomp_curcmd, the current command name being completed builtins/complete.def - new builtin, compopt, allows completion options for command names supplied as arguments or the current completion being executed to be modified. Suggested by Mika Fischer 3/30 ---- doc/{bash.1,bashref.texi},lib/readline/doc/rluser.texi - document new compopt builtin 4/5 --- support/shobj-conf - change solaris10 stanza to use -fPIC to fix 64-bit sparc_v9/solaris10 compilations. Fix from Fabian Groffen builtins/read.def - added `-i text' option, inserts `text' into line if using readline. Suggested by many, used some ideas from Kevin Pulo doc/{bash.1,bashref.texi} - document new `-i text' option to read builtin 4/7 --- lib/readline/bind.c - new settable variable, `history-size', sets the max number of entries in the history list doc/bash.1,lib/readline/doc/{rluser.texi,readline.3} - document new `history-size' settable readline variable 4/8 --- builtins/complete.def - change build_actions calling sequence to take a struct with `other' (non-action) flag arguments (-p, -r) - add support for `-E' option to build_actions and complete builtin -- modifies or displays (internal) `_EmptycmD_' completion spec bashline.c - change attempt_shell_completion to try programmable completion on an `empty' command line and return the results doc/bash.1,lib/readline/doc/rluser.texi - documented new `-E' option to `complete' 4/9 --- bashhist.c - new variable, `enable_history_list', used to reflect setting of `-o history' option - change bash_history_{enable,disable,reinit} to set enable_history_list as well as remember_on_history builtins/set.def - use `enable_history_list' instead of `remember_on_history' to keep value of `-o history' option builtins/evalstring.c - instead of unwind-protecting remember_on_history, use a function to restore it to the value of `enable_history_list' after parse_and_execute runs the commands in the string. This allows history to be turned off in a startup file, for instance. Problem reported by Dan Jacobson 4/11 ---- bashline.c - limited support for completing command words with globbing characters (only a single match completed on TAB, absolute or relative pathnames supported, no $PATH searching, some support for displaying possible matches, can be used with menu completion). Suggested by Harald Koenig print_cmd.c - change redirection printing to output r_err_and_out as `&>file', since the man page says that's the preferred form 4/12 ---- builtins/*.def - change long doc so the first line is a short description - add `Exit Status:' section to each longdoc describing exit values builtins/help.def - new `-d' option to print short description of each utility - new `-m' option to print description of each builtin in a pseudo-manpage format (inspired by ksh93) doc/{bash.1,bashref.texi} - document new `-d' and `-m' options to `help' builtins/mapfile.def - new builtin, `mapfile', imported from bash-4.0-devel branch tests/{mapfile.{data,right,tests},run-mapfile} - tests for `mapfile' builtin doc/{bash.1,bashref.texi} - added description of `mapfile' builtin MANIFEST,Makefile.in,builtins/Makefile.in - added entries for mapfile source files arrayfunc.[ch] - new function, bind_array_element, to support mapfile builtin 4/20 ---- expr.c - fix operator precendence in expcond(): term after the `:' is a conditional-expression, not a logical-OR-expression (using C terminology). Bug reported by 4/22 ---- bashintl.h - new P_ define for using ngettext to decide on plural forms (currently unused) 4/25 ---- execute_cmd.c - in execute_disk_command, if the command is not found, search for a shell function named `command_not_found_handle' and call it with the words in the command as arguments. Inspired by Debian feature. doc/{bash.1,bashref.texi} - document new command_not_found_handle behavior in COMMAND EXECUTION section configure.in - change default version to bash-4.0-devel 4/28 ---- variables.c - change push_func_var and push_exported_var to call stupidly_hack_special_variables if the temporary variable is going to be disposed. This undoes any internal changes caused by a local variable assignment in the environment or in a shell function. Bug reported by Morita Sho in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=478096 5/3 --- builtins/fc.def - fixed a problem caused by change of 1/21 to use remember_on_history, since it's turned off by parse_and_execute(), but can cause the last command in history to be deleted and leave last_hist pointing beyond the end of the history list. edit_and_execute_command can do this. bashline.c - new define, RL_BOOLEAN_VAR_VALUE, to take a readline boolean variable and get its value as 0 or 1 (consider making readline global) - put tty back into canonical mode before calling parse_and_execute in edit_and_execute_command and then back into raw mode after it returns. Fixes problem identified by . 5/4 --- lib/glob/glob.c - code to support `globstar' option: GX_GLOBSTAR and two internal flags. Changes to skipname, glob_vector, mbskipname, glob_filename. New function finddirs(). lib/glob/glob.h - new defines to support globstar code builtins/shopt.def - new shell option, `globstar', enables special handling of `**' in glob patterns -- matches all directories recursively pathexp.h - extern declaration for glob_star pathexp.c - break inline code out of quote_globbing_chars into a separate function to decide whether a character is a globbing char: glob_char_p - change shell_glob_filename to call glob_filename with the GX_GLOBSTAR flag if glob_star is set doc/{bash.1,bashref.texi} - document new `globstar' shell option arrayfunc.c - new function, broken out of quote_array_assignment_chars: quote_assign; extended from old code to make sure that globbing chars and chars in $IFS are quoted when displaying assignment statements, especially in compound array assignments 5/5 --- bashline.c - new variable, dircomplete_spelling, controls spelling correction of directory names when doing filename completion - change bash_directory_completion_hook to incorporate spelling correction if initial canonicalization of directory name fails builtins/shopt.def - new shell option, `dirspell', enables and disables spelling correction of directory names during word completion builtins/read.def - support for fractional timeout values (ival.uval); uses uconvert and falarm/setitimer config.h.in - new `HAVE_SETITIMER' define configure.in - look for setitimer(2), define HAVE_SETITIMER if found doc/{bash.1,bashref.texi} - document new `dirspell' shopt option - document new fractional values to `read -t timeout' 5/6 --- assoc.[ch] - new files, basic support for associative array implementation general.h - new extern declarations for sh_openpipe, sh_closepipe, trim_pathname general.c - new functions: sh_openpipe to create a pipe and move the file descriptors to a high range; sh_closepipe, to close pipe fds and clean up, and trim_pathname, to replace portions of a pathname with `...' (for prompting) jobs.c - don't set last_asynchronous_pid in child shell (messes up $!, among other things) parse.y,parser.h - moved definitions of parser flags to parser.h array.c - imported array_modcase (case-changing operations on arrays) from 4.0-devel branch array.h - new extern declaration for array_modcase lib/readline/complete.c - new variable, rl_menu_completion_entry_function, generator for rl_menu_complete - new menu completion `browsing' implementation, with several improvements over the old code. Inspired by Sami Pietila lib/readline/readline.h - extern declaration for rl_menu_completion_entry_function 5/8 --- lib/readline/complete.c - add support for a third argument to fnprint and print_filename, which supports replacing a specified portion of the pathnames printed when displaying possible completions with a `...' (or `___', if the prefix would be confused with a portion of the filename) - new variable, _rl_completion_prefix_display_length, sets the number of characters in a common prefix to be replaced with an ellipsis when displaying possible completions - add support to _rl_display_match_list to find the length of the common prefix of all items being displayed, and passing that value to print_filename for possible replacement with an ellipsis if that length is longer than _rl_completion_prefix_display_length lib/readline/bind.c - add support for retrieving value of history-size variable to _rl_get_string_variable_value - new bindable variable, completion-prefix-display-length. When displaying possible completions, matches with a common prefix longer than this value have the common prefix replaced with an ellipsis - support for retrieving value of completion-prefix-display-length variable to _rl_get_string_variable_value - new bindable variable, revert-all-at-newline: if enabled, causes all changes in history lines to be undone before readline returns after processing a newline doc/bash.1,lib/readline/doc/{readline.3,rluser.texi} - document new `completion-prefix-display-length' variable - document new `revert-all-at-newline' variable execute_cmd.c - change execute_builtin to not inherit the `-e' flag into commands executed by the `command' or `source/.' builtins if we are supposed to be ignoring the return value. This is like `eval'. Fixes bug reported by Hiroshi Fujishima 5/10 ---- variables.c - when reading the initial environment, don't create variables with names that are not valid shell identifiers. Fixes bug reported by Stephane Chazleas 5/13 ---- subst.c - fix string_quote_removal to gracefully handle the case where a backslash is the final character in the string (leaves the backslash in place). Fixes bug reported by Ian Robertson 5/16 ---- support/checkbashisms - Perl script that purports to check for bash-specific features in a shell script. Lifted from Debian via ubuntu 5/20 ---- lib/readline/display.c - in update_line, when deciding whether or not to adjust _rl_last_c_pos in a multibyte environment after printing the last line of a multiline prompt with invisible characters on the first and last lines, use the number of inivisible chars on the first line in the calculation deciding whether or not we're past the last invisible character and need to adjust the cursor position. Old code used the number of invisible chars on the last prompt line. Fixes bug reported by stuff@slinkp.com. - in update_line, when fixing _rl_last_c_pos after drawing the first line of the prompt, use the number of invisible chars on the first line as the offset, instead of the total number of invisible chars - use prompt_multibyte_characters, the number of multibyte chars in the prompt string, to short-circuit some relatively expensive multibyte text processing in rl_redisplay 5/21 ---- variables.c - new function, reinit_special_variables(), a hook for special vars that need their hook functions called when they're unset as a result of the shell reinitializing itself to run a script shell.c - shell_reinitialize now calls reinit_special_variables - shell_reinitialize now calls bashline_reset variables.h - new extern declaration for reinit_special_variables bashline.c - new function, bashline_reset(), called when the shell reinitializes in shell_reinitialize. Right now, just resets bash_readline_initialized to 0. bashline.h - new extern declaration for bashline_reset() 5/23 ---- bashhist.c - new function, bash_clear_history, clears the history and resets any associated internal bash state bashhist.h - extern declaration for bash_clear_history builtins/history.def - call bash_clear_history instead of clear_history for `history -c'. Fixes part of problem reported by Scott McDermott - decrement history_lines_this_session in delete_histent, called for `history -d' builtins/history.def,bashhist.[ch] - move delete_histent() to bashhist.c; rename to bash_delete_histent - move delete_last_history() to bashhist.c; rename to bash_delete_last_history() 5/25 ---- braces.c - add another parameter to mkseq(), the number of digits to put into each member of a numeric sequence (width), changes to determine any zero-padding go into expand_seqterm - changes to expand_seqterm to allow user-specified increments bashline.[ch],shell.c,sig.c - switched names of bashline_reinitialize and bashline_reset to better reflect their functions - when searching $PATH for directories to use for command completion, make sure to free `current_path' before going out of scope - new bindable function `dabbrev-expand', which is more or less menu completion using dynamic history completion as the generator - changes to bash_execute_unix_command to set variables for the executed command like programmable completion: READLINE_LINE (rl_line_buffer) and READLINE_POINT (rl_point) - change to bash_execute_unix_command to allow the executed command to change the readline line buffer by modifying the value of READLINE_LINE and to change rl_point by modifying the value of READLINE_POINT common.h - new SEVAL_ defines for later parse_string changes from 4.0-devel branch command.h - new defines for new &>> r_append_err_and_out redirection builtins/evalstring.c - new function, parse_string, parses a command from a passed string and returns the number of characters consumed. For satisfying Posix rules when parsing command substitutions, from bash-4.0-devel branch - split out common prolog code from parse_string and parse_and_execute into a separate function called from both parse.y - small changes to add symbols needed for parse_string - parser change to add `|&' as synonym for `2>&1 |'; translation is performed at parse time so |& never shows up in output of print_command, for instance. Picked up from zsh, merged in from bash-4.0-devel branch parse.y,{redir,copy_cmd,dispose_cmd,make_cmd,print_cmd}.c - implement new &>> r_append_err_and_out (like >>foo 2>&1); merged in from bash-4.0-devel branch doc/{bash.1,bashref.texi},lib/readline/doc/rluser.texi - document new optional increment in brace expansion - document new zero-padded fixed-width integer brace expansion - document new `dabbrev-expand' bindable readline command - document new effects of `bind -x' setting and reading the values of READLINE_LINE and READLINE_POINT - document new |& synonym for `2>&1 |' pipeline operator 5/26 ---- parse.y - recognize new ;& and ;;& case action list terminator tokens and implement them in the grammar, setting CASEPAT_FALLTHROUGH and CASEPAT_TESTNEXT flags as appropriate print_cmd.c - print new ;& and ;;& case clause action list terminators as appropriate execute_cmd.c - implement new case clause action list terminators: ;& - fall through to actions associated with next pattern list ;;& - fall through to tests in next pattern list doc/{bash.1,bashref.texi} - document new ;& and ;;& case clause action list terminators 5/28 ---- jobs.c - change waitchld so it treats SIGCHLD like SIGINT if `wait' is being executed, and allows wait to jump out before running any trap set on SIGCHLD. Fixes debian bug #483016 reported by Miroslav Rudisin - run_sigchld_trap is no longer static, so the trap code in trap.c can call it - change run_sigchld_trap to call set_impossible_sigchld_trap instead of just using a call to restore_default_signal jobs.h - new extern declaration for run_sigchld_trap trap.c - fix run_pending_traps to run a SIGCHLD trap if the trap handler isn't set to IMPOSSIBLE_TRAP_HANDLER - in trap_handler, don't reset the SIGCHLD trap handler to trap_handler if MUST_REINSTALL_SIGHANDLERS is defined - new function, set_impossible_sigchld_handler, sets the trap string associated with SIGCHLD to IMPOSSIBLE_TRAP_HANDLER; used as a sentinel by run_sigchld_trap and maybe_set_sigchld_handler - change maybe_set_sigchld_handler to set the SIGCHLD trap string only if the current value is IMPOSSIBLE_TRAP_HANDLER. This ensures that any traps on SIGCHLD set in a SIGCHLD handler will persist. Fixes debian bug #483016 reported by Miroslav Rudisin trap.h - new extern declaration for set_impossible_sigchld_trap 5/31 ---- parse.y - new function: parse_comsub(), parses $(...) by parsing command between parens and making sure the next token is `)'. From the bash-4.0-devel branch - new function: xparse_dolparen, helper function for parsing command substitutions in $(...). Called from subst.c to extract a command substitution during word expansion. From bash-4.0-devel branch - new function: rewind_input_stream(). Rewinds bash_input.location.string back to where it was before the shell parsed a $() command substitution. From bash-4.0-devel branch - changes to parse_matched_pair to combine most of the flag variables (was_dollar, in_comment, and so on) into a local flags word 6/2 --- parse.y - call trim_pathname, which retains only the last $PROMPT_DIRTRIM directories and replaces the intervening characters with `...', when expanding \w and \W doc/{bash.1,bashref.texi} - document the effect of setting PROMPT_DIRTRIM 6/3 --- builtins/ulimit.def - make the multiplier (block size) for -c and -f 512 bytes only if in Posix mode and 1024 bytes otherwise (as in previous versions). Uses POSIXBLK and BLOCK_SIZE defines to parameterize size based on value of posixly_correct doc/bashref.texi - document this addition to posix mode builtins/common.c - change get_numeric_arg to have a calling sequence and return value more closely mimicking general.c:legal_number(), with the addition of a flags word - add extra value for `fatal' argument to get_numeric_arg to force it to return failure to the caller rather than longjmping builtins/common.h - change prototype declaration for get_numeric_arg builtins/{break,shift}.def - change calls to get_numeric_arg to deal with new semantics and calling sequence builtins/history.def - display_history now returns an int - change calling sequence for get_numeric_arg in display_history - display_history now returns failure to the caller if get_numeric_arg detects an invalid number, rather than jumping back to the top level - use value returned by display_history as return status of history builtin, filtered through sh_chkwrite - history no longer aborts compound commands on invalid arguments. fixes problem reported by Chu Li {braces,subst}.c - extract_command_subst now takes a third flags argument; passed flags are ORd into flags passed to other functions; changed callers subst.h - move SX_* defines here from subst.c so parse.y:xparse_dolparen can see them and behave appropriately - extract_command_subst now takes a third flags argument; change prototype subst.c - change extract_command_subst to call xparse_dolparen when extracting a $() construct - change calls to extract_delimited_string to extract_command_subst as appropriate - if command_substitute returns a NULL word desc, don't call dispose_word_desc on it parse.y - change xparse_dolparen to use the SX_* flags now in subst.h 6/16 ---- subst.c - in quote_list, set W_HASQUOTEDNULL flag in the word if quote_string turns "" into CTLNUL - in dequote_list, turn off W_HASQUOTEDNULL flag in the word if dequote_string turns CTLNUL into "" - new function, string_list_pos_params, encapsulates everything needed to turn the positional parameters or an array indexed with '@' or '*' into a string, including taking care of quoting and using the first char of $IFS, when used in another expansion like pattern removal or pattern substitution - change list_remove_pattern, pos_params, pos_params_pat_subst to call string_list_pos_params. Fixes problems reported by Stephane Chazelas 6/22 ---- variables.h - include assoc.h for associative arrays - defines for case-modifying expansions and associative array variables - sh_var_assign_func_t functions now take an extra char * parameter 6/25 ---- variables.c - change declarations and definitions of sh_var_assign_func_t functions to add the extra char * parameter: null_assign, null_array_assign, assign_seconds, assign_random, assign_lineno, assign_subshell, assign_dirstack - change calls to var->assign_func to add extra char * argument - broke part of body of dispose_variable out into a new function, dispose_variable_value, which knows how to free all kinds of shell variable data - changes to deal with variables with the internal `nofree' attribute arrayfunc.c - change calls to var->assign_func to add extra char * argument - bind_array_var_internal now takes an extra `char *key' argument - additions for associative array implementation; from bash-4.0-devel tree arrayfunc.[ch],subst.c - expand_compound_array_assignment now takes the variable as the first argument (SHELL_VAR *); changed function definition and callers builtins/set.def - changes to handle associative arrays in `unset' {execute_cmd,command}.h - definitions for coproc implementation; from bash-4.0-devel tree variables.c - new functions for associative arrays: make_new_assoc_variable, make_local_assoc_variable 6/26 ---- variables.c - more infrastructure for associative arrays; from bash-4.0-devel tree - infrastructure for handling assignments to variables with case-modifying attributes; from bash-4.0-devel tree config.h.in - add #defines controlling case-modifying variable attributes and word expansions configure.in - add enable options for case-modifying variable attributes and word expansions (--enable-casemod-attributes and --enable-casemod-expansions, respectively); from bash-4.0-devel tree execute_cmd.c - add code to fix_assignment_words to handle assignment statements to "assignment builtins" that seem to be associative arrays. Imperfect subst.c - array_remove_pattern now takes a SHELL_VAR * as its first argument instead of an ARRAY *; from the bash-4.0-devel tree - changes to array_length_reference for associative arrays; from the bash-4.0-devel tree - changes to get_var_and_type for associative arrays; from the bash-4.0-devel tree - changes to parameter_brace_substring for associative arrays; from the bash-4.0-devel tree - changes to param_expand for associative arrays; from the bash-4.0-devel tree builtins/declare.def - changes for associative arrays: new `-A' option, changes to make local and global associative array variables; from the bash-4.0-devel tree 6/27 ---- execute_cmd.c - in execute_command_internal, when short-circuiting execution because `breaking' or `continuing' is non-zero, preserve the exit status by returning `last_command_exit_value' instead of an unconditional EXECUTION_SUCCESS. Fixes bug reported by Roman Rakus 6/28 ---- variables.c - fix get_var_and_type to appropriately handle references like ${varname[0]}, where `varname' is a scalar variable make_cmd.[ch],parse.y - make_here_document now takes a second argument: the current line number; changed caller (gather_here_documents) builtins/setattr.def - added support for associative arrays and the `-A' variable attribute option; from the bash-4.0-devel tree subst.c - change code that transforms `declare -A xxx=(yyy)' to perform the internal `declare -A xxx' before doing the variable assignment, because associative arrays have to be declared before being assigned to as such; uses new function make_internal_declare 6/30 ---- subst.[ch] - dequote_escapes is now external; add declaration in subst.h - remove_quoted_nulls is now external; add declaration in subst.h array.[ch] - new functions for completeness: array_dequote, array_dequote_escapes, array_remove_quoted_nulls - array_subrange now calls array_remove_quoted_nulls for "${array[*]}". Fixes bug reported by Vitor De Araujo - array_patsub now calls array_remove_quoted_nulls for "${array[*]}" - array_modcase now calls array_remove_quoted_nulls for "${array[*]}" - array_patsub now handles the mflags&MATCH_QUOTED case appropriately (that implies "${array[@]}") subst.c - new functions for case-modifying word expansion suppport: pos_params_casemod, parameter_brace_casemod; from bash-4.0-devel branch assoc.c - new functions for completeness: assoc_remove_quoted_nulls - assoc_patsub now calls assoc_remove_quoted_nulls for "${assoc[*]}" - assoc_modcase now calls assoc_remove_quoted_nulls for "${array[*]}" - assoc_patsub now handles the mflags&MATCH_QUOTED case appropriately (that implies "${assoc[@]}") 7/1 --- assoc.[ch] - new function, assoc_subrange: takes a hash table, converts it to a word list, and performs the subrange and indexing on that list - new functions for completeness: assoc_dequote, assoc_dequote_escapes subst.c - verify_substring_values now takes the variable SHELL_VAR * as its new first argument; changed callers - change verify_substring_values to handle associative arrays using the number of elements as the upper bound - brought in code to do case-modifying word expansions from bash-4.0-devel branch, conditional on CASEMOD_EXPANSIONS input.c - if the read(2) in getc_with_restart returns -1/EAGAIN, turn off non-blocking mode on the file descriptor and try again. Fixes problem reported by Glynn Clements 7/2 --- doc/{bash.1,bashref.texi} - documented new case-modifying word expansions make_cmd.c - change make_here_document to display a warning message including the start line of a here document if it ends up delimited by EOF. Addresses issue raised by Richard Neill subst.c - in do_assignment_internal, make sure the `invisible' attribute is unset before returning success 7/3 --- config-top.h - add `CASEMOD_CAPCASE' define to include or exclude the ~[~] word expansion and the `capcase' variable attribute (declare -c) builtins/declare.def - add support for manipulating the case-modifying attributes (new declare -clu); from bash-4.0-devel branch builtins/setattr.def - add support for reporting case-modifying attributes (-clu attributes); from bash-4.0-devel branch doc/{bash.1,bashref.texi} - specify that the read builtin timing out results in a return value greater than 128 - document new `-l' and `-u' options to declare/typeset/local. Leave `-c' undocumented for now 7/4 --- make_cmd.[ch] - make_coproc_command: construct a coproc; from bash-4.0-devel tree dispose_cmd.c - dispose coproc command; from bash-4.0-devel tree copy_cmd.c - copy a coproc command; from bash-4.0-devel tree print_cmd.c - print a coproc command; from bash-4.0-devel tree shell.c - dispoe the current coproc on shell exit; from bash-4.0-devel tree redir.c - when closing redirects as part of user redirections, check whether or not active coprocess fds are being closed and close the coproc if so; from bash-4.0-devel tree config.h.in - add define for COPROCESS_SUPPORT to include coprocesses configure.in - add support for configuring coprocesses into and out of the build jobs.c - in waitchld, check whether or not a coproc processs has exited; from the bash-4.0-devel tree 7/5 --- doc/bashref.texi - document new --enable-coprocesses option that includes coprocess support execute_cmd.c - add functions for coprocess support, including execute_coproc and code to call it when command->type == cm_coproc; from bash-4.0-devel tree lib/sh/fdprintf.c - new library function fdprintf(int fd, const char *format, ...); printf to a file descriptor {configure,config.h}.in - support for detecting fdprintf and compiling in replacement Makefile.in,lib/sh/Makefile.in - add rules to include fdprintf.o doc/{bash.1,bashref.texi} - documented coprocesses and `coproc' reserved word 7/7 --- subst.c - fix array_length_reference to use MB_STRLEN instead of STRLEN, so multibyte characters in array values are computed correctly. Fixes bug reported by Wang Xin 7/10 ---- jobs.c - new function, maybe_give_terminal_to (old, new, flags), sets the terminal pgrp to NEW if and only if it's currently set to OLD - call maybe_give_terminal_to when the parent sets the terminal pgrp to the pipeline pgrp in stop_pipeline, so we don't give the terminal to the new job's pgrp unless it's currently owned by the shell. Fixes race condition described by Joe Peterson , where parent bash may change tty pgrp after a grandchild (interactive bash child of su) has changed it to something else. The call to maybe_give_terminal_to makes explicit a previously-implicit assumption aclocal.m4 - remove dependency on writable /tmp by creating directories in build directory shell.c - make changes to how bash sets no_line_editing and running_under_emacs to deal with various emacs terminal emulators; use better check for `eterm', since bash sends $PWD to eterm with control sequences that confuse other programs. Problem reported by Micah Cowan 7/12 ---- print_cmd.c - break code that prints here-documents into two functions: print_heredoc_header, which prints the operator and delimiter, and print_heredoc_body, which prints the body text and closing delimiter - change print_redirection to call print_heredoc_{header,body} - sentinel variable, printing_connection, used when printing a command of type `connection' (|, &&, ||, etc.) - change print_redirection_list to save any here documents it finds while printing a connection and save them in `deferred_heredocs' - new function, print_deferred_heredocs, called from print_redirection in the cm_connection case, calls print_heredoc_header for all the here documents, then prints the operator (|, &&, ||, etc.), then the here-document body. This preserves syntactic correctness; the old code printed the control operator after the body of the here document. Fixes bug reported by 7/16 ---- locale.c - in set_locale_var, print a warning message if setlocale() fails any time it's called -- required some code restructuring 7/19 ---- support/shobj-conf - support for mingw32, contributed by Carlo Bramix 7/23 ---- execute_cmd.c - added support (currently unused) to manage a list of coprocs 7/25 ---- bashline.c - add extern declarations for literal_history and force_append_history builtins/shopt.def - include "bashhist.h" instead of having extern declarations for the appropriate history variables parser.h - new parser_state value: PST_HEREDOC, set when reading body of here- document in parse.y:read_secondary_line parse.y - set PST_HEREDOC bit in parser_state when reading a secondary line for the body of a here-document - change read_secondary_line to save lines in the body of a here- document in the shell history list if remember_on_history is set. Fixes bug reported by Gene Golub 8/4 --- configure.in - changed to 4.0-alpha lib/readline/readline.h - changed constants to reflect readline-6.0 version 8/11 ---- lib/readline/signals.c - make sure we don't use SIGWINCH without checking whether or not it's defined. Fix from Pedro Alves 8/12 ---- COPYING - updated to GPLv3; edits in every file with a copyright or license declaration to update to gpl3 version.c - update extended version info to latest gnu standard 8/17 ---- subst.c - change exp_jump_to_top_level to only call top_level_cleanup if parse_and_execute_level is 0. If it's not, the longjmp to parse_and_execute will run the unwind-protect stack. Fixes bug most recently reported by Roman Rakus 8/18 ---- support/config.{guess,sub} - updated to newer versions from autoconf-2.62 distribution 8/20 ---- subst.c - fixed parameter_brace_substring to differentiate between indexed and associative arrays when computing second offset, instead of assuming indexed array 8/21 ---- support/xcase.c - simple program to convert input from lower to uppercase and vice versa. Now used by coproc test suite, since `tr -u' is not portable. 8/22 ---- doc/bash.1 - fixed description of the bindable edit-and-execute commands to note they check $VISUAL first, instead of $FCEDIT. Fixed bug reported by [bash-4.0-alpha frozen] 8/28 ---- [bash-4.0-alpha released] 9/1 --- builtins/evalstring.c - fixed typo in parse_string (ostring used uninitialized). Bug reported by Andreas Schwab subst.c - fix return value of parameter_brace_expand to set the W_HASQUOTEDNULL flag in the returned WORD_DESC * if the return value from parameter_brace_remove_pattern is a quoted null string. Fixes bug reported by Andreas Schwab - set the W_HASQUOTEDNULL flag in the return value from parameter_brace_expand if the return value from parameter_brace_patsub is a quoted null string 9/6 --- builtins/read.def - change read -t 0 to return success if there is input available to be read -- allows scripts to poll for input. Uses input_avail libsh function 9/9 --- externs.h - fix extern fpurge declaration -- use HAVE_DECL_FPURGE instead of NEED_FPURGE_DECL, since the former is set by `configure' jobs.h - add extern declaration for close_pgrp_pipe - add a new job state JNONE (-1) to the enum jobs.c - include execute_cmd.h for extern declarations for coproc functions subst.c - include builtins/builtext.h for extern declarations for functions implementing builtins (e.g., declare_builtin) arrayfunc.c - include "pathexp.h" for extern declaration for glob_char_p braces.c - add extern declaration for `asprintf' lib/readline/rlprivate.h - add extern declarations for _rl_trace, _rl_tropen lib/sh/zgetline.c - add extern declarations for zread, zreadc lib/sh/mktime.c - include "bashansi.h" for string function declarations builtins/common.h - add extern declaration for parse_string trap.c - include jobs.h for extern declaration for run_sigchld_trap general.c - fix call to strtoimax in legal_number; if ep == string when function returns, the number was not converted, even if errno is not set. Fix from Paul Jarc 9/11 ---- [prayers for the victims of 9/11/2001] builtins/return.def - call no_options, as Posix requires. This also has the effect of disallowing negative return values unless they're prefixed by `--' 9/13 ---- builtins/bind.def - add an error message when bind is used without line editing active, instead of just returning an error status variables.c - make sure make_local_variable never creates visible variables with a value, whether or not a variable with the same name existed in a previous context. This is consistent with ksh93. Fix from 9/16 ---- execute_cmd.c - add call to CHECK_TERMSIG in shell_execve after the call to execve returns. Recommended by Roman Rakus - add QUIT check in execute_connection after executing first command in a `&' connection 9/22 ---- execute_cmd.c - new semaphore variable, executing_list, incremented every time a list (command1;command2 or command1 || command2 or command1 && command2) is executed; used as sentinel for rest of shell sig.c,builtins/evalstring.c - set executing_list to 0 when throwing execution back to top level; make sure to unwind-protect it in appropriate places jobs.c - if a pipeline is killed by SIGINT while executing a list (when executing_list is non-zero), make sure the shell acts as if an interrupt occurred. The behavior is dependent on the shell compatibility level being > 32 (bash-4.0 and above) 9/23 ---- redir.c - don't bother reporting an error with a file descriptor, even if the errno is EBADF, if the redirection error (e.g., NOCLOBBER) can't have anything to do with the fd. Fixes bug reported by "David A. Harding" , debian bug #499633. 9/24 ---- builtins/declare.def - make `declare [option] var' (and the `typeset' equivalent) create invisible variables, instead of assigning the null string to a visible variable. Fixes bug reported by Bernd Eggink 9/25 ---- builtins/common.[ch] - new function, builtin_warning(), like builtin_error but for warning messages builtins/bind.def - experimental: print a warning, but go on, if line editing not active when bind is invoked. Suggested by Rocky Bernstein 10/3 ---- test.c - use same_file instead of directly comparing st_dev and st_ino when comparing files in filecomp(). From mingw32 patches submitted by Hector Chu 10/4 ---- redir.c - in redirection_error(), use `error' instead of errno when comparing against EBADF. From mingw32 patches submitted by Hector Chu shell.c - in unset_bash_input(), reset bash_input.type to st_none after closing the default buffered fd. From mingw32 patches submitted by Hector Chu builtins/cd.def - ignore CDPATH when in privileged mode. Suggested by Paul Jarc variables.c - change sv_globignore to only act if privileged mode is not enabled. Suggested by Paul Jarc doc/bash.1,bashref.texi - document new treatment of CDPATH and GLOBIGNORE when privileged mode is enabled builtins/read.def - change prompt printing to occur after terminal is set to no-echo mode. Based on suggestion from Stephane Chazelas lib/readline/signals.c - new variables to keep track of special characters corresponding to SIGINT, SIGQUIT, and SIGTSTP - new variable to keep track of whether tty is echoing control characters corresponding to SIGINT, SIGQUIT, and SIGTSTP - new function, _rl_echo_signal_char(int sig) to display the tty special char generating SIGINT, SIGQUIT, or SIGTSTP. Based on idea and code from Joe Peterson - call rl_echo_signal_char in rl_signal_handler: if the terminal settings indicate it, readline will echo characters that generate keyboard signals lib/readline/rltty.c - set _rl_intr_char, _rl_quit_char, and _rl_susp_char to special characters that generate signals from keyboard - set _rl_echoctl if ECHOCTL tty flag is set lib/readline/rlprivate.h - extern declarations for _rl_intr_char, _rl_quit_char, and _rl_susp_char - extern declaration for _rl_echoctl lib/readline/readline.h - extern declaration for rl_echo_signal_char() lib/readline/doc/rltech.texi - document rl_echo_signal_handler(): available for applications that install their own signal handlers 10/5 ---- execute_cmd.c - fix errexit logic to not cause the shell to exit when a command in a pipeline fails. Fixes bug reported by Marcin Owsiany 10/14 ----- builtins/evalstring.c - don't short-circuit execution in parse_and_execute if we want to run an exit trap. Fixes bug reported by Steffen Kiess 10/18 ----- parse.y - fix error production to only call YYACCEPT if the shell is currently interactive and not in parse_and_execute (so parser errors in things like eval will correctly set $?). Fixes bug reported by marco-oweber@gmx.de execute_cmd.c - make sure variable name errors in execute_for_command and non- identifier function names in execute_intern_function set the return status to EX_BADUSAGE (2), not EX_USAGE (258) parser.h - new parser state, PST_REPARSE parse.y - turn PST_REPARSE on in parse_string_to_word_list - in parse_matched_pair, if parsing a single-quoted string and PST_REPARSE is set, don't requote CTLESC or CTLNUL. Fixes bug with compound array assignment using $'\x7f' reported by Antonio Macchi 10/23 ----- configure.in - define LOCAL_LDFLAGS as `-z interpose' on Solaris 8, 9, and 10 to allow the bash malloc to interpose the libc malloc when called by library functions pre-bound to the libc malloc. Suggested by Serge Dussud 10/26 ----- doc/bash.1 - add single-sentence descriptions to rest of parameter expansions. Suggested by Ken Irving 10/27 ----- subst.c - rearrange code in skip_to_delims to allow quote characters and other shell expansion characters to be delimiters - add new flags value for inverting search: skip to the next character NOT in the set of delimiters passed as an argument subst.h - define for new SD_INVERT flag value for skip_to_delims 10/28 ----- bashline.c - new bindable functions: shell-forward-word and shell-backward-word. Like forward-word and backward-word, but understand shell quoting and use shell metacharacters and whitespace as delimiters. Suggested by Andre Majorel - new bindable functions: shell-kill-word and shell-backward-kill-word. Like kill-word and backward-kill-word, but understand shell quoting and use shell metacharacters and whitespace as delimiters. Suggested by Andre Majorel doc/bash.1,lib/readline/doc/rluser.texi - documented shell-forward-word and shell-backward-word - documented shell-kill-word and shell-backward-kill-word 11/1 ---- redir.c - add extra argument to add_undo_redirect: fdbase. FD used to save a file descriptor must be > fdbase if fdbase >= SHELL_FD_BASE. A value of -1 for fdbase means to just use SHELL_FD_BASE. Fixes bug with 0<&10 reported by Clark Jian Wang 11/5 ---- unwind_prot.c - new function: have_unwind_protects(); returns 1 if unwind_protect_list is not empty unwind_prot.h - extern declaration for have_unwind_protects builtins/evalstring.c - in parse_and_execute_cleanup, make sure that we don't call run_unwind_frame and expect it to decrement parse_and_execute_level if there's no unwind_protect_list, since there's a while loop in throw_to_top_level that calls parse_and_execute_cleanup as long as parse_and_execute_level is non-zero 11/9 ---- variables.c - fix the assign function for COMP_WORDBREAKS to allocate new memory to store as the variable's value, to avoid freeing memory twice if the variable is unset after rl_completer_word_break_characters is freed and reallocated. Fix from Mike Stroyan 11/20 ----- general.c - new 'file_exists(fn)' primitive; just calls stat(2) general.h - new extern declaration for file_exists bashline.c - add `~' to rl_filename_quote_characters so make_quoted_replacement will call bash_quote_filename for words containing `~'. Then bash_quote_filename can make choices based on that - change quote_word_break_chars to backslash-quote the tilde in a filename with a leading tilde that exists in the current directory, since we want to inhibit tilde expansion in this case execute_cmd.c - call file_isdir from shell_execve instead of stat(2) directly bashhist.c - use file_exists and file_isdir primitives instead of calling stat 11/21 ----- redir.c - When undoing saving of non-standard file descriptors (>=3) using file descriptors >= SHELL_FD_BASE, we set the saving fd to be close-on-exec and use a flag (RX_SAVCLEXEC) to decide how to set close-on-exec when the fd is restored. Set flag in add_undo_redirect, check in do_redirection_internal. Fixes problem reported by Andreas Schwab 11/26 ----- subst.c - fix param_expand to have expansions of $@ and $* exit the shell if there are no positional parameters and `set -u' is enabled. Fixes bug reported by Dan Jacobson 11/27 ----- lib/readline/display.c - fix update_line to not call space_to_eol if current cursor position (_rl_last_c_pos) indicates that we're already at end of line. Partial fix for bug reported by Mike Frysinger - in update_line, don't call insert_some_chars if that will start before the last invisible character in the prompt string and not draw the entire prompt string. More of the partial fix for bug reported by Mike Frysinger - fix update_line to adjust _rl_last_c_pos by wrap_offset when adding characters beginning before the last invisible character in the prompt. New code is same as previously existed in a different code path. Rest of fix for bug from Mike Frysinger - fix assignment of newline breaks (inv_lbreaks) to correctly account for prompts longer than two screen lines containing invisible characters. The assumption is that part of the invisible characters are on the first line (prompt_invis_chars_first_line) and the remainder are on the last (wrap_offset - prompt_invis_chars_first_line). Fix is in rl_redisplay. part of fix for bug reported by "Wesley J. Landaker" in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=265182 [TENTATIVE] - fix _rl_move_cursor_relative to correctly offset `dpos' by `woff' when there are invisible characters on lines after the second by using (_rl_screenwidth*_rl_last_v_pos) when seeing whether or not we just wrote some invisible characters. Rest of fix for bug reported in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=265182 [TENTATIVE] 12/11 ----- sig.c - reset the execution context before running the exit trap in termsig_handler general.c - set and unset terminate_immediately like interrupt_immediately in bash_tilde_expand builtins/read.def - change terminate_immediately to a counter instead of a flag, as interrupt_immediately is used lib/readline/display.c - slight change to fix from 11/27 to deal with prompts longer than a screen line where the invisible characters all appear after the line wrap. Fixes bug reported by Andreas Schwab builtins/{echo,printf}.def - increment terminate_immediately at entry; decrement before returning. Fix for bug reported by Ralf.Wildenhues@gmx.de 12/16 ----- subst.c - fix off-by-one error in /dev/fd version of add_fifo_list; make sure we add to totfds when it is == fd, not just when fd > totfds. Fixes bug reported by marciso@gmail.com [bash-4.0-beta2 frozen] 12/29 ----- doc/{bash.1,bashref.texi} - document more clearly that when not in Posix mode, command substitution does not inherit the -e option. From bug report from Freddy Vulto {execute_cmd,sig,builtins/evalstring}.c - sentinel variable to keep track of whether or not we're supposed to ignore the failure status of a command executed in a command substitution even if the `-e' option is set: comsub_ignore_return - increment and decrement comsub_ignore_return in execute_simple_command before calling expand_words - in parse_and_execute, if comsub_ignore_return is non-zero and the SUBSHELL_COMSUB bit is set in subshell_environment, enable the CMD_IGNORE_RETURN flag in every command executed from the passed string. Fixes problem reported by Freddy Vulto - make sure to reset comsub_ignore_return every time we throw to the top level, like executing_list flag 1/2/2009 -------- parse.y - fix to rewind_input_stream to handle case of $(...) command substitution followed by a quoted literal newline. Report and fix from Andreas Schwab 1/7 --- subst.c - fix match_wpattern and match_upattern to prefix a `*' to the pattern even if it starts with a `*(' (if extglob is enabled) before checking whether or not it can match anywhere in the string. Fixes bug reported by os@sernet.de. [bash-4.0-rc1 frozen] 1/9 --- locale.c - since setlocale() doesn't set errno to anything meaningful, don't include the strerror() result in the error message if it fails - make sure the error messages printed when setlocale fails are localizable 1/11 ---- lib/readline/histexpand.c - make sure that every time history_no_expand_chars is tested, we also call the history_inhibit_expansion_function if it's set. Fixes bug reported by Yang Zhang 1/12 ---- trap.c - make sure to call parse_and_execute with the SEVAL_RESETLINE bit set in the flags so it will reset the line number when running the trap commands. Partial fix for bug reported by peter360@fastmail.us 1/14 ---- builtins/reserved.def - document `coproc' so it can be used with `help' builtin. Pointed out by Pierre Gaston lib/sh/casemod.c - added two new flags: CASE_UPFIRST and CASE_LOWFIRST to casemod the first character of the passed string and pass the rest through unchanged. Fixes bug reported by Jan Schampera externs.h - new defines for CASE_UPFIRST and CASE_LOWFIRST subst.c - use CASE_UPFIRST for ^ and CASE_LOWFIRST for , casemod operators builtins/mapfile.def - call zreset() before calling first zgetline(), to clean out any remaining data in local buffer used by zreadc. Fixes bug reported by Pierre Gaston 1/15 ---- lib/sh/zread.c - renamed zreadintr to zreadretry -- not perfect, but better - new functions: zreadintr, which just calls read so it can be interruptible, and zreadcintr, which is like zreadc but uses zreadintr to fill the buffer lib/sh/zgetline.c - in zgetline, when zread/zreadc return <= 0, make sure line is non-null before assigning to line[nr] builtins/mapfile.def - return an error right away if the supplied array variable name refers to a readonly or noassign array - set interrupt_immediately so calls to zgetline can be interrupted. Fixes bug reported by Pierre Gaston - if interactive, pass the SEVAL_INTERACT and SEVAL_NOHIST flags to parse_and_execute when calling callbacks. Fixes bug reported by Pierre Gaston - add `readarray' as a synonym for mapfile doc/{bash.1,bashref.texi} - document behavior of mapfile builtin adding index of array element to be assigned as additional argument to callback string. Reported by Pierre Gaston - document readarray as synonym for mapfile builtins/common.c - new error function, sh_ttyerror(set), prints an error message having to do with setting or getting terminal attributes builtins/read.def - print error message if read fails to set terminal attributes 1/16 ---- execute_cmd.c - new function, coproc_reap, calls coproc_dispose if sh_coproc is marked as COPROC_DEAD - new function, cpl_reap, disposes coprocs marked as COPROC_DEAD from coproc list - change coproc_pidchk to just mark the coproc as dead instead of calling coproc_dispose, so we don't call unsafe functions from a signal handler. Fixes bug reported by Andreas Schwab execute_cmd.h - new extern declaration for coproc_reap command.h - new flags for c_flags member of a struct coproc {jobs,nojobs}.c - add call to coproc_reap in cleanup_dead_jobs, which will do the right queueing or blocking of SIGCHLD trap.c - modify change from 1/12 to not reset the line number when running the DEBUG and RETURN traps 1/18 ---- lib/sh/casemod.c - change default operations to work on entire passed string instead of breaking into words at non-alpha-numerics. Use new CASE_USEWORDS flag to enable by-word behavior. Fixes bug reported by Jan Schampera builtins/printf.def - in vbprintf, bracket each call to vsnprintf (which uses the args passed to vbprintf) with SH_VA_START and va_end, so we can reninitialize the argument list for each call. This is actually what the C standard requires. Fixes bug that caused printf -b to `ignore' first % format specifier if it came first in the string. Reported by David Leverton builtins/mapfile.def - start the line count at 1, since it doesn't get incremented before (or after) reading the first line, so things like `mapfile -n 5 -c 1 -C 'echo foo' array < file' work right and call the callback after the first line is read. Fixes bug reported by Pierre Gaston 1/22 ---- lib/readline/complete.c - set _rl_interrupt_immediately non-zero before reading from the file system or calling an application-defined completion function lib/readline/signals.c - renamed rl_signal_handler to _rl_handle_signal; new version of rl_signal_handler that just calls _rl_handle_signal (for now) - new function _rl_signal_handler that calls _rl_handle_signal without any checking lib/readline/rlprivate.h - new extern declaration for _rl_signal_handler - new define, RL_CHECK_SIGNALS, checks whether or not _rl_caught_signal is set and calls _rl_signal_handler if so lib/readline/{bind,input,readline}.c - add RL_CHECK_SIGNALS in appropriate places lib/readline/signals.c - change rl_signal_handler to set a flag and return rather than run through the entire signal handling process. If _rl_interrupt_immediately is set, call the signal handling code right away instead of setting the flag. Initial fix for crash bug reported by Roman Rakus aclocal.m4 - new macro, BASH_TYPE_SIG_ATOMIC_T, tests for sig_atomic_t in , defines as int if not defined configure.in - call BASH_TYPE_SIG_ATOMIC_T - call AC_C_VOLATILE config.h.in - empty define for sig_atomic_t - empty define for volatile 1/27 ---- subst.c - audit calls to add_character and change to add_ifs_character (which quotes characters in $IFS). Affects primarily `:', `=', and `~'. Fixes bug reported by Jan Schampera ; fix suggested by Stephane Chazelas 2/1 --- configure.in - call AC_C_RESTRICT config.h.in - add empty defintion for `restrict' pcomplete.c - use unwind_protects around call to execute_shell_function in gen_shell_function_matches to prevent data corruption if throw_to_top_level is called. Bug report and fix from werner@suse.de. execute_cmd.c - don't clamp CPU usage at 100% in print_formatted_time. Bug reported by Linda Walsh 2/5 --- locale.c - in set_locale_var, set errno to 0 before calling setlocale(), and print strerror (errno) if setlocale fails and errno ends up non-zero 2/6 --- configure.in - backed out of solaris change from 10/23/2008 (adding `-z interpose' to LDFLAGS) due to solaris updates to fix a linker problem. Updatted by Serge Dussud 2/12 ---- execute_cmd.c - change execute_connection so failure of a pipeline will cause the shell to exit if -e is on. From discussion on austin-group mailing list - change execute_command_internal so failure of a user-specified subshell will cause the shell to exit if -e is on. From discussion on austin-group mailing list 2/13 ---- doc/{bash.1,bashref.texi} - clarified description of set -e option to accurately reflect current implementation 2/19 ---- print_cmd.c - fix print_deferred_heredocs to not print a space if the separator string is null - change print_deferred_heredocs to set `was_heredoc' after printing something - change connection printing code to only print the `;' separator if we haven't just printed a here-document - change connection printing code to print any deferred here documents after the rhs of the connection. Fixes bug reported by Bo Andresen [bash-4.0 frozen] 2/20 ---- [bash-4.0 released] 2/22 ---- parse.y - fix parse_comsub to not test a character for being a possible shell metacharacter if LEX_PASSNEXT flag is set. Fixes bug reported by Mike Frysinger pcomplete.c - add call to save_parser_state (accidentally dropped from patch) to gen_shell_function_matches. Fixes bug with bash_completion and file/directory completion reported by phil@Arcturus.universe Makefile.in - fix assignment to LDFLAGS_FOR_BUILD to match those in subdir Makefiles. Fixes bug reported by Mike Frysinger builtins/mapfile.def - make sure the callback quantum (-c option argument) is > 0. Fixes bug reported by Stephane Chazleas 2/23 ---- parse.y - fix save_token_state and restore_token_state to save and restore current_token. Fixes bug reported by Bernd Eggink builtins/exit.def - check jobs[i] before checking whether or not it's running when the checkjobs option is set and we're looking for running jobs at exit. Fixes bug reported by Mike Frysinger 2/24 ---- siglist.c - include bashintl.h for definition of _. Fixes bug reported by Greg Wooledge 2/25 ---- subst.c - new function, skip_matched_pair. Similar to skip_to_delim and the extract_XXX family - move skipsubscript here from arrayfunc.c; re-implement in terms of skip_matched_pair. Fixes bugs reported by arrayfunc.c - remove skipsubscript; moved to subst.c parse.y - change reset_parser to set current_token to '\n'. Rest of fix for bug reported by Bernd Eggink ; earlier fix on 2/23 2/26 ---- builtins/declare.def - when given something like array[x]=y (which sets making_array_special to 1), don't convert an associative array to an indexed array (line 493). Part of fix for bug reported by Pierre Gaston - if offset == 0, indicating that we do not have a valid assignment, make sure any `name' containing a `[' is a valid array reference before trying to go on. Not doing this leads to creating crazy variables like `name[foo[bar]=bax'. Rest of fix for bug reported by Pierre Gaston assoc.c - change assoc_to_assign to single-quote the array keys if `quoted' is non-zero. Makes things easier to read with weird characters in the key parse.y - fix parse_comsub to not set LEX_HEREDELIM when it sees "<<<". Fixes bug reported by Mike Frysinger 2/27 ---- parse.y - fix report_syntax_error to set last_command_exit_value to EX_BADUSAGE (2) instead of EX_USAGE (258), since there's nothing that will translate that to something < 128 before reading the next command. Partial fix for bug reported by Mike Frysinger sig.c - fix sigint_sighandler to set last_command_exit_value to sig+128 before calling throw_to_top_level. Rest of fix for bug reported by Mike Frysinger jobs.c - if fork() fails, set last_command_exit_value to 126 before calling throw_to_top_level execute_cmd.c - defer calling unlink_fifo_list in parent branch of execute_disk_command if we're executing in a shell function - change execute_function to call unlink_fifo_list before returning if it's the top-level function 3/2 --- builtins/read.def - if read times out, make sure we remove the top element from the unwind-protect stack (the free of input_string) and run the rest, to reset the tty and readline and alarm states. Then we jump to assigning the variables to any partial input. Fixes bug reported by Christopher F. A. Johnson 3/3 --- parse.y - break comment checking code into a common COMMENT_BEGIN define so we can use it in multiple places in parse_comsub - in parse_comsub, don't alter the LEX_RESWDOK flag if we read a `#' and we're checking comments, even though `#' isn't a `shell break' character. Fixes bug reported by Mike Frysinger braces.c - in expand_seqterm, decrease the total length of the rhs by the length of any (optional) increment, so we don't end up with unwanted zero padding because the rhs length is wrong. Fixes bug reported by Carl Albing 3/4 --- doc/{bash.1,bashref.texi} - changes to clean up some of the language describing the effects of terminal process groups on the ability to read from and write to the terminal 3/5 --- support/shobj-conf - add host_vendor to string tested in switch to handle things like gentoo/freebsd - beginning with version 7, FreeBSD no longer has /usr/bin/objformat or a.out binaries and libraries. It's always ELF. Fix from Timothy Redaelli parse.y - in parse_comsub, allow comments if we are ready to read a reserved word (tflags & LEX_RESWDOK), haven't read anything from one yet (lex_rwlen == 0) and the current character is a '#' 3/6 --- parse.y - new lex flag for parse_comsub: LEX_INWORD. Turn it off when we see a shell break character; turn it on or keep it on when not a break character. Keep track of word length (reset to 0 when we turn on LEX_INWORD when it was off). - don't use COMMENT_BEGIN in parse_comsub any more; test whether or not LEX_INWORD is set and lex_wlen == 0 in addition to tests for LEX_RESWDOK and lex_rwlen. Comments are valid when at the start of a word - move LEX_PASSNEXT code to the top of parse_comsub, so the rest of the function doesn't have to check for the flag at different places 3/7 --- parse.y - in parse_comsub, when looking for a reserved word (LEX_RESWDOK non-zero), and in a case statement, we can see either an esac or a pattern list. We handle an esac separately. We should turn off LEX_RESWDOK if we see anything but a newline, since we'll be reading a pattern list. Other part of fix for bug reported by Mike Frysinger (rest of fix on 3/3) 3/10 ---- {.,lib/readline}/doc/fdl.texi - updated to FDL version 1.3 3/11 ---- parse.y - when using the |& construct with a simple command preceding it, add the implicit redirection to the simple command's redirection list, since the redirections associated with the command struct are never executed. Fixes bug reported by Matt Zyzik 3/14 ---- execute_cmd.c - in execute_case_command, if ;& is used with no following pattern list, make sure we don't reference a NULL pointer. Bug report and fix from Clark Jian Wang parse.y - make parser_state global, so other files can use it - command_word_acceptable now returns non-zero if PST_REDIRLIST bit set in parser_state, so we accept assignment statements and perform alias expansion. Fix for bug reported by Vincent Lefevre (2/24/2009) parser.h - add PST_REDIRLIST flag, notes that parser is currently parsing a redirection list preceding a simple command make_cmd.c - make_simple_command now turns on PST_REDIRLIST in parser_state when creating a new simple command - make_simple_command turns off PST_REDIRLIST in parser_state if it adds a non-redirection to the command it's building - clean_simple_command turns off PST_REDIRLIST to make sure it's off subst.c - new flag for param_expand: PF_IGNUNBOUND, means to not exit if the variable is unbound even if `set -u' is enabled - change param_expand to not call err_unboundvar if the `pflags' argument has the PF_IGNUNBOUND bit set - parameter_brace_expand_word now takes an extra `pflags' argument to pass down to param_expand; changed callers - changed call to parameter_brace_expand_word in parameter_brace_expand to add PF_IGNUNBOUND flag so ${@:-foo} doesn't cause the shell to exit (but ${@} does) when there are no positional parameters. Fixes Debian bug 519165 from Dan Jacobson parse.y - add code to parse_comsub to allow here-documents within command substitutions to be delimited by the closing right paren, with the usual warning about here documents delimited by EOF on execution. Fixes regression from bash-3.2 noted in Red Hat bugzilla 485664 by Ralf Corsepius 3/15 ---- subst.c - string_list_dollar_at now checks for Q_PATQUOTE, which getpattern() uses to denote Q_DOUBLE_QUOTES (?). Fixes a=abcd echo "${a#$*}" when IFS= and args are `a b' as noted by Stephane Chazleas - param_expand now checks for Q_PATQUOTE and treats it identically to Q_DOUBLE_QUOTES when expanding $* - expand_word_unsplit now sets W_NOSPLIT in the flags of the word it passes to expand_word_internal if $IFS is NULL - expand_word_leave_quoted now sets expand_no_split_dollar_star and the W_NOSPLIT bit in the word flags before calling expand_word_internal if $IFS is NULL, just like expand_word_unsplit. It is now virtually identical to expand_word_unsplit. Rest of fix for problems reported by Stephane Chazleas 3/20 ---- trap.c - in _run_trap_internal, don't pass SEVAL_RESETLINE as flag to parse_and_execute if running the ERR trap (further modification of change from 1/12) execute_cmd.c - in execute_simple_command, set line_number to line_number_for_err_trap before calling run_error_trap. Part of fix for bug reported by Brian J. Murrell - change other places calling run_error_trap() to set and use line_number_for_err_trap 3/21 ---- builtins/fc.def - Even though command substitution through parse_and_execute turns off remember_on_history, command substitution in a shell when set -o history has been enabled (interactive or not) should use it in the last_hist calculation as if it were on. Same calculation in fc_gethnum and fc_builtin. Fixes bug reported by Ian Kelling sig.c - change termsig_sighandler to terminate immediately if it gets called twice with the same signal before termsig_handler gets called. This fixes the `looping on SIGSEGV' phenomenon reported by Linux users. parse.y - in read_secondary_line, don't try to add NULL lines to the history list. Report and patch from Lubomir Rintel 3/22 ---- sig.c - Augment change from 3/21 with explicit check for signals we *don't* want this to happen for. Patch from Lubomir Rintel 3/28 ---- array.c - in array_reference, return NULL immediately if the desired index is larger than the maximum - add cache of last array referenced and last array element referenced; use in array_reference to optimize case of sequential access; invalidated where necessary in other functions - array_rshift needs to set max_index to 0 if the array was empty before shifting in the new element 0 - array_shift needs to use element_index(a->head->prev) to set the max_index, not a simple decrement, to deal with sparse arrays 4/1 --- bashline.c - in bash_dequote_filename, return right away after copying the backslash if the last character in the string to be expanded is a backslash. The old code copied an extra NUL and overwrote the bounds checking. Fixes bug reported by Shawn Starr https://bugzilla.redhat.com/show_bug.cgi?id=488649 4/3 --- subst.c - in pat_subst.c, make sure to copy one character from the input string in the case of a null pattern match, since we substitute on the null match and then increment past the current character. Not doing this means that each character of the original string is replaced because of the null matches. Fixes debian bug reported bhy Louis-David Mitterrand http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=522160 lib/sh/winsize.c - incorporate contents of readline/rlwinsize.h to get all the various system dependencies right when trying to find TIOCGWINSZ. Fixes bug reported by Dan Price 4/6 --- doc/{bash.1,bashref.texi} - fix description of conditional `>' and `<' to remove statement that the comparison pays attention to the current locale -- it has always used strcmp lib/glob/glob.c - fixed a bug in glob_filename that caused glob_dir_to_array to be called to prepend a (globbed) directory name onto the results from glob_vector, which, if we were globbing `**', glob_vector has already done. Effect is to have the directory name(s) on there twice. Fixes "dir*/**" bug reported by Matt Zyzik 4/8 --- doc/{bash.1,bashref.texi} - fix short syntax summary of for command to reflect full bash syntax (which is a superset of Posix syntax). Fixes bug reported by Reuben Thomas 4/10 ---- {expr,subst}.c - make sure last_command_exit_value is set to EXECUTION_FAILURE before calling err_unboundvar, in case set -e is enabled and the shell exits from there. Fixes bug reported by Freddy Vulto and Piotr Zielinski 4/11 ---- jobs.c - in restore_pipeline, don't call discard_pipeline with a NULL argument trap.c - in run_debug_trap, make sure to save and restore the pipeline, pipeline_pgrp, and state of the pipeline around running the debug trap, then remove any job created by running the debug trap from the jobs table when it completes. Fixes for two bugs reported by lex@upc.ua 4/12 ---- lib/readline/signals.c - new functions to block and release SIGWINCH like the SIGINT blocking and releasing functions lib/readline/rlprivate.h - new extern declarations for _rl_block_sigwinch and _rl_release_sigwinch lib/readline/display.c - block SIGWINCH during redisplay like SIGINT. Should fix bug reported by Nicolai Lissner 4/13 ---- lib/readline/readline.h - new readline state variable: RL_STATE_REDISPLAYING lib/readline/display.c - in rl_redisplay, don't block SIGWINCH during redisplay; just set the REDISPLAYING state lib/readline/terminal.c - in rl_resize_terminal, don't call rl_redisplay_after_sigwinch() if we're already in the middle of redisplay (RL_STATE_REDISPLAYING). Fix for bug reported by Nicolai Lissner 4/15 ---- parse.y - fix parse_comsub to add check for \n when seeing whether the current character can change to a state where a reserved word is legal, since it is not a shell meta character. Fixes bug reported by Bernd Eggink . 4/17 ---- jobs.c - new functions to save and restore the pgrp_pipe (since there's only one): save_pgrp_pipe and restore_pgrp_pipe trap.c - run_debug_trap now saves and restores the pgrp_pipe before and after calling the debug trap - run_debug_trap now makes sure the terminal is owned by the pipeline pgrp after the debug trap runs. Rest of fix for bug reported by Oleksly Melnyk (lex@upc.ca) 4/19 ---- include/posixselect.h - new include file, encapsulates select(2) includes and defines for bash and readline. Inspired by patch from Mike Frysinger lib/sh/input_avail.c - include "posixselect.h" lib/readline/{input,parens}.c - include "posixselect.h" instead of using inline includes - use new USEC_TO_TIMEVAL define to make sure that values for timeouts greater than one second are handled properly lib/sh/fpurge.c - updated implementation, taken from gnulib 4/21 ---- lib/glob/glob.c - in finddirs, don't try to free a return value of glob_error_return from glob_vector. Bug and fix from werner@suse.de lib/readline/signals.c - in rl_echo_signal_char, check that SIGQUIT and SIGTSTP are defined before trying to use them. Bug report and fix from Volker Grabsch 4/24 ---- aclocal.m4 - add conditional inclusion of to BASH_CHECK_TYPE bashtypes.h,lib/sh/strto[iu]max.c - include if present for any existing declaration of intmax_t and uintmax_t. Fixes Interix problem reported by Markus Duft lib/sh/strindex.c,externs.h,builtins/common.h - renamed strindex to strcasestr to agree with other implementations (e.g., BSD, MacOS X, gnulib); changed callers lib/sh/{strindex.c,Makefile.in},Makefile.in - renamed strindex.c to strcasestr.c configure.in - add strcasestr to call to AC_REPLACE_FUNCS, take advantage of existing libc implementations config.h.in - add define for HAVE_STRCASESTR lib/sh/mbscmp.c - fix mbscmp to return correct values when the strings do not contain valid multibyte characters. Ideas from gnulib xstrchr.c - only compare current character against C if mblength == 1 {shell,variables}.c - changed some xstrchr calls back to strchr when the arguments cannot contain multibyte characters lib/sh/{xstrchr.c,Makefile.in},Makefile.in - renamed xstrchr to mbschr; renamed file to mbschr.c aclocal.m4 - change BASH_CHECK_MULTIBYTE to use AC_REPLACE_FUNCS(mbschr) externs.h - extern declarations for mbscmp and mbschr, conditional on the usual HAVE_MBSCMP and HAVE_MBSCHR defines general.h,{alias,arrayfunc,bashline,general,execute_cmd,subst}.c - changed calls to xstrchr to mbschr doc/bash.1 - use `pathname expansion' consistently, not `filename expansion' or `filename generation' doc/bashref.texi - use the phrase `filename expansion' consistently (since this is what the Gnu people prefer) instead of `pathname expansion' or `filename generation' aclocal.m4,config.h.in - check for mbscasecmp in BASH_CHECK_MULTIBYTE, define HAVE_MBSCASECMP if found lib/sh/{mbscasecmp.c,Makefile.in} - new file, case-insensitive multibyte string comparison externs.h - extern declaration for mbscasecmp 4/25 ---- lib/readline/display.c - in _rl_move_cursor_relative, don't adjust dpos by woff if it's already less than woff (don't want it less than 0) - in _rl_move_cursor_relative, short-circuit right away if the cursor is at columns 0 and `new' is 0 (doesn't matter if it's a multibyte locale or not, or whether there are invisible chars in the prompt) - in _rl_move_cursor_relative, go ahead and adjust dpos if prompt_physical_chars >= _rl_screenwidth (previous check was just > ) Fixes bug reported by Andreas Schwab 4/28 ---- lib/glob/glob.c - in glob_vector, don't add an empty pathname ("") if we're adding the currect directory to the dirlist and GX_NULLDIR is set -- we can just ignore it, since the passed directory name (".") was created by the caller. Fixes bug reported by Matt Zyzik 5/5 --- subst.c - make expansion of $@ and $* when set -u is in effect and there are no positional parameters be a non-fatal error. This is the consensus of the austin group, though it is not historical practice. Message from Geoff Clare <20090505091501.GA10097@squonk.masqnet> of 5 May 2009 and http://austingroupbugs.net/view.php?id=155 5/20 ---- lib/glob/glob.c - tentative fix to glob_filename to compensate for glob_vector putting null pathname at front of result vector when dflags&GX_NULLDIR. Current fix manually removes empty string element from front of result vector; a better fix would be to use a flag so glob_vector doesn't add it at all. Augments patch from 4/28, which appears to have broken some things. Fixes bug reported by Matt Zyzik 5/22 ---- lib/glob/glob.c - better fix for glob_filename; supersedes patch of 5/20. Now the code does not set GX_ADDCURDIR if directory_len == 0 and the function has not been called recursively ((flags & GX_ALLDIRS) == 0). Better fix for bug reported by Matt Zyzik Makefile.in - fix build race condition that occurs in some makes caused by libreadline.a and libhistory.a containing some of the same files (e.g., xmalloc.o) and conflicting when trying to build both at the same time. Reported by Mike Frysinger 5/25 ---- lib/readline/vi_mode.c - fix _rl_vi_initialize_line so that the loop counter is not unsigned (it doesn't matter, but it eliminates a compiler warning). Bug reported by Dave Caroline 5/26 ---- doc/{bash.1,bashref.texi} - add text to the description of array variables making it clear that an array variable is not considered set until a subscript has been assigned a value 5/29 ---- lib/readline/text.c - fix rl_change_case to handle case where mbrtowc doesn't find a valid multibyte character lib/readline/vi_mode.c - fix _rl_vi_change_mbchar_case to handle case where mbrtowc doesn't find a valid multibyte character lib/sh/casemod.c - fix sh_modcase to handle case where mbrtowc doesn't find a valid multibyte character lib/readline/mbutil.c - fix _rl_find_next_mbchar_internal to not call mbrtowc at the end of the string, since implementations return different values -- just break the loop immediately lib/readline/display.c - fix rl_redisplay to make same sort of cursor position adjustments based on multibyte locale and _rl_last_c_pos when performing horizontal scrolling rather than line wrapping. Probably still more to do. Fixes bug reported by jim@jim.sh 6/5 --- doc/{bash.1,bashref.texi} - added some more explanation of the inheritance of the ERR trap at the suggestion of Thomas Pospisek findcmd.c - use eaccess(2) if available in file_status to take other file access mechanisms such as ACLs into account. Patch supplied by werner@suse.de 6/12 ---- xmalloc.c - also calculate lowest brk() value the first time xmalloc/xrealloc (and their sh_ counterparts) are called - error messages consolidated into a single function (allocerr/ sh_allocerr) to avoid string duplication 6/16 ---- variables.c - changes to allow variables.c to be compiled if ALIAS is not defined. Bug and fix from John Gatewood Ham lib/sh/getcwd.c - fix so systems defining BROKEN_DIRENT_D_INO have the necessary defines. Fix from Jay Krell configure.in - add -D_ALL_SOURCE to interix CFLAGS for struct timezone definition. Bug and fix from John Gatewood Ham 6/29 ---- variables.c - change initialize_shell_variables to add environment variables with invalid names to the variables hash table, but marking them as invisible and imported - new function, export_environment_candidate. Used when creating the export environment for commands to include variables with invalid names inherited from the initial environment. Apparently this behavior is widespread - change make_var_export_array to use export_environment_candidate rather than visible_and_exported to test variables for inclusion in the export environment 7/1 --- builtins/read.def - fix a memory leak where the number of fields is not the same as the number of variables passed to `read'. Bug report from werner@suse.de builtins/command.def - move section of code that sets PATH from -p option before the verbose-handling section, so command -v and command -V honor the PATH set by command -p. Bug report and fix from ohki@gssm.otsuka.tsukuba.ac.jp 7/9 --- subst.c - change brace_expand_word_list to defer brace expansion on compound array assignments that are arguments to builtins like `declare', deferring the expansion until the assignment statement is processed. Fixes inconsistency reported by agriffis@n01se.net 7/16 ---- bashline.c - fix bash_execute_unix_command to set rl_point correctly based on READLINE_POINT. The old method of using save_point will not work because maybe_make_readline_line will change rl_point. Bug reported by Henning Bekel trap.c - fix _run_trap_internal and run_pending_traps to save and restore value of subst_assign_varlist so the dispose_words on it doesn't leave dangling pointers after the trap handler runs. Fixes bug reported by Marc Herbert 7/22 ---- subst.c - fix off-by-one error in pos_params when computing positional parameters beginning with index 0. Bug and fix from Isaac Good 7/24 ---- lib/readline/display.c - add code to _rl_move_cursor_relative and _rl_col_width to short- circuit a few special cases: prompt string and prompt string plus line contents, both starting from 0. Saves a bunch of calls to multibyte character functions using already-computed information. As a side effect, fixes bug reported by Lasse Karkkainen subst.c - fixed a problem in split_at_delims that could leave *cwp set to -1 if the line ends in IFS whitespace and SENTINEL is one of those whitespace characters. Fixes problem with setting COMP_CWORD for programmable completion reported by Ville Skytta bashline.c - change bash_execute_unix_command to clear the current line (if the terminal supplies the "ce" attribute) instead of moving to a new line. Inspired by report from Henning Bekel builtins/printf.def - changes to allow printf -v var to assign to array indices, the way the read builtin can. Suggested by Christopher F. A. Johnson lib/readline/complete.c - fix rl_old_menu_complete and rl_menu_complete to appropriately set and unset RL_STATE_COMPLETING while generating the list of matches. Fixes debian bug #538013 reported by Jerome Reybert 7/25 ---- execute_cmd.c - change execute_builtin to temporarily turn off and restore the ERR trap for the eval/source/command builtins in the same way as we temporarily disable and restore the setting of the -e option. Fixes bug reported by Henning Garus 7/27 ---- shell.c - add fflush(stdout) and fflush(stderr) to exit_shell before closing any file descriptors at exit time (e.g., coproc pipes) 7/30 ---- lib/readline/complete.c - new function rl_backward_menu_complete, just passes negative count argument to rl_menu_complete - change rl_menu_complete to act appropriately if rl_last_command is rl_backward_menu_complete, so we can cycle forward and backward through the list of completions lib/readline/doc/{readline.3,rluser.texi},doc/bash.1 - document new "menu-complete-backward" bindable readline function. Suggested by Jason Spiro lib/readline/vi_keymap.c - add binding of C-n to menu-complete and C-p to menu-complete-backward in vi-insert keymap, as suggested by Jason Spiro pcomplete.c - fixed a bug in programmable_completions: the options it returned from the compspec it found were set before generating the completions, which meant that any changes made by "compopt" were overridden and only in effect for the duration of the executing shell function rather than the entire completion. Fixes bug reported by Ville Skytta 7/31 ---- lib/readline/keymaps.c - fixed memory leak in rl_discard_keymap by freeing storage associated with hierarchical keymaps - new convenience function, rl_free_keymap, that calls rl_discard_keymap and frees the keymap passed as an argument lib/readline/util.c - new bindable keymap function, _rl_null_function, to be used internally lib/readline/rlprivate.h - extern declaration for _rl_null_function lib/readline/bind.c - fix rl_generic_bind in the case where we are trying to override a keymap with a null function (e.g., when trying to unbind it). We can't use a NULL function pointer in ANYOTHERKEY since that's indistinguishable from the keymap not having been overridden at all. We use _rl_null_function instead, which simply does nothing. We could add an rl_ding to it later. Fixes problem with hitting ESC repeatedly while in vi command mode reported by James Rowell builtins/bind.def - call rl_bind_keyseq instead of rl_set_key for -r option lib/readline/readline.c - Set vi_movement_keymap[ESC] to _rl_null_function after binding the arrow keys in bind_arrow_keys() to allow vi-mode users to hit ESC multiple times in vi command mode while still allowing the arrow keys to work 8/2 --- bashline.c - fix clear_hostname_list by setting hostname_list_initialized to 0 after freeing all list members. Fixes bug reported by Freddy Vulto lib/readline/display.c - in update_line, if we copy data from one line to another because we are wrapping a multibyte character from, say, the first line to the second, we need to update OMAX and the line indices to account for the moved data. Bug report and fix from Martin Hamrle 8/3 --- pcomplete.h - defines for EMPTYCMD ("_EmptycmD_") and DEFAULTCMD ("_DefaultCmD_") builtins/complete.def - change compopt_builtin to make -E work on the "empty" command completion - fix print_compitem and print_compopts to replace EMPTYCMD with -E - added -D (default) option to complete/compgen/compopt. No supporting code yet doc/bash.1,lib/readline/doc/rluser.texi - document new -D, -E options to compopt - document new -D option to complete/compgen shell.h - new define, EX_WEXPCOMSUB, value of 125 - new define, EX_RETRYFAIL, value of 124 (for programmable completion) subst.c - use EX_WEXPCOMSUB instead of literal 125 as exit status when a shell invoked to run wordexp(3) with the -n option supplied attempts a command substitution pcomplete.c - new define, PCOMP_RETRYFAIL, used to indicate a "failure, retry with next completion" status to the programmable completion code 8/4 --- pcomplete.c - changed gen_shell_function_matches to take an extra parameter indicating whether the specified shell function was not found or returned the special "fail/retry" status, and, if it was either, to not bother returning any matches list - changed gen_compspec_completions to take an extra parameter to pass through the "found" status from gen_shell_function_completions - new function gen_progcomp_completions to take care of searching for and evaluating a compspec for a particular word, saving its status, and returning to its caller (programmable_completions) whether or not to retry completion. This function also checks whether a retry changed the compspec associated with a command and short- circuits the retry if it has not - changed programmable_completions to try default completion (if set) if a specific completion was not found for a command - changed programmable_completions to implement "fail/retry" semantics for a shell function that returns 124 and changes the compspec associated with the command. All based on proposal and changes from Behdad Esfahbod (Red Hat bugzilla 475229) doc/bash.1,lib/readline/doc/rluser.texi - documented new dynamic programmable completion functionality 8/5 --- stringlib.c - first argument to substring() is now `const char *' externs.h - changed extern declaration for substring() subst.c - skipsubscript now takes a third FLAGS argument, passes to skip_matched_pair - skip_matched_pair now interprets flags&1 to mean not to parse matched pairs of quotes, backquotes, or shell word expansion constructs {subst,general,expr}.c - changed skipsubscript() callers assoc.c - changed assoc_to_assign to double-quote the key if it contains any shell metacharacters arrayfunc.c - use skipsubscript in quote_assign rather than quote any glob characters in the subscript of an array assignment - in assign_compound_array_list, call skipsubscript with a flags argument of 1 if assigning an associative array to avoid trying to re-parse quoted strings redir.c - set expanding_redir before expanding body of here documents and here strings to avoid looking for variables in temporary env 8/7 --- lib/readline/readline.c - in _rl_dispatch_callback, return value of -3 means that we have added to a key sequence, but there are previous matches in the sequence. Don't call _rl_subseq_result if we get a -3 from a previous context in the chain; just go back up the chain. Report and fix from bashline.c - fixes to history_completion_generator and bash_dabbrev_expand to make dabbrev-expand inhibit suppressing of appending space char to matches. Have to do it with the generator too because rl_menu_complete turns off suppressing the appended space in set_completion_defaults(). Suggestion from Dan Nicolaescu - suppress completion match sorting in bash_dabbrev_expand by setting rl_sort_completion_matches = 0. Suggestion from Dan Nicolaescu - don't qsort history match list in build_history_completion_array if dabbrev_expand_active == 1 - start the loop in build_history_completion_array that gathers words from history for possible completions from the end of the list rather than the beginning. It doesn't matter where you start if the results are sorted, and dabbrev-expand is supposed to offer the most recent completions first 8/12 ---- execute_cmd.c - change to execute_command_internal to make [[ ... ]] conditional command subject to settings of `set -e' and the ERR trap 8/14 ---- execute_cmd.c - change to execute_command_internal to make (( ... )) arithmetic command subject to settings of `set -e' and the ERR trap lib/readline/text.c - new bindable function, rl_skip_csi_sequence, reads the characters that make up a control sequence as defined by ECMA-48. Sequences are introduced by the Control Sequence Indicator (CSI) and contain a defined set of characters. Insert, End, Page Up and so on are CSI sequences. Report and code from Andy Koppe lib/readline/readline.h - extern declaration for rl_skip_csi_sequence lib/readline/funmap.c - new bindable command "skip-csi-sequence", runs rl_skip_csi_sequence doc/bash.1,lib/readline/doc/{readline.3,rluser.texi} - documented new bindable command "skip-csi-sequence", unbound by default builtins/evalfile.c - fix _evalfile to remove embedded null bytes from the file read into the string. Report and proposed fix from Roman Rakus {configure,config.h}.in - check for syslog(3), define HAVE_SYSLOG - check for syslog.h, define HAVE_SYSLOG_H config-top.h - new define SYSLOG_HISTORY, disabled by default config-bot.h - if HAVE_SYSLOG or HAVE_SYSLOG_H are not defined, undef SYSLOG_HISTORY bashhist.c - if SYSLOG_HISTORY is defined, call bash_syslog_history with the line added to the history in bash_add_history. - new function, bash_syslog_history(line), sends line to syslog at user.info. The line is truncated to send no more than 600 (SYSLOG_MAXLEN) bytes to syslog. Feature requested by many, and required by some national laws sig.c - in termsig_handler, resend SIGHUP to children if subshell_environment indicates we're a shell performing command or process substitution jobs.c - add CHECK_TERMSIG calls to wait_for in addition to the ones in waitchld() builtins/shopt.def - new functions set_bashopts, parse_bashopts, and initialize_bashopts to manage new environment variable $BASHOPTS, like $SHELLOPTS but for shopt options - change toggle_shopts to call set_bashopts after setting options, so $BASHOPTS reflects new values shell.c - call initialize_bashopts after calling initialize_shell_options at shell startup configure.in - new configure `enable' option --enable-exended-glob-default, to set the initial default value of the `extglob' shell option config.h - new define, EXTGLOB_DEFAULT, controlled by the `extended-glob-default' configure option pathexp.c - initialize extended_glob variable to EXTGLOB_DEFAULT doc/{bash.1,bashref.texi} - document new $BASHOPTS variable and its behavior doc/bashref.texi - document new --enable-extended-glob-default configure option 8/16 ---- print_cmd.c - new variables: xtrace_fd and xtrace_fp, the file descriptor and FILE * to which we send `set -x' tracing output. If fd == -1 then fp == STDERR, the default mode - new function xtrace_init, sets xtrace_fd == -1 and xtrace_fp = stderr - new function xtrace_set (fd, fp), sets xtrace_fd and xtrace_fp to the arguments - new function xtrace_reset, handles closing old xtrace fd/fp and moving them back to -1/stderr - new function xtrace_fdchck, calls xtrace_reset if the fd passed as an argument is xtrace_fd - change xtrace functions to fprintf to xtrace_fp instead of stderr shell.c - call xtrace_init() very early in main() variables.c - new special variable, BASH_XTRACEFD, holds file descriptor used for set -x trace output. Inspired by suggestion from Bruce Korb doc/{bash.1,bashref.texi} - added description of new BASH_XTRACEFD variable redir.c - add calls to xtrace_fdchk to the redirections that close file descriptors, so we notice if we close BASH_XTRACEFD and compensate accordingly (same places that call coproc_fdchk()) 8/18 ---- lib/readline/text.c - change to _rl_replace_text to add error checks: start must be <= end, and we don't call rl_insert_text if passed the empty string config.h.in - add define for HAVE_ICONV, already found by intl autoconf macros - add define for HAVE_LOCALE_CHARSET aclocal.m4 - add check for locale_charset() to BASH_CHECK_MULTIBYTE lib/sh/fnxform.c - new file with two public function: fnx_tofs and fnx_fromfs. Primarily intended for use on MacOS X, they use iconv to convert between whatever the current locale encoding is and "UTF-8-MAC", a special encoding on OS X in which all characters are decomposed unicode, as the HFS+ filesystem stores them. These functions return a pointer to a local buffer, allocated once and resized as necessary, to avoid too many allocations; callers should not free the return value, since it may be the string passed Makefile.in - make sure LIBICONV is set by autoconf (@LIBICONV@) and added to list of link libraries externs.h - new extern declarations for fnx_fromfs and fnx_tofs lib/glob/glob.c - convert the filename read using readdir() in glob_vector() using fnx_fromfs and use that value in the call to strmatch. This ensures that we're using the precomposed Unicode value of the filename rather than the native decomposed form. Original bug report from Len Lattanzi ; fix inspired by Guillaume Outters 8/19 ---- lib/readline/complete.c - new completion hook: rl_filename_rewrite_hook, can rewrite or modify filenames read from the filesystem before they are compared to the word to be completed lib/readline/readline.h - extern declaration for rl_filename_rewrite_hook lib/readline/doc/rltech.texi - document rl_filename_rewrite_hook bashline.c - new function, bash_filename_rewrite_hook, assigned to rl_filename_rewrite_hook. Calls fnx_fromfs to convert from filesystem format to "input" format. This makes completing filenames with accented characters work on Mac OS X 8/20 ---- lib/readline/bind.c - new bindable variable "skip-completed-text", bound to _rl_skip_completed_text. If enabled, it means to note when completing before the end of a word and skipping over characters after rl_point that match in both the completion to be inserted and the word being completed. It means that completing `Makefile' with the cursor after the `e' results in `Makefile' instead of `Makefilefile'. Inspired by an idea from Jared Yanovich from back in 2004 lib/readline/rlprivate.h - extern declaration for _rl_skip_completed_text lib/readline/complete.c - implement semantics of _rl_skip_completed_text in insert_match: skip characters in `replacement' that match chars in rl_line_buffer from the start of the word to be completed 8/21 ---- error.c - change parser_error to set last_command_exit_value to 2 before calling exit_shell (if set -e is enabled), so any exit or ERR trap gets the right value of $?. Suggestion from Stefano Lattarini braces.c - fix expand_seqterm so that a non-zero-prefixed term that's longer than a zero-prefixed term determines the length of each term in the brace-expanded sequence. This means that things like {01..100} will have three digits in all the elements of the expanded list. Fixes bug reported by Jeff Haemer 8/24 ---- {arrayfunc,variables}.c - when inserting a value into an associative array using syntax like T=v where T is an already-declared associative array using key "0", make sure the key is in newly-allocated memory so it can be freed when the variable is unset. Fixes bug reported as redhat 518644 by Jon Fairbairn 8/26 ---- lib/readline/funmap.c - add "old-menu-complete" binding for rl_old_menu_complete lib/readline/readline.h - add extern declaration for rl_old_menu_complete subst.c - fix memory leak when processing ${!prefix@}. Need to dispose all words in the word list created from all matching variable. Fixes bug reported by muszi@muszi.kite.hu. 8/29 ---- execute_cmd.c - add fflush(stdout) and fflush(stderr) to child coproc code before calling exit after execute_in_subshell 8/31 ---- lib/readline/{{bind,readline}.c,rlprivate.h} - new bindable variable, "echo-control-characters", enabled by default. This controls whether or not readline honors the tty ECHOCTL bit and displays characters corresponding to keyboard-generated signals. Controlled by _rl_echo_control_chars variable, declared in readline.c lib/readline/signals.c - if _rl_echo_control_chars == 0, don't go through _rl_echo_signal_char lib/readline/doc/{readline.3,rluser.texi} - document "echo-control-characters" bindable variable 9/1 --- lib/readline/histexpand.c - hist_string_extract_single_quoted now takes an additional argument: a flags word. The only defined value (flags & 1) allows backslash to quote the single quote. This is to inhibit history expansion inside $'...' containing an escaped single quote. - change history_expand to call hist_string_extract_single_quoted with flags == 1 if it sees $'. Fixes bug reported by Sean Donner 9/2 --- builtins/printf.def - add a call to sh_wrerror if ferror() succeeds in the PRETURN macro, to print an error message in the case that the final fflush fails (for instance, because it attempts to write data that didn't have a trailing newline). Fixes bug reported by Stefano Lattarini 9/7 --- arrayfunc.c - some fixes to assign_compound_array_list to avoid null pointer dereferences pointed out by clang/scan-build lib/glob/glob.c - fixes to udequote_pathname and wdequote_pathname to avoid possible null pointer dereferences pointed out by clang/scan-build lib/readline/undo.c - fix to _rl_copy_undo_list (function unused) to avoid deref of uninitialized pointer pointed out by clang/scan-build general.c - fix string_to_rlimtype so it works if passed a null pointer (though it never is) builtins/mapfile.def - fix to mapfile() to avoid possible null pointer dereference pointed out by clang/scan-build variables.c - fix to valid_exportstr to avoid possible null pointer dereferences pointed out by clang/scan-build bashline.c - fix to bash_execute_unix_command to avoid possible null pointer dereference if READLINE_LINE or READLINE_POINT is not bound 9/11 ---- [Prayers for the victimes of 9/11/2001] command.h - add `rflags' member to struct redirect to hold private flags and state information - change redirector to a REDIRECTEE instead of int to prepare for possible future changes {copy_cmd,dispose_cmd,make_cmd,print_cmd,redir}.c - changes resulting from type change of `redirector' member of struct redirect: change x->redirector to x->redirector.dest and add code where appropriate to deal with x->redirector.filename make_cmd.h - change extern declaration for make_redirection make_cmd.c - first argument of make_redirection is now a `REDIRECTEE' to prepare for possible future changes. First arg is now assigned directly to redirector member instead of assigning int to redirector.dest {make_cmd,redir}.c,parse.y - changes resulting from type change of first argument to make_redirection from int to REDIRECTEE. In general, changes are using REDIRECTEE sd and assigning old argument to sd.dest, then passing sd to make_redirection make_cmd.[ch],parse.y - add fourth argument to make_redirection: flags. Sets initial value of `rflags' member of struct redirect - changed all callers of make_redirection to add fourth argument of 0 9/15 ---- parse.y - change read_token_word to return REDIR_WORD for tokens of the form {var} where `var' is a valid shell identifier and the character following the } is a `<' or `>' - add REDIR_WORD versions of all input and output file redirections and here documents print_cmd.c - change input and output file redirection direction and here document cases of print_redirection to print a varname specification of the form {var} when appropriate. Still need to fix rest of cases redir.c - implement REDIR_VARASSIGN semantics for file input and output redirections and here documents 9/16 ---- parse.y - added REDIR_WORD versions of remaining redirection constructs except for err_and_out ones redir.c - handle REDIR_VARASSIGN semantics for rest of redirection constructs - accommodate REDIR_VARASSIGN when translating redirections - new function, redir_varvalue, does variable lookup for {v} when redirection needs the value (e.g., r_close_this) print_cmd.c - fix rest of cases to print {varname} when REDIR_VARASSIGN is set in redirect->rflags doc/{bash.1,bashref.texi} - document new {varname} REDIR_VARASSIGN form of redirections tests/vredir.{right,tests},vredir[1-5].sub - tests for new {varname} REDIR_VARASSIGN form of redirections 9/18 ---- subst.c - new flags argument to split_at_delims: these flags are ORd with SD_NOJMP and passed to skip_to_delim - change skip_to_delim to honor new SD_NOQUOTEDELIM flag by not checking whether or not single and double quotes are delimiters if it's set in passed flags until after skipping quoted strings. subst.h - change extern declaration for split_at_delims - new define for SD_NOQUOTEDELIM flag pcomplete.c - pass SD_NOQUOTEDELIM in flags argument to split_at_delims so single and double quotes, even though they're in rl_completer_word_break_characters, don't act as word delimiters for programmable completion. Fixes bug reported by Freddy Vulto lib/glob/glob.c - in glob_filename, after recursively scanning a directory specified with `**', turn off GX_ALLDIRS|GX_ADDCURDIR before calling glob_vector on the rest of the pathname, since it may not apply to the rest of the pattern. Turned back on if the filename makes it appropriate. Fixes bug reported by Anders Kaseorg redir.c - change execute_null_command to fork a child to execute if any of the commands redirections have the REDIR_VARASSIGN flag set, since those commands are not supposed to have side effects test.c - < and > binary operators will obey the locale by using strcoll if the TEST_LOCALE flag is passed to binary_test test.h - new define for TEST_LOCALE execute_cmd.c - execute_cond_node sets TEST_LOCALE so [[ str1 < str2 ]] (and >) obey the locale. Fixes bug/incompatibility reported by Greg Wooledge doc/{bash.1,bashref.texi} - documented [[ command new locale-sensitive treatment of < and > 9/24 ---- configure.in - add "darwin10" cases like darwin8 and darwin9 to handle linking with included readline and history libraries 9/26 ---- lib/readline/display.c - modify change of 7/24 to use prompt_physical_chars instead of prompt_visible_length to account for visible multibyte characters in the line (usually in the prompt). Fixes debian bug #547264 reported by Pietro Battiston - add flags argument to _rl_col_width; changed callers. flags > 0 means that it's ok to use the already-computed prompt information; flags == 0 means that we're expanding the prompt and we should not short-circuit parse.y - in decode_prompt_string, when expanding \w and \W on Mac OS X, use fnx_fromfs to convert from "filesystem" form to "input" form. This makes $PWD with multibyte characters work in the prompt string on Mac OS X lib/sh/fnxform.c - in fnx_fromfs and fnx_tofs, use templen instead of outlen as last argument in calls to iconv, since outlen is used to keep track of the size of the buffer, and iconv potentially modifies its `outbytesleft' argument 9/29 ---- subst.c - make skip_to_delim understand how to skip over process substitution constructs the way it skips $(...) command substitution 9/30 ---- lib/readline/terminal.c - don't set the `terminal has meta key' flag if the `MT' capability is available; that means something completely different 10/1 ---- builtins/help.def - make sure width is at least 7, since we pass `width/2 - 3' to strncpy as the length argument. Terminal widths <= 6 are converted to 80. Fixes bug reported by Chris Hall configure.in - changed version to 4.1-alpha subst.h - new flag for skip_to_delim: SD_NOSKIPCMD, which means to not skip over embedded command and process substitutions, but rather to look for delimiters within them subst.c - implement semantics of SD_NOSKIPCMD in skip_to_delim bashline.c - call skip_to_delim with SD_NOSKIPCMD from find_cmd_start, so programmable completion can use the completion defined for `b' for command lines like "a $(b c". Fixes inconsistency/bug reported by Freddy Vulto parser.h - replace unused PST_CMDTOKEN parser state value with PST_EXTPAT, means currently parsing an extended glob pattern (extglob) parse.y - fix cond_node() so that extended_glob is set before parsing the rhs of the `==' or `!=' operators. For ksh93 compatibility. - reset extended_glob to global value (saved in parse_cond_command()) in reset_parser() 10/5 ---- jobs.c - change waitchld() to only interrupt the wait builtin when the shell receives SIGCHLD in Posix mode. It's a posix requirement, but makes easy things hard to do, like run a SIGCHLD trap for every exiting child. Change prompted by question from Alex Efros doc/bashref.texi - document new posix mode behavior about SIGCHLD arriving while the wait builtin is executing when a trap on SIGCHLD has been set 10/6 ---- lib/readline/histexpand.c - fix hist_expand to keep from stopping history expansion after the first multibyte character (a `break' instead of a `continue'). Fixes debian bug (#549933) reported by Nikolaus Schulz 10/8 ---- builtins/read.def - implement new `-N nchars' option: read exactly NCHARS characters, ignoring any delimiter, and don't split the result on $IFS. Feature requested by Richard Stallman doc/{bash.1,bashref.texi} - document new `read -N' option 10/9 ---- lib/readline/bind.c - new bindable variable, "enable-meta-key", controls whether or not readline enables any meta modifier key the terminal claims to support. Suggested by Werner Fink lib/readline/doc/{readline.3,rluser.texi},doc/bash.1 - document new readline "enable-meta-key" bindable variable 10/10 ----- trap.c - new function, free_trap_string(), does what it says and turns off SIG_TRAPPED flag without changing signal disposition [bash-4.1-alpha frozen] 10/16 ----- builtins/mapfile.def - return an error if the variable passed is not an indexed array. Fixes bug reported by Nick Hobson - change help text to make it clear that an indexed array is required doc/{bash.1,bashref.texi} - changed description of mapfile to note that the array variable argument must be an indexed array, and mapfile will return an error if it is not subst.c - change expand_string_unsplit and expand_string_leave_quoted to add the (previously unused) W_NOSPLIT2 flag to the created word - change expand_word_internal to understand W_NOSPLIT2 to mean that we're not going to split on $IFS, so we should not quote any characters in IFS that we add to the result string. Fixes bug reported by Enrique Perez-Terron - change cond_expand_word similarly. Fixes rest of bug reported by Enrique Perez-Terron parse.y - save and restore value of last_command_subst_pid around call to expand_prompt_string in decode_prompt_string. Fixes bug that causes $? to be set wrong when using a construct like false || A=3 when set -x is enabled and $PS4 contains a command substitution. Reported by Jeff Haemer 10/17 ----- execute_cmd.c - in execute_in_subshell, make sure we set setjmp(return_catch) before running the command, in case the command or its word expansion calls jump_to_top_level. Fixes bug reported by Nils Bernhard subst.c - new PF_NOSPLIT2 flag for param_expand - parameter_brace_expand takes a new `pflags' argument, before the `output' parameters; passes to param_expand as necessary - change parameter_brace_expand to call parameter_brace_expand_word with the PF_NOSPLIT2 flag if the pflags argument to parameter_brace_expand has it set parse.y - change report_syntax_error to set last_command_exit_value to EX_BADSYNTAX if parse_and_execute_level is > 0, indicating a syntax error while we're executing a dot script, eval string, trap command, etc. builtins/evalstring.c - in parse_and_execute, if parse_command() returns non-zero, indicating a parse error, print a warning message if the conditions would require a posix-mode shell to abort (parse error in a `.' script or eval string) 10/19 ----- builtins/evalfile.c - even if the `check binary' flag is not passed to _evalfile, return an error after reading 128 null characters if called by `source', on the assumption that it's probably a binary file. [This will be in bash-4.1-beta] 10/24 ----- [bash-4.1-alpha released] bashline.c - don't call command_substitution_completion_function if we're completing a substring delimited by a single quote. Fixes bug reported by bash-bugs@atu.cjb.net lib/readline/complete.c - make sure _rl_skip_completed_text defaults to 0, as the documentation states (incorrect in bash-4.1-alpha) - in insert_match, skip over a close quote in the replacement text if the character at point when completion is invoked is a single quote. Fixes complaint from bash-bugs@atu.cjb.net 10/26 ----- shell.c - in main, make sure "$EMACS" is non-null before calling strstr on its value. Fixes Red Hat bug 530911 submitted by Mitchell Berger builtins/mapfile.def - don't save callback commands in shell history. Suggested by Jan Schampera mailcheck.c - in file_mod_date_changed, make sure the modification time is later than the saved modification date, not just that it's not equal. Fix from Evgeniy Dushistov - in file_access_date_changed, make sure the access time is later than the saved access time, not just that it's not equal 10/27 ----- builtins/shopt.def - added new `compat40' compatibility variable, with associated changes to shell_compatibility_level(), since the default compatibility level is now 41 test.c - make the < and > operators to [[ use strcoll() only if the shell compatibility level is greater than 40 (it is 41 by default in bash-4.1) 10/28 ----- support/shobj-conf - decrease the default version of FreeBSD that creates shared libraries to 4.x. Advice from Peter Jeremy 11/2 ---- parse.y - change parse_comsub to free `heredelim' and set it to 0 whenever the comsub scanner finds the end of a here document. Really need to implement a stack of here doc delimiters like in the parser (can we use redir_stack here, too?) - fix parse_comsub to not attempt to read another here doc delimiter after seeing a shell break character (that is not newline) if we already have one. Fixes Debian bash bug #553485, submitted by Samuel Hym 11/3 ---- variables.c - fix bind_variable_internal to call a variable's dynamic 'set function' with the right arguments depending on whether its an associative array, an indexed array, or a scalar. Fixes Ubuntu bug #471504 https://bugs.launchpad.net/ubuntu/+source/bash/+bug/471504 reported by AJ Slater [bash-4.1-beta frozen] 11/11 ----- builtins/printf.def - in getintmax(), in the case of a conversion error, return the partial value accumulated so far, which is suppose to be what strtoimax/strtoll/strtol returns 11/17 ----- [bash-4.1-beta released] 11/18 ----- builtins/{common.h,shopt.def},shell.c - changed shopt variable "set functions" to take the option name as the first argument; changed function prototypes and callers builtins/shopt.def - change set_compatibility_level() to turn off other compatNN options when one is set -- enforce mutual exclusivity. Fixes problem noted by Jan Schampera 11/19 ----- lib/readline/rltty.c - make sure prepare_terminal_settings() tests for the presence of ECHOCTL before using it. Fixes bug reported by Joachim Schmitz config-top.h - new WORDEXP_OPTION define (off by default) shell.c - don't include the --wordexp option or the supporting function (run_wordexp) if WORDEXP_OPTION is not defined. Suggested by Aharon Robbins execute_cmd.c - in execute_cond_node, turn on comsub_ignore_return if the flags indicate we're ignoring the return value before calling cond_expand_word. Fixes bug reported by Anirban Sinha 11/20 ----- lib/sh/snprintf.c,builtins/printf.def - change check for HAVE_ASPRINTF and HAVE_SNPRINTF to check if value is 1 or 0 rather than whether they are defined or not. This allows a value of 0 to enable function replacement configure.in,aclocal.m4 - new autoconf macro, BASH_FUNC_SNPRINTF, checks for snprintf present and working as C99 specifies with a zero length argument. Idea from Greg Wooledge - new macro BASH_FUNC_VSNPRINTF, does same thing for vsnprintf 11/25 ----- subst.c - in command_substitute, only tell parse_and_execute to reset the line number in an interactive shell if sourcelevel == 0 -- we'll use the line numbers from the sourced file execute_cmd.c - in execute_simple_command, only subtract function_line_number from line_number if sourcelevel == 0. If sourcing, we'll use the line numbers from the sourced file. Fixes bug reported by Hugo Mildenberger builtins/declare.def - in declare_internal, call bind_assoc_variable instead of bind_array_variable in the case of declare -A foo=bar. Fixes bug reported by Bernd Eggink . 11/27 ----- lib/readline/util.c - change declaration for _rl_walphabetic to use prototype, assuming that any system with multibyte characters has a compiler that can handle prototypes. Fix for AIX compilation problem reported by Nick Hillman 11/28 ----- execute_cmd.c - make funcnest file-scope static and unwind-protect its value in execute_function, so it can be used as a real measure of function call nesting general.c - fix off-by-one error in trim_pathname that caused it to short-circuit when PROMPT_DIRTRIM == number of directories - 1. Fixes bug reported by Dennis Williamson 11/29 ----- jobs.c - when fork() returns -1/EAGAIN, call waitchld(-1, 0) so the shell can reap any dead jobs before trying fork again. Currently disabled until bash-4.2 development starts lib/readline/complete.c - when incrementing _rl_interrupt_immediately, make sure it's greater than 0 before decrementing it. In practice, not a problem, but the right way to do it. Suggested by Jan Kratochvil lib/readline/signals.c - make sure rl_signal_handler doesn't set rl_caught_signal if _rl_interrupt_immediately is set, so RL_CHECK_SIGNALS doesn't cause it to be processed twice. Suggested by Jan Kratochvil - if the callback interface is being used, use the code path that immediately handles signals. This restores the readline-5.2 behavior. Fixes GDB readline bug reported by Jan Kratochvil 12/18 ----- [bash-4.1-rc1 released] 12/22 ----- config-top.h - don't have SYSLOG_HISTORY enabled by default lib/sh/Makefile.in - add explicit dependency on pathnames.h for parallel make support externs.h - add extern declaration for xtrace_fdchk lib/sh/snprintf.c - add local prototype declarations for isinf, isnan if we are providing local definitions lib/sh/fnxform.c - add extern declaration for get_locale_var if HAVE_LOCALE_CHARSET not defined execute_cmd.c - define NEED_FPURGE_DECL so we pick up any extern declaration for fpurge (e.g., if the system doesn't provide it) builtins/shopt.def - correct prototype and declaration for set_shellopts_after_change so it's the correct type for shopt_set_func_t - add new function shopt_enable_hostname_completion that is the correct type for shopt_set_func_t; just calls enable_hostname_completion and returns its result 12/26 ----- doc/{bash.1,bashref.texi} - add \E and \" escape sequences to ANSI-C quoting description. Suggested by Aharon Robbins 12/29 ----- doc/bash.1 - make sure shell and environment variable names are always in `small caps' bold. Suggested by Aharon Robbins 12/30 ----- {execute_cmd.c,parse.y,Makefile} - changes for building minimal configuration from Matthias Klose [bash-4.1 frozen] 12/31 ----- [bash-4.1 released] 1/5/2010 -------- doc/bashref.texi - document compat32 and compat40 shopt options. Omission pointed out by Dilyan Palauzov 1/6 --- lib/readline/complete.c - use `convfn' (converted filename) instead of entry->d_name (filename read from file system) when adding partial or full completions to the command line. Bug and fix from Guillaume Outters 1/7 --- builtins/printf.def - fix prototype in extern declaration for vsnprintf. Fix for bug reported by Yann Rouillard 1/9 --- parse.y - fix shell_getc to handle alias expansions containing quoted newlines. Problems in bash-4.1 with aliases containing quoted newlines in the middle of and at the end of their expansion. Fix for bug reported by Jonathan Claggett - change mk_alexpansion to not append a space to an alias expansion ending with a newline. Works with shell_getc 1/11 ---- lib/glob/Makefile.in - add dependencies on shell.h and pathnames.h. From Mike Frysinger 1/15 ---- doc/{bash.1,{bashref,version}.texi},lib/readine/doc/rluser.texi - some typo fixes from Aharon Robbins - added descriptions of ENV, COPROC, and MAPFILE variables - added descriptions of READLINE_LINE and READLINE_POINT 1/21 ---- arrayfunc.c - free `akey', the word-expanded index into the assoc array to avoid mem leak in array_value_internal - free index into assoc array in unbind_array_element - change array_value_internal to take an additional argument: an arrayind_t *. If not null, an index to an indexed array is returned there. If not an indexed array or subscript is @ or *, the value is unchanged 1/22 ---- builtins/ulimit.def - include if we found it during configure and we don't have resources. Fixes omission reported by Joachim Schmitz {configure,config.h}.in - check for , define HAVE_ULIMIT_H if found lib/sh/oslib.c - include for extern declaration for kill(2) if HAVE_KILLPG not defined jobs.c - if HAVE_KILLPG is not defined, add an extern prototype decl for killpg() 1/24 ---- print_cmd.c - when printing here-string redirections, don't quote the string. The original quotes, if any, are still in place and don't need to be requoted. Fixes bug reported by Arfrever Frehtes Taifersar Arahesis subst.c - fix array_length_reference to return 0 for variables that have not been assigned a value. Fixes bug reported by Mart Frauenlab , but is not backwards compatible arrayfunc.[ch] - change array_value to take a new arrayind_t *indp parameter like get_array_value; changed extern prototype declaration subst.c - changed callers of array_value to add extra parameter expr.c - change expr_streval to set a new `lvalue' parameter with information about the string being evaluated: string, value, array index (if any), variable evaluated (if set). - saving and restoring current context now saves and restores the current `lvalue' - new function expr_bind_array_element, binds an array element with an already-computed index to a specified value - anywhere we set the current token to a string (STR), save and set the current lvalue - change calls to expr_bind_variable to check whether or not the current lvalue indicates an indexed array was evaluated, and, if so, call expr_bind_array_element using the already-computed index (curlval.ind). Fixes problems with dynamic variables (e.g., RANDOM) in array indices with {pre,post}-{inc,dec}rement and op= operators reported by 1/25 ---- expr.c - fix subexpr() to initialize curlval and lastlval when resetting all of the rest of the expression-parsing variables 1/26 ---- builtins/setattr.def - in show_var_attributes, if the variable is not set (value == 0), don't print `name=""', just print `name'. Pointed out by Mart Frauenlab arrayfunc.c - fix array_keys to return NULL if the variable is not set or invisible. Pointed out by Mart Frauenlab - change array_value_internal to return NULL for variable which has not been set 1/30 ---- bashline.c - in command_word_completion_function, don't call glob_pattern_p on hint -- use the already-computed `globpat'. At this point, hint might contain an already-dequoted globbing character, but glob_matches will be NULL. Fixes bug reported by coyote@wariat.org.pl 2/5 --- builtins/exec.def - set extern variable "exec_argv0" to the argument to -a shell.c - if exec_argv0 is set, set dollar_vars[0] to it and set it to NULL, assuming it was set by `exec -a'. `exec -a foo' now sets $0 to foo in an executable shell script without a leading `#!' (fixes longstanding bug) 2/8 --- variables.c - in push_func_var, if a variable is in a special builtin's temporary environment and needs to be propagated because we're in Posix mode, or we just need to propagate a variable, and we are executing in a function without any local variables (so the function-local variable context has no variable hash table), make sure we create a hash table so we have a place to save the variable to be propagated. Fixes bug reported by Crestez Dan Leonard . 2/18 ---- builtins/hash.def - change add_hashed_command to remove the command being looked up from the hash table before trying to add it. That way, if it's not found, there won't be anything remaining in the hash table 2/26 ---- trap.[ch] - move IMPOSSIBLE_TRAP_HANDLER define to trap.h so other parts of the shell can use it parse.y - change yy_readline_get to use IMPOSSIBLE_TRAP_HANDER instead of NULL as a sentinel value for the SIGINT signal handler - make sure yy_readline_get resets interrupt_immediately to 0 after calling readline() using the same criteria it used to set it to 1 before the call -- make the code symmetric. Suggested by Werner Fink builtins/read.def - move assignment to `retval' before decrement of interrupt_immediately and terminate_immediately and call to discard_unwind_frame - move assign_vars label before decrement of interrupt_immediately and terminate_immediately so those variables get reset appropriately if read -t times out subst.h - new define for Q_DOLBRACE, indicates double-quoted ${...} subst.c - in parameter_brace_expand, before calling parameter_brace_expand_rhs, add Q_DOLBRACE to `quoted' if we're within double quotes. - in expand_word_internal, if the Q_DOLBRACE flag is set, remove a backslash escaping a }. Result of a Posix discussion on the austin-group list 2/27 ---- variables.c - new functions to save and restore the PIPESTATUS variable's internal array: save_pipestatus_array and restore_pipestatus_array variables.h - new extern declarations for save_pipestatus_array and restore_pipestatus_array trap.c - in run_pending_traps, _run_trap_internal, and run_exit_trap, save and restore $PIPESTATUS while traps are running. Fixes bug reported by Florian Bruhin parse.y - use save_pipestatus_array and restore_pipestatus_array in save_parser_state and restore_parser_state, respectively, replacing inline code lib/readline/histfile.c - fix callers of history_filename to be prepared to cope with it returning NULL - change history_filename to return NULL if $HOME is not set, rather than trying to write the history file in the current directory. This is the default directory, used only if the application does not specify a history filename. Changed due to long-ago (unsent) bug report from OpenBSD {Makefile,config.h,configure}.in,externs.h,lib/sh/{dprintf.c,Makefile.in} - change fdprintf to dprintf, which is the Posix standard interface, look for it with configure, replace it if not available 2/28 ---- command.h - add new subshell flag, SUBSHELL_RESETTRAP. Indicates to the trap builtin that the shell is executing a command substitution and should free the trap strings we left unfreed by reset_signal_handlers() trap.c - free_trap_string() and free_trap_strings() are now compiled in builtins/trap.def - if changing a signal disposition and the SUBSHELL_RESETTRAP flag is set in subshell_environment, free the trap strings left unfreed by reset_signal_handlers subst.c - in command_substitute, set the SUBSHELL_RESETTRAP flag. This change is for Austin Group Posix interpretation 53 (http://austingroupbugs.net/view.php?id=53) 3/7 --- lib/sh/{Makefile.in,strchrnul.c},Makefile.in - implementation of strchrnul, from gnulib configure.in,config.h.in - look for strchrnul and compile in version in lib/sh/strchrnul.c if not available - look for mbsnrtowcs and define HAVE_MBSNRTOWCS if available lib/sh/xmbsrtowcs.c - new function, xdupmbstowcs2, fast version of xdupmbstowcs used when mbsnrtowcs is available and the indices are not required. Called from xdupmbstowcs as required. Initial patch from <0xe2.0x9a.0x9b@gmail.com> 3/22 ---- print_cmd.c - call print_deferred_heredocs virtually every time a recursive call to make_command_string_internal is made so here documents get printed correctly when they are attached to commands inside compound commands such as for and while. Fixes bug reported by Mike Frysinger 3/25 ---- builtins/printf.def - fix have_precision case in PF macro to call printf with precision instead of fieldwidth argument. Fixes bug reported by Rob Robason 3/26 ---- trap.[ch] - new function, signal_is_hard_ignored, returns true if the shell inherited SIG_IGN as a signal's disposition - new function, set_original_signal (sig, handler), provides interface for rest of shell to set original_signals[sig] = handler execute_cmd.c - execute_disk_command needs to call reset_terminating_signals in the child process before resetting traps with restore_original_signals builtins/trap.def - call initialize_terminating_signals before calling display_traps for trap -p or trap without any other arguments. Possible future use lib/readline/complete.c - rl_filename_completion_function needs to call rl_filename_dequoting_function with `dirname' (which has already been tilde-expanded) instead of `users_dirname', because it calls opendir with `dirname'. Fixes bug reported by Stefan H. Holek 3/27 ---- sig.c - experimental change to set_signal_handler: when setting the SIGCHLD handler, set the SA_RESTART flag so that interruptible system calls get restarted after a child dies. Fixes bug reported by Tomas Trnka , but needs further evaluation lib/sh/eaccess.c - eaccess(2) apparently does only half the job: it does not check that the permission bits on a file actually allow, for instance, execution. Need to augment with a call to sh_stataccess if eaccess returns success on FreeBSD. Fixes FreeBSD problem reported by Jonan Hattne 3/28 ---- parse.y,bashline.c,externs.h - history_delimiting_chars now takes a const char * as an argument: the line being added to the history. Changed callers parse.y - bash_add_history should not add a semicolon separator if the current history entry already ends in a newline. It can introduce syntax errors (e.g., when it results in a null command before a close brace). Fixes bug reported by Andreas Schwab parse.y - history_delimiting_chars needs to return a newline instead of a semicolon if it thinks the current line starts a here document (if it contains `<<'). Also keeps track of the fact with a new static variable, LAST_WAS_HEREDOC, so it can return the right sequence of newlines later for the here-document body. Fixes bug reported by Andreas Schwab 3/29 ---- lib/sh/eaccess.c - if the system has faccessat, sh_eaccess will now use it in preference to all other options 3/30 ---- subst.h - new string_extract and extract_dollar_brace_string flag value: SX_POSIXEXP, set if the shell is expanding one of the new Posix pattern removal word expansions parser.h - new definitions for "word expansion state", shared between parse.y and subst.c subst.c - include parser.h 4/9 --- builtins/declare.def - make sure declare_internal calls bind_assoc_variable with newly- allocated memory for the key argument when using an implicit key of "0". Bug report and fix from Andreas Schwab 4/14 ---- lib/readline/input.c - restructure the rl_event_hook loop in rl_read_key to call the event hook after rl_gather_tyi() returns and rl_get_char has a chance to collect the input. Previous behavior was to call the event hook before attempting to read input. Problem reported by Anant Shankar 4/15 ---- builtins/fc.def - fc_builtin needs to check whether the calculation of last_hist leaves hlist[last_hist] == 0, and keep decrementing it until it leaves a non-null history entry or goes < 0. Currently only does this if saved_command_line_count > 0, indicating we're trying to edit a multi-line command. Fixes bug reported by Roman Rakus 4/17 ---- subst.c - new process substitution helper functions: unlink_fifo - closes a single FD or FIFO num_fifos - returns number of open FDs or active FIFOs copy_fifo_list - returns a bitmap of open FDs or active FIFOs by index into appropriate list (dev_fd_list or fifo_list) close_new_fifos - take a bitmap saved by copy_fifo_list and call unlink_fifo on any FD or FIFO open at the time of the call that is not marked as active in list execute_cmd.c - execute_builtin_or_function: use new framework to close process substitution FDs or FIFOs created by a shell builtin or shell function. Fixes bug reported by Charles Duffy doc/{bash.1,bashref.texi} - document 'C and "C constants for printf builtin 4/22 ---- lib/readline/complete.c - new function to return screenwidth for use when displaying possible matches: complete_get_screenwidth; changed uses of _rl_screenwidth to use complete_get_screenwidth(). - change complete_get_screenwidth to query (readline-private) _rl_completion_colums, $COLUMNS, then _rl_screenwidth in that order - change rl_display_match_list to deal with limit < 0 (which implies that cols == 0) when _rl_screenwidth > 0 lib/readline/bind.c - new bindable variable: completion-display-width, controls the number of columns used when displaying completions with new sv_compwidth function to call when value is set or unset lib/readline/doc/{readline.3,rltech.texi} - documented completion-display-width variable 4/23 ---- execute_cmd.c - change execute_in_subshell to reset trap handlers without freeing the trap strings and set SUBSHELL_RESETTRAP. In line with Austin Group interp #53 (trap in a subshell). - ditto for execute_simple_command where it can be determined that the shell is going to run a builtin or function in a subshell trap.c - new function, get_all_original_signals, retrieves the original signal disposition for all signals trap.h - extern declaration for get_all_original_signals builtins/trap.def - change showtrap to display signals that are "hard ignored" as trap commands to ignore them, even though that trap command would be a no-op. Partial fix for feature request from Siddhesh Poyarekar - change trap_builtin to call get_all_original_signals before displaying traps. This will show inherited ignored signals. Rest of feature request from Siddhesh Poyarekar lib/readline/histexpand.c - fix history_tokenize_word so that it understands $(...) and the <(...) and >(...) expansions as a single word - change history_tokenize_word so that it understands extended shell globbing patterns as a single word. Code is very similar to $(...) code above. Bug reported by Rajeev V. Pillai 4/24 ---- lib/readline/vi_mode.c - add checks to rl_vi_char_search to make sure we've already done a search if the command is `;' or `,', and return immediately if we have not. Fixes bug reported by Eric Ho lib/readline/text.c - make sure `dir' is in the valid range before searching in _rl_char_search_internal. Range checks in the code depend on it being non-zero 5/3 --- lib/readline/complete.c - in rl_complete_internal, if show-all-if-ambiguous or show-all-if-unmodified are set (what_to_do == '!' or '@', respectively), and the common match prefix is shorter than the text being completed, inhibit inserting the match. The guess is that replacing text with a shorter match will not be wanted 5/20 ---- lib/sh/unicode.c - new file, with unicode character conversion-related code. Will be used to implement \u and \U escapes where appropriate, and for other unicode-related functions in the future 5/21 ---- builtins/printf.def - add code to handle \u and \U escapes in format strings and arguments processed by the %b format specifier lib/sh/strtrans.c - add code to handle \u and \U escapes as unicode characters, works for both `echo -e' and $'...' doc/{bash.1,bashref.texi} - document new \u and \U escape sequences for $'...' and echo (printf defers to the system's man page or Posix) 5/24 ---- execute_cmd.c - change execute_disk_command to return a status, instead of just leaving it in `last_command_exit_value', since the parent's return value is sometimes used (e.g., when a restricted shell refuses to run a command with a `/'). Fixes bug reported by David Pitt 5/25 ---- bashline.c - change bash completion functions to save and restore the value of rl_ignore_some_completions_function, and set it to the bash default of filename_completion_ignore where appropriate. Fixes bug reported by Henning Bekel variables.c - new convenience function: find_global_variable (name). Looks for NAME in the global variables table, skipping any local and temporary environment variables builtins/declare.def - add new -g option to declare/typeset/local, forces variables to be created or modified at the global scope when executing inside a shell function. Requested by many, most recently by konsolebox@gmail.com 5/27 ---- test.c - added new `-v var' unary test operator; returns TRUE if var is set (i.e., has been assigned a value). Works in both test builtin and [[ conditional command doc/{bash.1,bashref.texi} - documented new `-v var' unary conditional operator tests/test.tests - added tests for new -v var operator builtins/kill.def - change kill builtin so -PID (pgrp specification) following a -s sig or -n sig option is not interpreted as a signal specification. Fixes bug reported by Roman Rakus builtins/evalstring.c - in parse_and_execute, if parse_command() returns non-zero, indicating a parse error, exit the shell if the conditions require a posix-mode non-interactive shell to abort (parse error in a `.' script or eval string). Bash-4.1 only printed a warning. This is from Austin Group interp 114 doc/bashref.texi - add note to the posix mode section of the texinfo manual noting the changed behavior for `.' and `eval' parse.y - change time_command_acceptable to allow TIME token to appear after BANG token (to allow `! time foo', which is supposed to be valid) - change pipeline_command production to allow multiple instances of `!' (which toggle inverting the return status) and `time' (which have no effect) execute_cmd.c - In posix mode, `time' without a following pipeline prints the elapsed user, system, and real time for the shell and its children since the shell was invoked. It's like `times' but obeys the setting of TIMEFORMAT. A future revision of Posix will require this doc/{bashref.texi,bash.1} - document new posix mode use of `time' parse.y - add production to pipeline_command that permits `!' by itself to be equivalent to `false' (and, with the changes above, permits `! !' to be roughly equivalent to `true'). A future revision of Posix will require this 5/28 ---- parse.y - fix \W prompt expansion to use memmove instead of strcpy, since the source and target strings overlap (though you think it wouldn't matter, since the overlapping regions are never touched at the same time). Fixes bug reported by Stéphane Jourdoi parse.y - Posix interp 217 states that $(( must be parsed first as an arithmetic expansion, so avoid attempting to parse it as a nested command substitution. Fixes bug reported by several, most recently subst.c - change extract_delimited_string to process nested $( as a possible command substitution, but only if already parsing an arithmetic expansion. Rest of fix for Posix interp 217 - change parameter_brace_expand_rhs to make the := expansion operator perform quote removal and both assign the result to the variable and return it as the result of the expansion, rather than assign the value after quote removal but return the value before quote removal. Posix interp 221 - introduce new internal quoting flag: Q_DOLBRACE. Denotes a double- quoted ${...} expansion. In this case, Posix interp 221 requires that a backslash quoting an embedded `}' be removed, even though it's not one of the characters marked as special inside double quotes. Set in parameter_brace_expand, used by expand_word_internal. parse.y - introduce new parsing state, P_DOLBRACE, set when parsing a ${...} expansion - set a "dolbrace operator state" in parse_matched_pair to decide whether the lexer is reading the param, op, or word in ${paramOPword}. Will be used to decide whether or not to treat single quotes specially in a double-quoted "${...} 5/29 ---- parse.y - change parse_matched_pair so that a single quote appearing in a double-quoted ${...} expansion is not special unless the expansion operator is `#[#]' or `%[%]'. Posix interp 221 subst.c - change string_extract_double_quoted so that a single quote appearing in a double-quoted ${...} expansion is not special unless the expansion operator is `#[#]' or `%[%]'. Posix interp 221 doc/bashref.texi - document posix-mode effects of Posix interp 221 - add section describing GNU parallel as requested by Stallman lib/readline/complete.c - broke code that compares filenames read from the file system (and possibly converted) to words being completed out into a separate function: complete_fncmp - augment complete_fncmp to treat hyphen and underscore as equivalent when comparing filenames if _rl_completion_case_map is set lib/readline/rlprivate.h - new extern declaration for _rl_completion_case_map lib/readline/util.c - change _rl_strnicmp to return the difference between the characters, like strcasecmp, and not modify the pointers it is passed - change _rl_stricmp to not modify the pointers it is passed lib/readline/bind.c - new bindable variable, "completion-case-map", toggles value of _rl_completion_case_map lib/readline/doc/{rluser.texi,readline.3} - document new bindable readline variable "completion-case-map" execute_cmd.c - change execute_function to reset funcnest and jump back to top level if funcnest exceeds funcnest_max - use funcnest_max as a max function nesting level, if set to numeric value greater than 0 (defaults to 0, so inactive) variables.c - new variable FUNCNEST, controls funcnest_max value if set to numeric value > 0 sig.c - reset funcnest to 0 when throw_to_top_level occurs doc/{bash.1,bashref.texi} - document FUNCNEST variable and its effect on function execution lib/readline/funmap.c - add new bindable command names to avoid case-insensitive matching problems between, for instance, vi-fword and vi-fWord: vi-forward-word vi-forward-bigword vi-backward-word vi-backward-bigword vi-end-word vi-end-bigword Suggested in a different form in 2006 (!) by Servatius Brandt builtins/mapfile.def - run_callback now takes a new third argument: curline, the line currently being read and about to be assigned - the callback function/command now takes an additional argument: the line to be assigned to the array index. Feature suggested by Dennis Williamson doc/{bash.1,bashref.texi} - document new additional `line' argument to mapfile callback 5/30 ---- builtins/printf.def - add new %(fmt)T format specifier, where FMT is a strftime format. Argument is number of seconds since the epoch, with -1 meaning current time (roughly date +%s) and -2 meaning shell start time (roughly $SECONDS, unless it's been assigned a value or unset). Fieldwidth and precision are preserved, strftime result is printed as with %[-][[fieldwidth][.[precision]]]s doc/{bash.1,bashref.texi} - document new %(datefmt)T printf format specifier and special arguments builtins/hash.def - don't permit programs with slashes to be entered into the hash table at all, even with the -p option. Inconsistency pointed out by Jan Schampera builtins/shopt.def - add `compat41' option in preparation for bash-4.2 6/6 --- lib/readline/vi_mode.c - finish restructuring rl_vi_domove and the functions that call it so it works in callback mode, including numeric arguments. Requested a long time ago by Bob Rossi lib/readline/callback.c - arrange to call appropriate callback when readline state indicates RL_STATE_VIMOTION, so vi motion commands like `cw' and `d2w' are handled in callback mode lib/sh/wcswidth.c - replacement wcswidth implementation aclocal.m4 - add REPLACE_FUNCS(wcswidth) to BASH_CHECK_MULTIBYTE execute_cmd.c - fix select_query and print_index_and_element to compute correct display width of select list elements in presence of multibyte characters. Bug reported by Bernd Eggink builtins/cd.def - add posix-mandated -e option; currently ignored in most circumstances doc/{bash.1,bashref.texi} - document new `cd -e' option 6/12 ---- arrayfunc.c - change array_value_internal to treat negative subscripts to indexed arrays, offset from array_max_index(x) + 1, so foo[-1] is the last element of $foo subst.c - Change verify_substring_values to allow negative length specifications when using string variables or array members. Negative lengths mean to return characters from OFFSET until (${#var} - N) for {var:offset:-N}. Feature requested by Richard Neill doc/{bash.1,bashref.texi} - document new behavior of negative subscripts to indexed arrays - document new behavior of negative LENGTH in substring expansion configure.in - change version to bash-4.2-devel variables.c - make sure initialize_shell_variables calls sv_xtracefd if BASH_XTRACEFD is inherited in the shell environment. Fixes but reported by 6/13 ---- lib/readline/complete.c - change get_y_or_n to always return 1 when in callback mode, so we don't do a blocking read. Have to wait until readline-7.0 to add a state so we can use callbacks, since that will change public interface 6/17 ---- subst.c - fix memory leak in parameter_brace_expand: when performing pattern removal with parameter_brace_remove_pattern, make sure `name' is freed. Fixes bug reported by oyvindh@dhampir.no 6/23 ---- {parse.y,subst.c} - make the ${param//pat/rep}, ${param^pat}, and ${param,pat} expansions require single quotes and double quotes to match when within double quotes. This way every expansion except the Posix ones behaves as bash has always behaved subst.c - change remove_upattern and remove_wpattern to return their first argument if nothing matches, change callers to allocate memory appropriately - change remove_pattern to short-circuit and return copy of PARAM if remove_wpattern returns its first argument (indicating no match) rather than convert back to multibyte string, allocating new memory twice and calling wcsrtombs 6/24 ---- execute_cmd.c - add missing initializers for sh_coproc to eliminate a compiler warning. Patch from Werner Fink 6/27 ---- parse.y - add `TIMEIGN' token to handle `time -p -- ...'. Pointed out by Laszlo Ersek on austin-group list 6/28 ---- jobs.c - treat a shell with (subshell_environment&SUBSHELL_PIPE) != 0 like a command substitution in wait_for and act like we received a SIGINT if a job we're waiting for dies of SIGINT. Fixes bug reported by Ilya Basin 7/2 --- jobs.c - if fork() fails in make_child, try to reap some dead children before retrying execute_cmd.c - change execute_pipeline to run the last command of a non-asynchronous pipeline in the current shell environment if the `lastpipe' shell option is enabled and job control is not active. Code from Werner Fink parse.y - Posix says (issue 267) that time is not recognized as a keyword if the next token begins with a `-' doc/{bash.1,bashref.texi} - changed the descriptions of BASH_SOURCE, BASH_LINENO, and FUNCNAME as proposed in Ubuntu bug 591677. - document new `lastpipe' shell option that runs last command of a pipeline in the current shell environment - document new posix-mode behavior with `time -p' 7/5 --- aclocal.m4 - new autoconf test WEXITSTATUS_OFFSET, bit offset in status word returned by wait() of the process's exit status jobs.[ch] - change stop_pipeline to return the actual index of the job just created and added to the jobs table, instead of the current job - job_exit_status and job_exit_signal are now global functions, with extern declarations in jobs.h - append_process: new utility function for use by the lastpipe code, takes info, creates a PROCESS from them, and adds it to the end of the passed job id's pipeline. lastpipe code uses it to add a dummy process for the last command in the pipeline - freeze_jobs_list: new utility function so rest of shell can freeze the jobs list. Used by the lastpipe code execute_cmd.c - changes to lastpipe code to make `pipefail' option, $PIPESTATUS, and $? work correctly. Uses append_process and job_exit_status 7/10 ---- subst.c - when performing pattern substitution word expansions, a `&' in the replacement string is replaced by the text matched by the pattern. The `&' can be quoted with a backslash to inhibit the expansion. CURRENTLY DISABLED 7/13 ---- pcomplib.[ch] - new member for struct compspec: lcommand. for future use 7/15 ---- parse.y - fix problem in parse_comsub where extra space was added to here-doc delimiter if the first word in the comsub contained a `/'. Fixes bug reported by Alex Khesin 7/20 ---- parse.y - change reserved_word_acceptable to return success if the last two tokens read were `function WORD'. Allows function definitions like function good [[ -x foo ]];. Fixes bug reported by Linda Walsh doc/{bash.1,bashref.texi} - change function definition meta-syntax to make it clearer, rather than let the text note the optional portions 7/24 ---- bashhist.c - change bash_history_inhibit_expansion() to suppress history expansion for $! parameter expansion. Fixes debian bug #589745 submitted by Frank Heckenbach lib/readline/terminal.c - change rl_resize_terminal to always fetch the new terminal size and only force the redisplay if _rl_echoing_p is non-zero. Fixes bug reported by Balazs Kezes 7/25 ---- lib/readline/xfree.c - new file, contains definition of xfree moved from xmalloc.c 7/28 ---- variables.c - check suspect return values from bind_variable before trying to use the returned SHELL_VAR *. Changes to: initialize_shell_variables, bind_int_variable, FIND_OR_MAKE_VARIABLE. Fixes bug reported by Roman Rakus 7/31 ---- lib/readline/rltty.c - fix rl_prep_terminal and rl_deprep_terminal to use fileno(stdout) if rl_instream is NULL. Fixes bug reported by Otto Allmendinger otto.allmendinger@googlemail.com 8/2 --- lib/sh/casemod.c - if the passed string is NULL or empty, return it immediately. Fixes bug reported by Dennis Williamson subst.c - fix pat_subst to cope with the passed string being NULL arrayfunc.h - added flag values for array_value_internal and its callers; converted array_value_internal `allow_all' parameter into a general flags word - get_array_value now takes a flags value - changed array_value internal to use *indp as an index to use if the AV_USEIND flag is set, rather than recomputing it subst.c - get_var_and_type takes two new parameters: a flags word and an index that represents an already-computed index for an array reference (just indexed arrays so far). Index is used and passed to array_value if flags includes AV_USEIND - parameter_brace_expand_word takes a new argument: the already- computed index; returns W_ARRAYIND if word expanded is being used as an array index - changed parameter_brace_casemod, parameter_brace_patsub, parameter_brace_substring, parameter_brace_remove_pattern to take new flags and index arguments from parameter_brace_expand_word. They pass the new parameters along to get_var_and_type to use an already-computed array index if necessary. Fixes bug where array indexes are computed twice reported by Andrew Benton doc/bash.1,lib/readline/doc/{history.3,hsuser.texi} - modified description of history event designators to clarify that all non-absolute event designators are relative to the current position in the history list. Question raised by Frank Heckenbach as debian bash bug 590012 8/5 --- subst.c - remove code that does not add a quoted null when the input string is partially quoted; subsequent word splitting may require it. Fixes bug reported by Eric Blake 8/12 ---- lib/glob/gmisc.c - move match_pattern_wchar and match_pattern_char to new file in glob library - new functions: wmatchlen(pat, max) and umatchlen(pat, max), computes number of characters PAT will match. Returns the number of chars that will be matched or -1 if the match length is indeterminate (i.e., contains a `*') subst.c - use umatchlen/wmatchlen in match_upattern/match_wpattern to bound the number of match attempts in large strings to (usually) one, depending on match length. Fixes performance problems with pattern substitution in large strings noted by Yi Yan . Can be applied to remove_[uw]pattern also 8/13 ---- bashhist.c - in maybe_append_history, change check for history_lines_this_session so that we append the lines to the file if it's equal to the value returned by where_history(). This means that without this change, the history won't be appended if all the lines in the history list were added in the current session since the last time the history file was read or written. Fixes bug reported by Bruce Korb shell.h,parse.y - add prompt_string_pointer to the parser_state struct saved and restored by {save,restore}_parser_state. Fixes both bugs exposed by bash_completion and completion of open backquotes reported by Egmont Koblinger subst.h - new flag for skip_to_delim: SD_EXTGLOB. Skip extended globbing patterns while looking for ending delimiter subst.c - when passed the SD_EXTGLOB flag, skip_to_delim skips over extended globbing patterns (when extended_glob is set) while looking for a character in the delimiter set pathexp.c - split_ignorespec: new function to replace calls to extract_colon_unit in setup_ignore_patterns. uses skip_to_delim with the SD_EXTGLOB flag to skip over extended globbing patterns in variables like HISTIGNORE and GLOBIGNORE. Fixes bug reported by Dimitar DIMITROV and Greg Wooledge 8/28 ---- lib/readline/rlprivate.h - add members to search_cxt to save _rl_keymap - new flag for isearch context: SF_CHGKMAP, set if we changed the keymap while reading a character for the search string that translated to a command lib/readline/isearch.c - save current readline keymap in cxt->keymap and cxt->okeymap in _rl_scxt_alloc - in _rl_isearch_dispatch, only check for cxt->lastc as a member of cxt->search_terminators if it's > 0 (i.e., not an isearch opcode) 9/3 --- support/signames.c - add Solaris SIGJVM1 and SIGJVM2. Update from Stefan Teleman shell.c - instead of closing all fds 3-20 at shell startup, just set them to be close-on-exec. Report from Rainer Müller lib/readline/isearch.c - in _rl_isearch_dispatch, if the current character maps to ISKMAP, move to the indicated keymap (using cxt->keymap) and go on to read another character. Fixes problem reported by Davor Cubranic - in _rl_isearch_dispatch, after translating key to possible opcode, restore _rl_keymap from cxt->okeymap if necessary - in _rl_isearch_dispatch, use key sequences that map to default functions that ^G, ^W, and ^Y map to as equivalent to those chars lib/readline/complete.c - new variable, _rl_menu_complete_prefix_first, zero by default - change menu_complete to display common prefix (matches[0]) first before cycling through rest of match list if _rl_menu_complete_prefix_first is non-zero. Suggested by Sami Pietila lib/readline/bind.c - new bindable readline variable, "menu-complete-display-prefix", controls setting of _rl_menu_complete_prefix_first doc/{bash.1,bashref.texi},lib/readline/doc/{readline.3,rluser.texi} - added description of menu-complete-display-prefix bindable readline variable 9/17 ---- configure.in - remove AM_PATH_LISPDIR call since we don't use that bash debugger any more. Suggested by Mike Frysinger 10/6 ---- findcmd.c - change executable_file to set errno to EISDIR if the passed name is a directory builtins/exec.def - change exec_builtin to report appropriate error message if the file argument is a directory. Noted by Eric Blake in a message to austin-group builtins/source.def - change source_builtin to make sure the shell exits if the file is not found when in a non-interactive shell running in posix mode and source_searches_cwd == 0 (as posix mode makes it by default). Pointed out in http://thread.gmane.org/gmane.comp.shells.dash/291/focus=392 by Jilles Tjoelker execute_cmd.c - set executing_command_builtin in execute_builtin if the builtin is command_builtin. Unwind-protected in execute_function_or_builtin (like executing_builtin variable). Available for rest of shell builtins/{source.def,evalfile.c} - make sure that non-interactive posix mode shells exit if the file argument to `.' is not found only if they are not being executed by the command builtin (executing_command_builtin == 0). This is how `command' can cancel effects of special builtin exit properties in the case of `dot file not found' 10/13 ----- lib/sh/strtrans.c - pass \c through unchanged if not escaping for `echo -e' and they are the final two characters in the string 10/15 ----- subst.c - extract_dollar_brace_string: fix problem with single quotes in unquoted ${...} for Posix compliance 10/16 ----- builtins/exec.def - catch return value from shell_execve; don't print duplicate error message if return value is EX_NOTFOUND. Make sure exit status from exec is 127 if command is not found execute_cmd.c - fix typo (`saved_redirects' should be `saved redirects') in execute_function_or_builtin `command exec' case. Typo caused too much of the unwind-protect stack to be discarded - in same execute_function_or_builtin case, don't discard the `saved redirects' frame unconditionally; only discard it if saved_redirects is non-null in the `command exec' case. Fixes sh -c 'command exec; exit 1' hanging bug uncovered by FreeBSD sh test cases 10/18 ----- subst.c - when in posix mode, shell should not exit if a variable assignment error (e.g., assigning to readonly variable) occurs preceding a command that is not a special builtin. Fixes bug uncovered by FreeBSD sh test cases - when in posix mode, the ${!?} and ${!#} expansions are not indirect expansions, but posix word expansions involving the `!' variable parse.y - fix parse_comsub so that it does not skip backslash-newline when parsing a comment 10/19 ----- subst.c - fix parameter_brace_expand so that an attempt to use the % or # expansions on an unset variable with -u set will cause a non- interactive shell to abort. Posix change - fix parameter_brace_expand so that an attempt to use pattern substitution or case modification expansions on an unset variable with -u set will cause and unbound variable error and make a non-interactive shell abort - change parameter_brace_expand_length to return INTMAX_MIN if a positional parameter is unset and -u is set - if parameter_brace_expand_length returns INTMAX_MIN when -u is set, treat it as an unbound variable error and make a non-interactive shell abort. Posix change - change parameter_brace_expand_length to return INTMAX_MIN if an implicit reference to array[0] is made ${#array} and array[0] is not set when -u is set 10/20 ----- builtins/cd.def - Posix 2008 says that if no matching directories are found in $CDPATH, use the directory name passed as an operand and go on. Posix change doc/bashref.texi - change Posix mode section with latest additions and removals 11/4 ---- lib/readline/complete.c - fix rl_menu_complete and rl_old_menu_complete to keep incrementing match_list_index by match_list_size as long as it's < 0. Fixes bug reported by jeenuv@gmail.com braces.c - make mkseq() take intmax_t arguments for sequence start and end and make sure it's passed intmax_t values. Fixes bug reported by Pete Gregory sig.c - if termsig_handler is called when terminate_immediately == 1, assume we're being called as a signal handler and set history_lines_this_session to 0 to inhibit history file being written on shell exit. Fixes long-standing bug most recently observed by Andreas Schwab 11/5 ---- redir.c - add_undo_close_redirect now returns int, 0 on success, non-zero on failure. Currently always succeeds - new macro REDIRECTION_ERROR to make do_redirection_internal return value of errno - change do_redirection_internal to call REDIRECTION_ERROR after saving file descriptor and make do_redirection_internal return error if add_undo_redirect or add_undo_close redirect fails. This makes failure to save a file descriptor a redirection error and the shell behaves appropriately. Fixes bug reported by Eric Blake bashline.c - modify bash_forward_shellword to correctly handle quoted strings, especially if point is in a quoted string when function is invoked. Fixes bug reported by Daniel Colascione configure.in - change version to 4.2-alpha 11/7 ---- lib/readline/text.c - in rl_insert, if we're not in the multibyte code path, don't try to optimize and insert all of the available typeahead input if we're reading input from a macro. Fixes bug reported by Andre Majorel lib/readline/text.c - break out multibyte guts of rl_forward_char into a separate function _rl_forward_char_internal that does nothing but calculate the new value of point - change rl_forward_char to call _rl_forward_char_internal instead of having equivalent code inline lib/readline/rlprivate.h - new extern declaration for _rl_forward_char_internal lib/readline/vi_mode.c - change _rl_vi_append_forward to call _rl_forward_char_internal to set rl_point, instead of calling rl_forward_char. When at the end of the line, rl_forward_char will ring the bell. Fixes debian bash bug 601042, reported by Alan J. Greenberger 11/14 ----- subst.c - fix match_upattern to use correct test to immediately break out of loop (when potential match length is greater than number of chars remaining in the string) in MATCH_ANY case 11/15 ----- subst.c - include "typemax.h" to make sure we have a definition of INTMAX_MIN 11/16 ----- lib/sh/unicode.c - make sure `localconv' isn't declared on machines without iconv - add stub_charset for systems that don't have locale_charset: looks up LC_CTYPE, returns everything after last `.', "UTF-8" if the value is exactly "UTF-8", and "ASCII" otherwise 11/20 ----- lib/readline/vi_mode.c - in rl_domove_motion_callback, make sure to use m->key instead of key, which is not initialized and should not be used. Bug report from Andreas Schwab - in rl_vi_domove, make assignment to `m' explicit instead of relying on evaluation order semantics, since the C standard leaves them unspecified. Bug report from Andreas Schwab 11/21 ----- lib/sh/shquote.c - sh_single_quote and sh_double_quote now take a const char * argument. Fixes problem pointed out by Joachim Schmitz externs.h - change extern declarations for sh_single_quote and sh_double_quote lib/sh/strchrnul.c - make sure that return value is cast to (char *) if we're using a part of the passed (const char *) argument. Fixes problem pointed out by Joachim Schmitz lib/glob/gmisc.c - fix a typo that mixed up defines for LPAREN and RPAREN. Bug and fix from Andreas Schwab - use WLPAREN and WRPAREN in multibyte character environments - fixed typos using L'cc' in a non-wide-char environment lib/readline/complete.c - fix rl_filename_completion_function to dequote users_dirname if there is a filename dequoting function (as well as dirname), since users_dirname gets tacked back onto the beginning of the possible completions and then requoted. Bug reported by Andreas Schwab 11/22 ----- lib/readline/parens.c - the `blink-matching-paren' variable should default to off 11/23 ----- subst.h - add extern declaration for close_new_fifos() lib/sh/fnxform.c - fix curencoding to return the character past the `.', not a string beginning with `.' lib/sh/unicode.c - fix stub_charset to do the same cut-off at `@' as curencoding(). These two functions should be combined builtins/printf.def - document new %(datefmt)T modifier in help text 11/24 ----- parse.y - fix `W' case in decode_prompt_string: memmove was copying one too few bytes and missed the closing NUL. Bug report from Tim Mooney 11/26 ----- subst.c - in expand_word_internal, don't add quoted nulls to partially- quoted strings if the word will not be subjected to word splitting later (which will remove the quoted null). Fixes bug reported by Rocky Bernstein 11/28 ----- subst.c - change multibyte case of match_pattern to revert to match_upattern if neither the pattern nor the string has any multibyte characters alias.c - fix tests of backslash-escaped characters in skipquotes, skipws, rd_token to check for backslash at EOS and not go past the end. Fixes debian bug 603696 reported by Tim Small include/shmbchar.h - new file, mbchar.h from gnulib minus the include lib/sh/shmbchar.c - new file, mbchar.c from gnulib with additions - moved mbstrlen from subst.c to here, changed initialization of mbs - change mbstrlen to use is_basic to avoid calls to mbrlen for ASCII chars; code hints from gnulib - don't copy mbs and mbsbak if we're not calling mbrlen 11/29 ----- lib/glob/smatch.c - change xstrmatch to use internal_strmatch() if the pattern and string don't have any multibyte characters 11/30 ----- include/shmbutil.h - change ADVANCE_CHAR and ADVANCE_CHAR_P macros to use is_basic and only call mbrlen and copy state and state_bak if is_basic returns false (non-ASCII). Called all over the place. - change rest of macros except BACKUP_CHAR and BACKUP_CHAR_P in the same way 12/2 ---- subst.c - audit all calls to string_list and make sure caller can handle a NULL return value. Fixes bug reported by David Rochberg general.h - change sh_wassign_func_t to take an additional argument: an int flags word subst.c - change do_word_assignment to take an additional argument to match wassign_func_t; change callers - change call to (*assign_func) in expand_word_list_internal to match new wassign_func_t prototype - (*assign_func) passes 1 as additional arg if the simple command is a builtin or function, in which case the assignment to the temporary env should take effect variables.c - change assign_in_env to take an additional argument to match wassign_func_t; change callers - move call to sv_ifs from dispose_temporary_env to dispose_used_env_vars; we don't need to do it if called from merge_temporary_env 12/3 ---- variables.c - change dispose_temporary_env to maintain a list (tempvar_list) of variables that need to be handled specially. If a variable that gets freed by push_temp_var or propagate_temp_var is one of the variables that the shell handles specially (IFS, LANG, etc.), it's stored on the list. For each variable in this list, dispose_temp_var calls stupidly_hack_special_variables. - assign_in_env calls stupidly_hack_special_variables if flags arg is non-zero, so variable assignments affect current shell execution environment if a builtin or function is being executed. Fixes bug reported by Bruno Haible 12/5 ---- subst.c - use mbsmbchar on both string and pattern in match_pattern instead of strlen and mbstrlen; only go through the strings once 12/6 ---- lib/readline/kill.c - in rl_yank_last_arg, only switch directions if the `count' argument is < 0, not < 1. This makes explicit count arguments of 0 work as expected. Fixes bug reported by Dennis Williamson doc/bash.1,lib/readline/doc/{readline.3,rluser.texi} - fix documentation for yank-last-arg to make it clear how the count argument is set and how second and subsequent calls treat any numeric argument doc/{bash.1,bashref.texi} - slight changes to the description of test - change \(bv to `|'; it seems that many `internationalized' versions of groff don't render that as a vertical bar. Fixes Debian bug 603805 12/10 ----- configure.in - changed release status to 4.2-beta [bash-4.2-beta frozen] 12/18 ----- redir.c - change REDIRECTION_ERROR macro to accept a third argument: an additional file descriptor to close before returning and error (pass -1 to do nothing) - change calls to REDIRECTION_ERROR to close appropriate file descriptors. Fixes bug reported by Andreas Schwab - make sure to close any file descriptors opened for REDIR_VARASSIGN before returning an error 12/19 ----- expr.c - move processing of unary `-' and `+' to exp1 from exp0 to avoid precedence problems. Fixes bug reported by <12bric@gmail.com> 12/21 ----- lib/sh/unicode.c - fixed reversed arguments to wctomb. Reported by Andreas Schwab