diff options
Diffstat (limited to 'src/google/protobuf/io')
21 files changed, 2046 insertions, 259 deletions
diff --git a/src/google/protobuf/io/coded_stream.cc b/src/google/protobuf/io/coded_stream.cc index 6a91a13..f6a8453 100644 --- a/src/google/protobuf/io/coded_stream.cc +++ b/src/google/protobuf/io/coded_stream.cc @@ -43,7 +43,7 @@ #include <limits.h> #include <google/protobuf/io/zero_copy_stream.h> #include <google/protobuf/stubs/common.h> -#include <google/protobuf/stubs/stl_util-inl.h> +#include <google/protobuf/stubs/stl_util.h> namespace google { @@ -56,10 +56,36 @@ static const int kMaxVarintBytes = 10; static const int kMaxVarint32Bytes = 5; +inline bool NextNonEmpty(ZeroCopyInputStream* input, + const void** data, int* size) { + bool success; + do { + success = input->Next(data, size); + } while (success && *size == 0); + return success; +} + } // namespace // CodedInputStream ================================================== +CodedInputStream::~CodedInputStream() { + if (input_ != NULL) { + BackUpInputToCurrentPosition(); + } + + if (total_bytes_warning_threshold_ == -2) { + GOOGLE_LOG(WARNING) << "The total number of bytes read was " << total_bytes_read_; + } +} + +// Static. +int CodedInputStream::default_recursion_limit_ = 100; + + +void CodedOutputStream::EnableAliasing(bool enabled) { + aliasing_enabled_ = enabled && output_->AllowsAliasing(); +} void CodedInputStream::BackUpInputToCurrentPosition() { int backup_bytes = BufferSize() + buffer_size_after_limit_ + overflow_bytes_; @@ -89,8 +115,7 @@ inline void CodedInputStream::RecomputeBufferLimits() { CodedInputStream::Limit CodedInputStream::PushLimit(int byte_limit) { // Current position relative to the beginning of the stream. - int current_position = total_bytes_read_ - - (BufferSize() + buffer_size_after_limit_); + int current_position = CurrentPosition(); Limit old_limit = current_limit_; @@ -124,10 +149,9 @@ void CodedInputStream::PopLimit(Limit limit) { legitimate_message_end_ = false; } -int CodedInputStream::BytesUntilLimit() { +int CodedInputStream::BytesUntilLimit() const { if (current_limit_ == INT_MAX) return -1; - int current_position = total_bytes_read_ - - (BufferSize() + buffer_size_after_limit_); + int current_position = CurrentPosition(); return current_limit_ - current_position; } @@ -136,13 +160,22 @@ void CodedInputStream::SetTotalBytesLimit( int total_bytes_limit, int warning_threshold) { // Make sure the limit isn't already past, since this could confuse other // code. - int current_position = total_bytes_read_ - - (BufferSize() + buffer_size_after_limit_); + int current_position = CurrentPosition(); total_bytes_limit_ = max(current_position, total_bytes_limit); - total_bytes_warning_threshold_ = warning_threshold; + if (warning_threshold >= 0) { + total_bytes_warning_threshold_ = warning_threshold; + } else { + // warning_threshold is negative + total_bytes_warning_threshold_ = -1; + } RecomputeBufferLimits(); } +int CodedInputStream::BytesUntilTotalBytesLimit() const { + if (total_bytes_limit_ == INT_MAX) return -1; + return total_bytes_limit_ - CurrentPosition(); +} + void CodedInputStream::PrintTotalBytesLimitError() { GOOGLE_LOG(ERROR) << "A protocol message was rejected because it was too " "big (more than " << total_bytes_limit_ @@ -223,6 +256,14 @@ bool CodedInputStream::ReadStringFallback(string* buffer, int size) { buffer->clear(); } + int closest_limit = min(current_limit_, total_bytes_limit_); + if (closest_limit != INT_MAX) { + int bytes_to_limit = closest_limit - CurrentPosition(); + if (bytes_to_limit > 0 && size > 0 && size <= bytes_to_limit) { + buffer->reserve(size); + } + } + int current_buffer_size; while ((current_buffer_size = BufferSize()) < size) { // Some STL implementations "helpfully" crash on buffer->append(NULL, 0). @@ -289,11 +330,16 @@ inline const uint8* ReadVarint32FromArray(const uint8* buffer, uint32* value) { uint32 b; uint32 result; - b = *(ptr++); result = (b & 0x7F) ; if (!(b & 0x80)) goto done; - b = *(ptr++); result |= (b & 0x7F) << 7; if (!(b & 0x80)) goto done; - b = *(ptr++); result |= (b & 0x7F) << 14; if (!(b & 0x80)) goto done; - b = *(ptr++); result |= (b & 0x7F) << 21; if (!(b & 0x80)) goto done; - b = *(ptr++); result |= b << 28; if (!(b & 0x80)) goto done; + b = *(ptr++); result = b ; if (!(b & 0x80)) goto done; + result -= 0x80; + b = *(ptr++); result += b << 7; if (!(b & 0x80)) goto done; + result -= 0x80 << 7; + b = *(ptr++); result += b << 14; if (!(b & 0x80)) goto done; + result -= 0x80 << 14; + b = *(ptr++); result += b << 21; if (!(b & 0x80)) goto done; + result -= 0x80 << 21; + b = *(ptr++); result += b << 28; if (!(b & 0x80)) goto done; + // "result -= 0x80 << 28" is irrevelant. // If the input is larger than 32 bits, we still need to read it all // and discard the high-order bits. @@ -323,8 +369,8 @@ bool CodedInputStream::ReadVarint32Slow(uint32* value) { bool CodedInputStream::ReadVarint32Fallback(uint32* value) { if (BufferSize() >= kMaxVarintBytes || - // Optimization: If the varint ends at exactly the end of the buffer, - // we can detect that and still use the fast path. + // Optimization: We're also safe if the buffer is non-empty and it ends + // with a byte that would terminate a varint. (buffer_end_ > buffer_ && !(buffer_end_[-1] & 0x80))) { const uint8* end = ReadVarint32FromArray(buffer_, value); if (end == NULL) return false; @@ -359,16 +405,17 @@ uint32 CodedInputStream::ReadTagSlow() { // For the slow path, just do a 64-bit read. Try to optimize for one-byte tags // again, since we have now refreshed the buffer. - uint64 result; + uint64 result = 0; if (!ReadVarint64(&result)) return 0; return static_cast<uint32>(result); } uint32 CodedInputStream::ReadTagFallback() { - if (BufferSize() >= kMaxVarintBytes || - // Optimization: If the varint ends at exactly the end of the buffer, - // we can detect that and still use the fast path. - (buffer_end_ > buffer_ && !(buffer_end_[-1] & 0x80))) { + const int buf_size = BufferSize(); + if (buf_size >= kMaxVarintBytes || + // Optimization: We're also safe if the buffer is non-empty and it ends + // with a byte that would terminate a varint. + (buf_size > 0 && !(buffer_end_[-1] & 0x80))) { uint32 tag; const uint8* end = ReadVarint32FromArray(buffer_, &tag); if (end == NULL) { @@ -379,7 +426,9 @@ uint32 CodedInputStream::ReadTagFallback() { } else { // We are commonly at a limit when attempting to read tags. Try to quickly // detect this case without making another function call. - if (buffer_ == buffer_end_ && buffer_size_after_limit_ > 0 && + if ((buf_size == 0) && + ((buffer_size_after_limit_ > 0) || + (total_bytes_read_ == current_limit_)) && // Make sure that the limit we hit is not total_bytes_limit_, since // in that case we still need to call Refresh() so that it prints an // error. @@ -417,8 +466,8 @@ bool CodedInputStream::ReadVarint64Slow(uint64* value) { bool CodedInputStream::ReadVarint64Fallback(uint64* value) { if (BufferSize() >= kMaxVarintBytes || - // Optimization: If the varint ends at exactly the end of the buffer, - // we can detect that and still use the fast path. + // Optimization: We're also safe if the buffer is non-empty and it ends + // with a byte that would terminate a varint. (buffer_end_ > buffer_ && !(buffer_end_[-1] & 0x80))) { // Fast path: We have enough bytes left in the buffer to guarantee that // this read won't cross the end, so we can skip the checks. @@ -430,20 +479,30 @@ bool CodedInputStream::ReadVarint64Fallback(uint64* value) { // processors. uint32 part0 = 0, part1 = 0, part2 = 0; - b = *(ptr++); part0 = (b & 0x7F) ; if (!(b & 0x80)) goto done; - b = *(ptr++); part0 |= (b & 0x7F) << 7; if (!(b & 0x80)) goto done; - b = *(ptr++); part0 |= (b & 0x7F) << 14; if (!(b & 0x80)) goto done; - b = *(ptr++); part0 |= (b & 0x7F) << 21; if (!(b & 0x80)) goto done; - b = *(ptr++); part1 = (b & 0x7F) ; if (!(b & 0x80)) goto done; - b = *(ptr++); part1 |= (b & 0x7F) << 7; if (!(b & 0x80)) goto done; - b = *(ptr++); part1 |= (b & 0x7F) << 14; if (!(b & 0x80)) goto done; - b = *(ptr++); part1 |= (b & 0x7F) << 21; if (!(b & 0x80)) goto done; - b = *(ptr++); part2 = (b & 0x7F) ; if (!(b & 0x80)) goto done; - b = *(ptr++); part2 |= (b & 0x7F) << 7; if (!(b & 0x80)) goto done; + b = *(ptr++); part0 = b ; if (!(b & 0x80)) goto done; + part0 -= 0x80; + b = *(ptr++); part0 += b << 7; if (!(b & 0x80)) goto done; + part0 -= 0x80 << 7; + b = *(ptr++); part0 += b << 14; if (!(b & 0x80)) goto done; + part0 -= 0x80 << 14; + b = *(ptr++); part0 += b << 21; if (!(b & 0x80)) goto done; + part0 -= 0x80 << 21; + b = *(ptr++); part1 = b ; if (!(b & 0x80)) goto done; + part1 -= 0x80; + b = *(ptr++); part1 += b << 7; if (!(b & 0x80)) goto done; + part1 -= 0x80 << 7; + b = *(ptr++); part1 += b << 14; if (!(b & 0x80)) goto done; + part1 -= 0x80 << 14; + b = *(ptr++); part1 += b << 21; if (!(b & 0x80)) goto done; + part1 -= 0x80 << 21; + b = *(ptr++); part2 = b ; if (!(b & 0x80)) goto done; + part2 -= 0x80; + b = *(ptr++); part2 += b << 7; if (!(b & 0x80)) goto done; + // "part2 -= 0x80 << 7" is irrelevant because (0x80 << 7) << 56 is 0. // We have overrun the maximum size of a varint (10 bytes). The data // must be corrupt. - return NULL; + return false; done: Advance(ptr - buffer_); @@ -483,13 +542,13 @@ bool CodedInputStream::Refresh() { "CodedInputStream::SetTotalBytesLimit() in " "google/protobuf/io/coded_stream.h."; - // Don't warn again for this stream. - total_bytes_warning_threshold_ = -1; + // Don't warn again for this stream, and print total size at the end. + total_bytes_warning_threshold_ = -2; } const void* void_buffer; int buffer_size; - if (input_->Next(&void_buffer, &buffer_size)) { + if (NextNonEmpty(input_, &void_buffer, &buffer_size)) { buffer_ = reinterpret_cast<const uint8*>(void_buffer); buffer_end_ = buffer_ + buffer_size; GOOGLE_CHECK_GE(buffer_size, 0); @@ -528,7 +587,8 @@ CodedOutputStream::CodedOutputStream(ZeroCopyOutputStream* output) buffer_(NULL), buffer_size_(0), total_bytes_(0), - had_error_(false) { + had_error_(false), + aliasing_enabled_(false) { // Eagerly Refresh() so buffer space is immediately available. Refresh(); // The Refresh() may have failed. If the client doesn't write any data, @@ -582,6 +642,23 @@ uint8* CodedOutputStream::WriteRawToArray( } +void CodedOutputStream::WriteAliasedRaw(const void* data, int size) { + if (size < buffer_size_ + ) { + WriteRaw(data, size); + } else { + if (buffer_size_ > 0) { + output_->BackUp(buffer_size_); + total_bytes_ -= buffer_size_; + buffer_ = NULL; + buffer_size_ = 0; + } + + total_bytes_ += size; + had_error_ |= !output_->WriteAliasedRaw(data, size); + } +} + void CodedOutputStream::WriteLittleEndian32(uint32 value) { uint8 bytes[sizeof(value)]; @@ -825,6 +902,13 @@ int CodedOutputStream::VarintSize64(uint64 value) { } } +uint8* CodedOutputStream::WriteStringWithSizeToArray(const string& str, + uint8* target) { + GOOGLE_DCHECK_LE(str.size(), kuint32max); + target = WriteVarint32ToArray(str.size(), target); + return WriteStringToArray(str, target); +} + } // namespace io } // namespace protobuf } // namespace google diff --git a/src/google/protobuf/io/coded_stream.h b/src/google/protobuf/io/coded_stream.h index e5f6161..50a03a1 100644 --- a/src/google/protobuf/io/coded_stream.h +++ b/src/google/protobuf/io/coded_stream.h @@ -110,14 +110,27 @@ #define GOOGLE_PROTOBUF_IO_CODED_STREAM_H__ #include <string> -#ifndef _MSC_VER -#include <sys/param.h> -#endif // !_MSC_VER +#ifdef _MSC_VER + #if defined(_M_IX86) && \ + !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST) + #define PROTOBUF_LITTLE_ENDIAN 1 + #endif + #if _MSC_VER >= 1300 + // If MSVC has "/RTCc" set, it will complain about truncating casts at + // runtime. This file contains some intentional truncating casts. + #pragma runtime_checks("c", off) + #endif +#else + #include <sys/param.h> // __BYTE_ORDER + #if defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN && \ + !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST) + #define PROTOBUF_LITTLE_ENDIAN 1 + #endif +#endif #include <google/protobuf/stubs/common.h> -#include <google/protobuf/stubs/common.h> // for GOOGLE_PREDICT_TRUE macro -namespace google { +namespace google { namespace protobuf { class DescriptorPool; @@ -157,6 +170,9 @@ class LIBPROTOBUF_EXPORT CodedInputStream { // successfully and the stream's byte limit. ~CodedInputStream(); + // Return true if this CodedInputStream reads from a flat array instead of + // a ZeroCopyInputStream. + inline bool IsFlat() const; // Skips a number of bytes. Returns false if an underlying read error // occurs. @@ -217,11 +233,22 @@ class LIBPROTOBUF_EXPORT CodedInputStream { // Read a tag. This calls ReadVarint32() and returns the result, or returns // zero (which is not a valid tag) if ReadVarint32() fails. Also, it updates // the last tag value, which can be checked with LastTagWas(). - // Always inline because this is only called in once place per parse loop + // Always inline because this is only called in one place per parse loop // but it is called for every iteration of said loop, so it should be fast. // GCC doesn't want to inline this by default. uint32 ReadTag() GOOGLE_ATTRIBUTE_ALWAYS_INLINE; + // This usually a faster alternative to ReadTag() when cutoff is a manifest + // constant. It does particularly well for cutoff >= 127. The first part + // of the return value is the tag that was read, though it can also be 0 in + // the cases where ReadTag() would return 0. If the second part is true + // then the tag is known to be in [0, cutoff]. If not, the tag either is + // above cutoff or is 0. (There's intentional wiggle room when tag is 0, + // because that can arise in several ways, and for best performance we want + // to avoid an extra "is tag == 0?" check here.) + inline std::pair<uint32, bool> ReadTagWithCutoff(uint32 cutoff) + GOOGLE_ATTRIBUTE_ALWAYS_INLINE; + // Usually returns true if calling ReadVarint32() now would produce the given // value. Will always return false if ReadVarint32() would not return the // given value. If ExpectTag() returns true, it also advances past @@ -248,8 +275,8 @@ class LIBPROTOBUF_EXPORT CodedInputStream { // zero, and ConsumedEntireMessage() will return true. bool ExpectAtEnd(); - // If the last call to ReadTag() returned the given value, returns true. - // Otherwise, returns false; + // If the last call to ReadTag() or ReadTagWithCutoff() returned the + // given value, returns true. Otherwise, returns false; // // This is needed because parsers for some types of embedded messages // (with field type TYPE_GROUP) don't actually know that they've reached the @@ -298,7 +325,10 @@ class LIBPROTOBUF_EXPORT CodedInputStream { // Returns the number of bytes left until the nearest limit on the // stack is hit, or -1 if no limits are in place. - int BytesUntilLimit(); + int BytesUntilLimit() const; + + // Returns current position relative to the beginning of the input stream. + int CurrentPosition() const; // Total Bytes Limit ----------------------------------------------- // To prevent malicious users from sending excessively large messages @@ -314,8 +344,9 @@ class LIBPROTOBUF_EXPORT CodedInputStream { // cause integer overflows is 512MB. The default limit is 64MB. Apps // should set shorter limits if possible. If warning_threshold is not -1, // a warning will be printed to stderr after warning_threshold bytes are - // read. An error will always be printed to stderr if the limit is - // reached. + // read. For backwards compatibility all negative values get squashed to -1, + // as other negative values might have special internal meanings. + // An error will always be printed to stderr if the limit is reached. // // This is unrelated to PushLimit()/PopLimit(). // @@ -336,15 +367,20 @@ class LIBPROTOBUF_EXPORT CodedInputStream { // something unusual. void SetTotalBytesLimit(int total_bytes_limit, int warning_threshold); + // The Total Bytes Limit minus the Current Position, or -1 if there + // is no Total Bytes Limit. + int BytesUntilTotalBytesLimit() const; + // Recursion Limit ------------------------------------------------- // To prevent corrupt or malicious messages from causing stack overflows, // we must keep track of the depth of recursion when parsing embedded // messages and groups. CodedInputStream keeps track of this because it // is the only object that is passed down the stack during parsing. - // Sets the maximum recursion depth. The default is 64. + // Sets the maximum recursion depth. The default is 100. void SetRecursionLimit(int limit); + // Increments the current recursion depth. Returns true if the depth is // under the limit, false if it has gone over. bool IncrementRecursionDepth(); @@ -420,7 +456,8 @@ class LIBPROTOBUF_EXPORT CodedInputStream { // // Note that this feature is ignored when parsing "lite" messages as they do // not have descriptors. - void SetExtensionRegistry(DescriptorPool* pool, MessageFactory* factory); + void SetExtensionRegistry(const DescriptorPool* pool, + MessageFactory* factory); // Get the DescriptorPool set via SetExtensionRegistry(), or NULL if no pool // has been provided. @@ -444,7 +481,7 @@ class LIBPROTOBUF_EXPORT CodedInputStream { int overflow_bytes_; // LastTagWas() stuff. - uint32 last_tag_; // result of last ReadTag(). + uint32 last_tag_; // result of last ReadTag() or ReadTagWithCutoff(). // This is set true by ReadTag{Fallback/Slow}() if it is called when exactly // at EOF, or by ExpectAtEnd() when it returns true. This happens when we @@ -469,6 +506,11 @@ class LIBPROTOBUF_EXPORT CodedInputStream { // Maximum number of bytes to read, period. This is unrelated to // current_limit_. Set using SetTotalBytesLimit(). int total_bytes_limit_; + + // If positive/0: Limit for bytes read after which a warning due to size + // should be logged. + // If -1: Printing of warning disabled. Can be set by client. + // If -2: Internal: Limit has been reached, print full size when destructing. int total_bytes_warning_threshold_; // Current recursion depth, controlled by IncrementRecursionDepth() and @@ -521,12 +563,13 @@ class LIBPROTOBUF_EXPORT CodedInputStream { bool ReadStringFallback(string* buffer, int size); // Return the size of the buffer. - uint32 BufferSize() const; + int BufferSize() const; static const int kDefaultTotalBytesLimit = 64 << 20; // 64MB static const int kDefaultTotalBytesWarningThreshold = 32 << 20; // 32MB - static const int kDefaultRecursionLimit = 64; + + static int default_recursion_limit_; // 100 by default. }; // Class which encodes and writes binary data which is composed of varint- @@ -554,7 +597,7 @@ class LIBPROTOBUF_EXPORT CodedInputStream { // char text[] = "Hello world!"; // // int coded_size = sizeof(magic_number) + -// CodedOutputStream::Varint32Size(strlen(text)) + +// CodedOutputStream::VarintSize32(strlen(text)) + // strlen(text); // // uint8* buffer = @@ -610,6 +653,9 @@ class LIBPROTOBUF_EXPORT CodedOutputStream { // Write raw bytes, copying them from the given buffer. void WriteRaw(const void* buffer, int size); + // Like WriteRaw() but will try to write aliased data if aliasing is + // turned on. + void WriteRawMaybeAliased(const void* data, int size); // Like WriteRaw() but writing directly to the target array. // This is _not_ inlined, as the compiler often optimizes memcpy into inline // copy loops. Since this gets called by every field with string or bytes @@ -621,8 +667,21 @@ class LIBPROTOBUF_EXPORT CodedOutputStream { void WriteString(const string& str); // Like WriteString() but writing directly to the target array. static uint8* WriteStringToArray(const string& str, uint8* target); + // Write the varint-encoded size of str followed by str. + static uint8* WriteStringWithSizeToArray(const string& str, uint8* target); + // Instructs the CodedOutputStream to allow the underlying + // ZeroCopyOutputStream to hold pointers to the original structure instead of + // copying, if it supports it (i.e. output->AllowsAliasing() is true). If the + // underlying stream does not support aliasing, then enabling it has no + // affect. For now, this only affects the behavior of + // WriteRawMaybeAliased(). + // + // NOTE: It is caller's responsibility to ensure that the chunk of memory + // remains live until all of the data has been consumed from the stream. + void EnableAliasing(bool enabled); + // Write a 32-bit little-endian integer. void WriteLittleEndian32(uint32 value); // Like WriteLittleEndian32() but writing directly to the target array. @@ -667,6 +726,21 @@ class LIBPROTOBUF_EXPORT CodedOutputStream { // If negative, 10 bytes. Otheriwse, same as VarintSize32(). static int VarintSize32SignExtended(int32 value); + // Compile-time equivalent of VarintSize32(). + template <uint32 Value> + struct StaticVarintSize32 { + static const int value = + (Value < (1 << 7)) + ? 1 + : (Value < (1 << 14)) + ? 2 + : (Value < (1 << 21)) + ? 3 + : (Value < (1 << 28)) + ? 4 + : 5; + }; + // Returns the total number of bytes written since this object was created. inline int ByteCount() const; @@ -682,6 +756,7 @@ class LIBPROTOBUF_EXPORT CodedOutputStream { int buffer_size_; int total_bytes_; // Sum of sizes of all buffers seen so far. bool had_error_; // Whether an error occurred during output. + bool aliasing_enabled_; // See EnableAliasing(). // Advance the buffer by a given number of bytes. void Advance(int amount); @@ -690,6 +765,10 @@ class LIBPROTOBUF_EXPORT CodedOutputStream { // Advance(buffer_size_). bool Refresh(); + // Like WriteRaw() but may avoid copying if the underlying + // ZeroCopyOutputStream supports it. + void WriteAliasedRaw(const void* buffer, int size); + static uint8* WriteVarint32FallbackToArray(uint32 value, uint8* target); // Always-inlined versions of WriteVarint* functions so that code can be @@ -735,8 +814,7 @@ inline bool CodedInputStream::ReadVarint64(uint64* value) { inline const uint8* CodedInputStream::ReadLittleEndian32FromArray( const uint8* buffer, uint32* value) { -#if !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST) && \ - defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN +#if defined(PROTOBUF_LITTLE_ENDIAN) memcpy(value, buffer, sizeof(*value)); return buffer + sizeof(*value); #else @@ -751,8 +829,7 @@ inline const uint8* CodedInputStream::ReadLittleEndian32FromArray( inline const uint8* CodedInputStream::ReadLittleEndian64FromArray( const uint8* buffer, uint64* value) { -#if !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST) && \ - defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN +#if defined(PROTOBUF_LITTLE_ENDIAN) memcpy(value, buffer, sizeof(*value)); return buffer + sizeof(*value); #else @@ -771,9 +848,8 @@ inline const uint8* CodedInputStream::ReadLittleEndian64FromArray( } inline bool CodedInputStream::ReadLittleEndian32(uint32* value) { -#if !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST) && \ - defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN - if (GOOGLE_PREDICT_TRUE(BufferSize() >= sizeof(*value))) { +#if defined(PROTOBUF_LITTLE_ENDIAN) + if (GOOGLE_PREDICT_TRUE(BufferSize() >= static_cast<int>(sizeof(*value)))) { memcpy(value, buffer_, sizeof(*value)); Advance(sizeof(*value)); return true; @@ -786,9 +862,8 @@ inline bool CodedInputStream::ReadLittleEndian32(uint32* value) { } inline bool CodedInputStream::ReadLittleEndian64(uint64* value) { -#if !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST) && \ - defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN - if (GOOGLE_PREDICT_TRUE(BufferSize() >= sizeof(*value))) { +#if defined(PROTOBUF_LITTLE_ENDIAN) + if (GOOGLE_PREDICT_TRUE(BufferSize() >= static_cast<int>(sizeof(*value)))) { memcpy(value, buffer_, sizeof(*value)); Advance(sizeof(*value)); return true; @@ -811,6 +886,45 @@ inline uint32 CodedInputStream::ReadTag() { } } +inline std::pair<uint32, bool> CodedInputStream::ReadTagWithCutoff( + uint32 cutoff) { + // In performance-sensitive code we can expect cutoff to be a compile-time + // constant, and things like "cutoff >= kMax1ByteVarint" to be evaluated at + // compile time. + if (GOOGLE_PREDICT_TRUE(buffer_ < buffer_end_)) { + // Hot case: buffer_ non_empty, buffer_[0] in [1, 128). + // TODO(gpike): Is it worth rearranging this? E.g., if the number of fields + // is large enough then is it better to check for the two-byte case first? + if (static_cast<int8>(buffer_[0]) > 0) { + const uint32 kMax1ByteVarint = 0x7f; + uint32 tag = last_tag_ = buffer_[0]; + Advance(1); + return make_pair(tag, cutoff >= kMax1ByteVarint || tag <= cutoff); + } + // Other hot case: cutoff >= 0x80, buffer_ has at least two bytes available, + // and tag is two bytes. The latter is tested by bitwise-and-not of the + // first byte and the second byte. + if (cutoff >= 0x80 && + GOOGLE_PREDICT_TRUE(buffer_ + 1 < buffer_end_) && + GOOGLE_PREDICT_TRUE((buffer_[0] & ~buffer_[1]) >= 0x80)) { + const uint32 kMax2ByteVarint = (0x7f << 7) + 0x7f; + uint32 tag = last_tag_ = (1u << 7) * buffer_[1] + (buffer_[0] - 0x80); + Advance(2); + // It might make sense to test for tag == 0 now, but it is so rare that + // that we don't bother. A varint-encoded 0 should be one byte unless + // the encoder lost its mind. The second part of the return value of + // this function is allowed to be either true or false if the tag is 0, + // so we don't have to check for tag == 0. We may need to check whether + // it exceeds cutoff. + bool at_or_below_cutoff = cutoff >= kMax2ByteVarint || tag <= cutoff; + return make_pair(tag, at_or_below_cutoff); + } + } + // Slow path + last_tag_ = ReadTagFallback(); + return make_pair(last_tag_, static_cast<uint32>(last_tag_ - 1) < cutoff); +} + inline bool CodedInputStream::LastTagWas(uint32 expected) { return last_tag_ == expected; } @@ -867,7 +981,9 @@ inline bool CodedInputStream::ExpectAtEnd() { // If we are at a limit we know no more bytes can be read. Otherwise, it's // hard to say without calling Refresh(), and we'd rather not do that. - if (buffer_ == buffer_end_ && buffer_size_after_limit_ != 0) { + if (buffer_ == buffer_end_ && + ((buffer_size_after_limit_ != 0) || + (total_bytes_read_ == current_limit_))) { last_tag_ = 0; // Pretend we called ReadTag()... legitimate_message_end_ = true; // ... and it hit EOF. return true; @@ -876,6 +992,10 @@ inline bool CodedInputStream::ExpectAtEnd() { } } +inline int CodedInputStream::CurrentPosition() const { + return total_bytes_read_ - (BufferSize() + buffer_size_after_limit_); +} + inline uint8* CodedOutputStream::GetDirectBufferForNBytesAndAdvance(int size) { if (buffer_size_ < size) { return NULL; @@ -915,8 +1035,7 @@ inline uint8* CodedOutputStream::WriteVarint32SignExtendedToArray( inline uint8* CodedOutputStream::WriteLittleEndian32ToArray(uint32 value, uint8* target) { -#if !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST) && \ - defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN +#if defined(PROTOBUF_LITTLE_ENDIAN) memcpy(target, &value, sizeof(value)); #else target[0] = static_cast<uint8>(value); @@ -929,8 +1048,7 @@ inline uint8* CodedOutputStream::WriteLittleEndian32ToArray(uint32 value, inline uint8* CodedOutputStream::WriteLittleEndian64ToArray(uint64 value, uint8* target) { -#if !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST) && \ - defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN +#if defined(PROTOBUF_LITTLE_ENDIAN) memcpy(target, &value, sizeof(value)); #else uint32 part0 = static_cast<uint32>(value); @@ -983,12 +1101,21 @@ inline int CodedOutputStream::VarintSize32SignExtended(int32 value) { } inline void CodedOutputStream::WriteString(const string& str) { - WriteRaw(str.data(), str.size()); + WriteRaw(str.data(), static_cast<int>(str.size())); +} + +inline void CodedOutputStream::WriteRawMaybeAliased( + const void* data, int size) { + if (aliasing_enabled_) { + WriteAliasedRaw(data, size); + } else { + WriteRaw(data, size); + } } inline uint8* CodedOutputStream::WriteStringToArray( const string& str, uint8* target) { - return WriteRawToArray(str.data(), str.size(), target); + return WriteRawToArray(str.data(), static_cast<int>(str.size()), target); } inline int CodedOutputStream::ByteCount() const { @@ -1017,7 +1144,7 @@ inline void CodedInputStream::DecrementRecursionDepth() { if (recursion_depth_ > 0) --recursion_depth_; } -inline void CodedInputStream::SetExtensionRegistry(DescriptorPool* pool, +inline void CodedInputStream::SetExtensionRegistry(const DescriptorPool* pool, MessageFactory* factory) { extension_pool_ = pool; extension_factory_ = factory; @@ -1031,7 +1158,7 @@ inline MessageFactory* CodedInputStream::GetExtensionFactory() { return extension_factory_; } -inline uint32 CodedInputStream::BufferSize() const { +inline int CodedInputStream::BufferSize() const { return buffer_end_ - buffer_; } @@ -1044,12 +1171,12 @@ inline CodedInputStream::CodedInputStream(ZeroCopyInputStream* input) last_tag_(0), legitimate_message_end_(false), aliasing_enabled_(false), - current_limit_(INT_MAX), + current_limit_(kint32max), buffer_size_after_limit_(0), total_bytes_limit_(kDefaultTotalBytesLimit), total_bytes_warning_threshold_(kDefaultTotalBytesWarningThreshold), recursion_depth_(0), - recursion_limit_(kDefaultRecursionLimit), + recursion_limit_(default_recursion_limit_), extension_pool_(NULL), extension_factory_(NULL) { // Eagerly Refresh() so buffer space is immediately available. @@ -1070,21 +1197,24 @@ inline CodedInputStream::CodedInputStream(const uint8* buffer, int size) total_bytes_limit_(kDefaultTotalBytesLimit), total_bytes_warning_threshold_(kDefaultTotalBytesWarningThreshold), recursion_depth_(0), - recursion_limit_(kDefaultRecursionLimit), + recursion_limit_(default_recursion_limit_), extension_pool_(NULL), extension_factory_(NULL) { // Note that setting current_limit_ == size is important to prevent some // code paths from trying to access input_ and segfaulting. } -inline CodedInputStream::~CodedInputStream() { - if (input_ != NULL) { - BackUpInputToCurrentPosition(); - } +inline bool CodedInputStream::IsFlat() const { + return input_ == NULL; } } // namespace io } // namespace protobuf + +#if defined(_MSC_VER) && _MSC_VER >= 1300 + #pragma runtime_checks("c", restore) +#endif // _MSC_VER + } // namespace google #endif // GOOGLE_PROTOBUF_IO_CODED_STREAM_H__ diff --git a/src/google/protobuf/io/coded_stream_inl.h b/src/google/protobuf/io/coded_stream_inl.h index e9799d4..41dc10e 100644 --- a/src/google/protobuf/io/coded_stream_inl.h +++ b/src/google/protobuf/io/coded_stream_inl.h @@ -37,8 +37,9 @@ #define GOOGLE_PROTOBUF_IO_CODED_STREAM_INL_H__ #include <google/protobuf/io/coded_stream.h> +#include <google/protobuf/io/zero_copy_stream_impl_lite.h> #include <string> -#include <google/protobuf/stubs/stl_util-inl.h> +#include <google/protobuf/stubs/stl_util.h> namespace google { namespace protobuf { @@ -50,8 +51,12 @@ inline bool CodedInputStream::InternalReadStringInline(string* buffer, if (BufferSize() >= size) { STLStringResizeUninitialized(buffer, size); - memcpy(string_as_array(buffer), buffer_, size); - Advance(size); + // When buffer is empty, string_as_array(buffer) will return NULL but memcpy + // requires non-NULL pointers even when size is 0. Hench this check. + if (size > 0) { + memcpy(mutable_string_data(buffer), buffer_, size); + Advance(size); + } return true; } diff --git a/src/google/protobuf/io/coded_stream_unittest.cc b/src/google/protobuf/io/coded_stream_unittest.cc index 7d29833..0cfeb88 100644 --- a/src/google/protobuf/io/coded_stream_unittest.cc +++ b/src/google/protobuf/io/coded_stream_unittest.cc @@ -44,7 +44,6 @@ #include <google/protobuf/testing/googletest.h> #include <gtest/gtest.h> #include <google/protobuf/io/zero_copy_stream_impl.h> -#include <google/protobuf/stubs/strutil.h> // This declares an unsigned long long integer literal in a portable way. @@ -125,6 +124,13 @@ namespace { class CodedStreamTest : public testing::Test { protected: + // Helper method used by tests for bytes warning. See implementation comment + // for further information. + static void SetupTotalBytesLimitWarningTest( + int total_bytes_limit, int warning_threshold, + vector<string>* out_errors, vector<string>* out_warnings); + + // Buffer used during most of the tests. This assumes tests run sequentially. static const int kBufferSize = 1024 * 64; static uint8 buffer_[kBufferSize]; }; @@ -208,6 +214,33 @@ TEST_2D(CodedStreamTest, ReadTag, kVarintCases, kBlockSizes) { EXPECT_EQ(kVarintCases_case.size, input.ByteCount()); } +// This is the regression test that verifies that there is no issues +// with the empty input buffers handling. +TEST_F(CodedStreamTest, EmptyInputBeforeEos) { + class In : public ZeroCopyInputStream { + public: + In() : count_(0) {} + private: + virtual bool Next(const void** data, int* size) { + *data = NULL; + *size = 0; + return count_++ < 2; + } + virtual void BackUp(int count) { + GOOGLE_LOG(FATAL) << "Tests never call this."; + } + virtual bool Skip(int count) { + GOOGLE_LOG(FATAL) << "Tests never call this."; + return false; + } + virtual int64 ByteCount() const { return 0; } + int count_; + } in; + CodedInputStream input(&in); + input.ReadTag(); + EXPECT_TRUE(input.ConsumedEntireMessage()); +} + TEST_1D(CodedStreamTest, ExpectTag, kVarintCases) { // Leave one byte at the beginning of the buffer so we can read it // to force the first buffer to be loaded. @@ -649,14 +682,197 @@ TEST_F(CodedStreamTest, ReadStringImpossiblyLargeFromStringOnHeap) { EXPECT_FALSE(coded_input.ReadString(&str, 1 << 30)); } +TEST_1D(CodedStreamTest, ReadStringReservesMemoryOnTotalLimit, kBlockSizes) { + memcpy(buffer_, kRawBytes, sizeof(kRawBytes)); + ArrayInputStream input(buffer_, sizeof(buffer_), kBlockSizes_case); + + { + CodedInputStream coded_input(&input); + coded_input.SetTotalBytesLimit(sizeof(kRawBytes), sizeof(kRawBytes)); + EXPECT_EQ(sizeof(kRawBytes), coded_input.BytesUntilTotalBytesLimit()); + + string str; + EXPECT_TRUE(coded_input.ReadString(&str, strlen(kRawBytes))); + EXPECT_EQ(sizeof(kRawBytes) - strlen(kRawBytes), + coded_input.BytesUntilTotalBytesLimit()); + EXPECT_EQ(kRawBytes, str); + // TODO(liujisi): Replace with a more meaningful test (see cl/60966023). + EXPECT_GE(str.capacity(), strlen(kRawBytes)); + } + + EXPECT_EQ(strlen(kRawBytes), input.ByteCount()); +} + +TEST_1D(CodedStreamTest, ReadStringReservesMemoryOnPushedLimit, kBlockSizes) { + memcpy(buffer_, kRawBytes, sizeof(kRawBytes)); + ArrayInputStream input(buffer_, sizeof(buffer_), kBlockSizes_case); + + { + CodedInputStream coded_input(&input); + coded_input.PushLimit(sizeof(buffer_)); + + string str; + EXPECT_TRUE(coded_input.ReadString(&str, strlen(kRawBytes))); + EXPECT_EQ(kRawBytes, str); + // TODO(liujisi): Replace with a more meaningful test (see cl/60966023). + EXPECT_GE(str.capacity(), strlen(kRawBytes)); + } + + EXPECT_EQ(strlen(kRawBytes), input.ByteCount()); +} + +TEST_F(CodedStreamTest, ReadStringNoReservationIfLimitsNotSet) { + memcpy(buffer_, kRawBytes, sizeof(kRawBytes)); + // Buffer size in the input must be smaller than sizeof(kRawBytes), + // otherwise check against capacity will fail as ReadStringInline() + // will handle the reading and will reserve the memory as needed. + ArrayInputStream input(buffer_, sizeof(buffer_), 32); + + { + CodedInputStream coded_input(&input); + + string str; + EXPECT_TRUE(coded_input.ReadString(&str, strlen(kRawBytes))); + EXPECT_EQ(kRawBytes, str); + // Note: this check depends on string class implementation. It + // expects that string will allocate more than strlen(kRawBytes) + // if the content of kRawBytes is appended to string in small + // chunks. + // TODO(liujisi): Replace with a more meaningful test (see cl/60966023). + EXPECT_GE(str.capacity(), strlen(kRawBytes)); + } + + EXPECT_EQ(strlen(kRawBytes), input.ByteCount()); +} + +TEST_F(CodedStreamTest, ReadStringNoReservationSizeIsNegative) { + memcpy(buffer_, kRawBytes, sizeof(kRawBytes)); + // Buffer size in the input must be smaller than sizeof(kRawBytes), + // otherwise check against capacity will fail as ReadStringInline() + // will handle the reading and will reserve the memory as needed. + ArrayInputStream input(buffer_, sizeof(buffer_), 32); + + { + CodedInputStream coded_input(&input); + coded_input.PushLimit(sizeof(buffer_)); + + string str; + EXPECT_FALSE(coded_input.ReadString(&str, -1)); + // Note: this check depends on string class implementation. It + // expects that string will always allocate the same amount of + // memory for an empty string. + EXPECT_EQ(string().capacity(), str.capacity()); + } +} + +TEST_F(CodedStreamTest, ReadStringNoReservationSizeIsLarge) { + memcpy(buffer_, kRawBytes, sizeof(kRawBytes)); + // Buffer size in the input must be smaller than sizeof(kRawBytes), + // otherwise check against capacity will fail as ReadStringInline() + // will handle the reading and will reserve the memory as needed. + ArrayInputStream input(buffer_, sizeof(buffer_), 32); + + { + CodedInputStream coded_input(&input); + coded_input.PushLimit(sizeof(buffer_)); + + string str; + EXPECT_FALSE(coded_input.ReadString(&str, 1 << 30)); + EXPECT_GT(1 << 30, str.capacity()); + } +} + +TEST_F(CodedStreamTest, ReadStringNoReservationSizeIsOverTheLimit) { + memcpy(buffer_, kRawBytes, sizeof(kRawBytes)); + // Buffer size in the input must be smaller than sizeof(kRawBytes), + // otherwise check against capacity will fail as ReadStringInline() + // will handle the reading and will reserve the memory as needed. + ArrayInputStream input(buffer_, sizeof(buffer_), 32); + + { + CodedInputStream coded_input(&input); + coded_input.PushLimit(16); + + string str; + EXPECT_FALSE(coded_input.ReadString(&str, strlen(kRawBytes))); + // Note: this check depends on string class implementation. It + // expects that string will allocate less than strlen(kRawBytes) + // for an empty string. + EXPECT_GT(strlen(kRawBytes), str.capacity()); + } +} + +TEST_F(CodedStreamTest, ReadStringNoReservationSizeIsOverTheTotalBytesLimit) { + memcpy(buffer_, kRawBytes, sizeof(kRawBytes)); + // Buffer size in the input must be smaller than sizeof(kRawBytes), + // otherwise check against capacity will fail as ReadStringInline() + // will handle the reading and will reserve the memory as needed. + ArrayInputStream input(buffer_, sizeof(buffer_), 32); + + { + CodedInputStream coded_input(&input); + coded_input.SetTotalBytesLimit(16, 16); + + string str; + EXPECT_FALSE(coded_input.ReadString(&str, strlen(kRawBytes))); + // Note: this check depends on string class implementation. It + // expects that string will allocate less than strlen(kRawBytes) + // for an empty string. + EXPECT_GT(strlen(kRawBytes), str.capacity()); + } +} + +TEST_F(CodedStreamTest, + ReadStringNoReservationSizeIsOverTheClosestLimit_GlobalLimitIsCloser) { + memcpy(buffer_, kRawBytes, sizeof(kRawBytes)); + // Buffer size in the input must be smaller than sizeof(kRawBytes), + // otherwise check against capacity will fail as ReadStringInline() + // will handle the reading and will reserve the memory as needed. + ArrayInputStream input(buffer_, sizeof(buffer_), 32); + + { + CodedInputStream coded_input(&input); + coded_input.PushLimit(sizeof(buffer_)); + coded_input.SetTotalBytesLimit(16, 16); + + string str; + EXPECT_FALSE(coded_input.ReadString(&str, strlen(kRawBytes))); + // Note: this check depends on string class implementation. It + // expects that string will allocate less than strlen(kRawBytes) + // for an empty string. + EXPECT_GT(strlen(kRawBytes), str.capacity()); + } +} + +TEST_F(CodedStreamTest, + ReadStringNoReservationSizeIsOverTheClosestLimit_LocalLimitIsCloser) { + memcpy(buffer_, kRawBytes, sizeof(kRawBytes)); + // Buffer size in the input must be smaller than sizeof(kRawBytes), + // otherwise check against capacity will fail as ReadStringInline() + // will handle the reading and will reserve the memory as needed. + ArrayInputStream input(buffer_, sizeof(buffer_), 32); + + { + CodedInputStream coded_input(&input); + coded_input.PushLimit(16); + coded_input.SetTotalBytesLimit(sizeof(buffer_), sizeof(buffer_)); + EXPECT_EQ(sizeof(buffer_), coded_input.BytesUntilTotalBytesLimit()); + + string str; + EXPECT_FALSE(coded_input.ReadString(&str, strlen(kRawBytes))); + // Note: this check depends on string class implementation. It + // expects that string will allocate less than strlen(kRawBytes) + // for an empty string. + EXPECT_GT(strlen(kRawBytes), str.capacity()); + } +} + // ------------------------------------------------------------------- // Skip const char kSkipTestBytes[] = "<Before skipping><To be skipped><After skipping>"; -const char kSkipOutputTestBytes[] = - "-----------------<To be skipped>----------------"; TEST_1D(CodedStreamTest, SkipInput, kBlockSizes) { memcpy(buffer_, kSkipTestBytes, sizeof(kSkipTestBytes)); @@ -947,9 +1163,11 @@ TEST_F(CodedStreamTest, TotalBytesLimit) { ArrayInputStream input(buffer_, sizeof(buffer_)); CodedInputStream coded_input(&input); coded_input.SetTotalBytesLimit(16, -1); + EXPECT_EQ(16, coded_input.BytesUntilTotalBytesLimit()); string str; EXPECT_TRUE(coded_input.ReadString(&str, 16)); + EXPECT_EQ(0, coded_input.BytesUntilTotalBytesLimit()); vector<string> errors; @@ -964,7 +1182,9 @@ TEST_F(CodedStreamTest, TotalBytesLimit) { "A protocol message was rejected because it was too big", errors[0]); coded_input.SetTotalBytesLimit(32, -1); + EXPECT_EQ(16, coded_input.BytesUntilTotalBytesLimit()); EXPECT_TRUE(coded_input.ReadString(&str, 16)); + EXPECT_EQ(0, coded_input.BytesUntilTotalBytesLimit()); } TEST_F(CodedStreamTest, TotalBytesLimitNotValidMessageEnd) { @@ -995,6 +1215,60 @@ TEST_F(CodedStreamTest, TotalBytesLimitNotValidMessageEnd) { EXPECT_FALSE(coded_input.ConsumedEntireMessage()); } +// This method is used by the tests below. +// It constructs a CodedInputStream with the given limits and tries to read 2KiB +// of data from it. Then it returns the logged errors and warnings in the given +// vectors. +void CodedStreamTest::SetupTotalBytesLimitWarningTest( + int total_bytes_limit, int warning_threshold, + vector<string>* out_errors, vector<string>* out_warnings) { + ArrayInputStream raw_input(buffer_, sizeof(buffer_), 128); + + ScopedMemoryLog scoped_log; + { + CodedInputStream input(&raw_input); + input.SetTotalBytesLimit(total_bytes_limit, warning_threshold); + string str; + EXPECT_TRUE(input.ReadString(&str, 2048)); + } + + *out_errors = scoped_log.GetMessages(ERROR); + *out_warnings = scoped_log.GetMessages(WARNING); +} + +TEST_F(CodedStreamTest, TotalBytesLimitWarning) { + vector<string> errors; + vector<string> warnings; + SetupTotalBytesLimitWarningTest(10240, 1024, &errors, &warnings); + + EXPECT_EQ(0, errors.size()); + + ASSERT_EQ(2, warnings.size()); + EXPECT_PRED_FORMAT2(testing::IsSubstring, + "Reading dangerously large protocol message. If the message turns out to " + "be larger than 10240 bytes, parsing will be halted for security reasons.", + warnings[0]); + EXPECT_PRED_FORMAT2(testing::IsSubstring, + "The total number of bytes read was 2048", + warnings[1]); +} + +TEST_F(CodedStreamTest, TotalBytesLimitWarningDisabled) { + vector<string> errors; + vector<string> warnings; + + // Test with -1 + SetupTotalBytesLimitWarningTest(10240, -1, &errors, &warnings); + EXPECT_EQ(0, errors.size()); + EXPECT_EQ(0, warnings.size()); + + // Test again with -2, expecting the same result + SetupTotalBytesLimitWarningTest(10240, -2, &errors, &warnings); + EXPECT_EQ(0, errors.size()); + EXPECT_EQ(0, warnings.size()); +} + + TEST_F(CodedStreamTest, RecursionLimit) { ArrayInputStream input(buffer_, sizeof(buffer_)); CodedInputStream coded_input(&input); @@ -1032,6 +1306,7 @@ TEST_F(CodedStreamTest, RecursionLimit) { EXPECT_FALSE(coded_input.IncrementRecursionDepth()); // 7 } + class ReallyBigInputStream : public ZeroCopyInputStream { public: ReallyBigInputStream() : backup_amount_(0), buffer_count_(0) {} diff --git a/src/google/protobuf/io/gzip_stream.cc b/src/google/protobuf/io/gzip_stream.cc index 84d277f..fe1f331 100644 --- a/src/google/protobuf/io/gzip_stream.cc +++ b/src/google/protobuf/io/gzip_stream.cc @@ -73,6 +73,17 @@ GzipInputStream::~GzipInputStream() { zerror_ = inflateEnd(&zcontext_); } +static inline int internalInflateInit2( + z_stream* zcontext, GzipInputStream::Format format) { + int windowBitsFormat = 0; + switch (format) { + case GzipInputStream::GZIP: windowBitsFormat = 16; break; + case GzipInputStream::AUTO: windowBitsFormat = 32; break; + case GzipInputStream::ZLIB: windowBitsFormat = 0; break; + } + return inflateInit2(zcontext, /* windowBits */15 | windowBitsFormat); +} + int GzipInputStream::Inflate(int flush) { if ((zerror_ == Z_OK) && (zcontext_.avail_out == 0)) { // previous inflate filled output buffer. don't change input params yet. @@ -89,14 +100,7 @@ int GzipInputStream::Inflate(int flush) { zcontext_.next_in = static_cast<Bytef*>(const_cast<void*>(in)); zcontext_.avail_in = in_size; if (first) { - int windowBitsFormat = 0; - switch (format_) { - case GZIP: windowBitsFormat = 16; break; - case AUTO: windowBitsFormat = 32; break; - case ZLIB: windowBitsFormat = 0; break; - } - int error = inflateInit2(&zcontext_, - /* windowBits */15 | windowBitsFormat); + int error = internalInflateInit2(&zcontext_, format_); if (error != Z_OK) { return error; } @@ -127,9 +131,21 @@ bool GzipInputStream::Next(const void** data, int* size) { return true; } if (zerror_ == Z_STREAM_END) { - *data = NULL; - *size = 0; - return false; + if (zcontext_.next_out != NULL) { + // sub_stream_ may have concatenated streams to follow + zerror_ = inflateEnd(&zcontext_); + if (zerror_ != Z_OK) { + return false; + } + zerror_ = internalInflateInit2(&zcontext_, format_); + if (zerror_ != Z_OK) { + return false; + } + } else { + *data = NULL; + *size = 0; + return false; + } } zerror_ = Inflate(Z_NO_FLUSH); if ((zerror_ == Z_STREAM_END) && (zcontext_.next_out == NULL)) { @@ -183,16 +199,6 @@ GzipOutputStream::GzipOutputStream(ZeroCopyOutputStream* sub_stream, Init(sub_stream, options); } -GzipOutputStream::GzipOutputStream( - ZeroCopyOutputStream* sub_stream, Format format, int buffer_size) { - Options options; - options.format = format; - if (buffer_size != -1) { - options.buffer_size = buffer_size; - } - Init(sub_stream, options); -} - void GzipOutputStream::Init(ZeroCopyOutputStream* sub_stream, const Options& options) { sub_stream_ = sub_stream; @@ -251,8 +257,7 @@ int GzipOutputStream::Deflate(int flush) { } error = deflate(&zcontext_, flush); } while (error == Z_OK && zcontext_.avail_out == 0); - if (((flush == Z_FULL_FLUSH) || (flush == Z_FINISH)) - && (zcontext_.avail_out != sub_data_size_)) { + if ((flush == Z_FULL_FLUSH) || (flush == Z_FINISH)) { // Notify lower layer of data. sub_stream_->BackUp(zcontext_.avail_out); // We don't own the buffer anymore. @@ -294,10 +299,11 @@ int64 GzipOutputStream::ByteCount() const { } bool GzipOutputStream::Flush() { - do { - zerror_ = Deflate(Z_FULL_FLUSH); - } while (zerror_ == Z_OK); - return zerror_ == Z_OK; + zerror_ = Deflate(Z_FULL_FLUSH); + // Return true if the flush succeeded or if it was a no-op. + return (zerror_ == Z_OK) || + (zerror_ == Z_BUF_ERROR && zcontext_.avail_in == 0 && + zcontext_.avail_out != 0); } bool GzipOutputStream::Close() { diff --git a/src/google/protobuf/io/gzip_stream.h b/src/google/protobuf/io/gzip_stream.h index 65dbc5b..7ee24bc 100644 --- a/src/google/protobuf/io/gzip_stream.h +++ b/src/google/protobuf/io/gzip_stream.h @@ -45,6 +45,7 @@ #include <zlib.h> +#include <google/protobuf/stubs/common.h> #include <google/protobuf/io/zero_copy_stream.h> namespace google { @@ -144,12 +145,6 @@ class LIBPROTOBUF_EXPORT GzipOutputStream : public ZeroCopyOutputStream { ZeroCopyOutputStream* sub_stream, const Options& options); - // DEPRECATED: Use one of the above constructors instead. - GzipOutputStream( - ZeroCopyOutputStream* sub_stream, - Format format, - int buffer_size = -1) GOOGLE_ATTRIBUTE_DEPRECATED; - virtual ~GzipOutputStream(); // Return last error message or NULL if no error. @@ -165,6 +160,13 @@ class LIBPROTOBUF_EXPORT GzipOutputStream : public ZeroCopyOutputStream { // necessary. // Compression may be less efficient stopping and starting around flushes. // Returns true if no error. + // + // Please ensure that block size is > 6. Here is an excerpt from the zlib + // doc that explains why: + // + // In the case of a Z_FULL_FLUSH or Z_SYNC_FLUSH, make sure that avail_out + // is greater than six to avoid repeated flush markers due to + // avail_out == 0 on return. bool Flush(); // Writes out all data and closes the gzip stream. diff --git a/src/google/protobuf/io/printer.cc b/src/google/protobuf/io/printer.cc index c7d3074..d2bf3f5 100644 --- a/src/google/protobuf/io/printer.cc +++ b/src/google/protobuf/io/printer.cc @@ -35,7 +35,6 @@ #include <google/protobuf/io/printer.h> #include <google/protobuf/io/zero_copy_stream.h> #include <google/protobuf/stubs/common.h> -#include <google/protobuf/stubs/strutil.h> namespace google { namespace protobuf { @@ -51,8 +50,8 @@ Printer::Printer(ZeroCopyOutputStream* output, char variable_delimiter) } Printer::~Printer() { - // Only BackUp() if we're sure we've successfully called Next() at least once. - if (buffer_size_ > 0) { + // Only BackUp() if we have called Next() at least once and never failed. + if (buffer_size_ > 0 && !failed_) { output_->BackUp(buffer_size_); } } @@ -132,6 +131,17 @@ void Printer::Print(const char* text, Print(vars, text); } +void Printer::Print(const char* text, + const char* variable1, const string& value1, + const char* variable2, const string& value2, + const char* variable3, const string& value3) { + map<string, string> vars; + vars[variable1] = value1; + vars[variable2] = value2; + vars[variable3] = value3; + Print(vars, text); +} + void Printer::Indent() { indent_ += " "; } @@ -158,7 +168,7 @@ void Printer::WriteRaw(const char* data, int size) { if (failed_) return; if (size == 0) return; - if (at_start_of_line_) { + if (at_start_of_line_ && (size > 0) && (data[0] != '\n')) { // Insert an indent. at_start_of_line_ = false; WriteRaw(indent_.data(), indent_.size()); diff --git a/src/google/protobuf/io/printer.h b/src/google/protobuf/io/printer.h index de08538..5be4854 100644 --- a/src/google/protobuf/io/printer.h +++ b/src/google/protobuf/io/printer.h @@ -82,7 +82,11 @@ class LIBPROTOBUF_EXPORT Printer { // Like the first Print(), except the substitutions are given as parameters. void Print(const char* text, const char* variable1, const string& value1, const char* variable2, const string& value2); - // TODO(kenton): Overloaded versions with more variables? Two seems + // Like the first Print(), except the substitutions are given as parameters. + void Print(const char* text, const char* variable1, const string& value1, + const char* variable2, const string& value2, + const char* variable3, const string& value3); + // TODO(kenton): Overloaded versions with more variables? Three seems // to be enough. // Indent text by two spaces. After calling Indent(), two spaces will be diff --git a/src/google/protobuf/io/printer_unittest.cc b/src/google/protobuf/io/printer_unittest.cc index 580a53d..76fb944 100644 --- a/src/google/protobuf/io/printer_unittest.cc +++ b/src/google/protobuf/io/printer_unittest.cc @@ -220,7 +220,7 @@ TEST(Printer, Indenting) { } // Death tests do not work on Windows as of yet. -#ifdef GTEST_HAS_DEATH_TEST +#ifdef PROTOBUF_HAS_DEATH_TEST TEST(Printer, Death) { char buffer[8192]; @@ -231,9 +231,33 @@ TEST(Printer, Death) { EXPECT_DEBUG_DEATH(printer.Print("$unclosed"), "Unclosed variable name"); EXPECT_DEBUG_DEATH(printer.Outdent(), "without matching Indent"); } -#endif // GTEST_HAS_DEATH_TEST +#endif // PROTOBUF_HAS_DEATH_TEST -TEST(Printer, WriteFailure) { +TEST(Printer, WriteFailurePartial) { + char buffer[17]; + + ArrayOutputStream output(buffer, sizeof(buffer)); + Printer printer(&output, '$'); + + // Print 16 bytes to almost fill the buffer (should not fail). + printer.Print("0123456789abcdef"); + EXPECT_FALSE(printer.failed()); + + // Try to print 2 chars. Only one fits. + printer.Print("<>"); + EXPECT_TRUE(printer.failed()); + + // Anything else should fail too. + printer.Print(" "); + EXPECT_TRUE(printer.failed()); + printer.Print("blah"); + EXPECT_TRUE(printer.failed()); + + // Buffer should contain the first 17 bytes written. + EXPECT_EQ("0123456789abcdef<", string(buffer, sizeof(buffer))); +} + +TEST(Printer, WriteFailureExact) { char buffer[16]; ArrayOutputStream output(buffer, sizeof(buffer)); diff --git a/src/google/protobuf/io/strtod.cc b/src/google/protobuf/io/strtod.cc new file mode 100644 index 0000000..cf0c6e1 --- /dev/null +++ b/src/google/protobuf/io/strtod.cc @@ -0,0 +1,113 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2008 Google Inc. All rights reserved. +// http://code.google.com/p/protobuf/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include <google/protobuf/io/strtod.h> + +#include <cstdio> +#include <cstring> +#include <string> + +#include <google/protobuf/stubs/common.h> + +namespace google { +namespace protobuf { +namespace io { + +// ---------------------------------------------------------------------- +// NoLocaleStrtod() +// This code will make you cry. +// ---------------------------------------------------------------------- + +namespace { + +// Returns a string identical to *input except that the character pointed to +// by radix_pos (which should be '.') is replaced with the locale-specific +// radix character. +string LocalizeRadix(const char* input, const char* radix_pos) { + // Determine the locale-specific radix character by calling sprintf() to + // print the number 1.5, then stripping off the digits. As far as I can + // tell, this is the only portable, thread-safe way to get the C library + // to divuldge the locale's radix character. No, localeconv() is NOT + // thread-safe. + char temp[16]; + int size = sprintf(temp, "%.1f", 1.5); + GOOGLE_CHECK_EQ(temp[0], '1'); + GOOGLE_CHECK_EQ(temp[size-1], '5'); + GOOGLE_CHECK_LE(size, 6); + + // Now replace the '.' in the input with it. + string result; + result.reserve(strlen(input) + size - 3); + result.append(input, radix_pos); + result.append(temp + 1, size - 2); + result.append(radix_pos + 1); + return result; +} + +} // namespace + +double NoLocaleStrtod(const char* text, char** original_endptr) { + // We cannot simply set the locale to "C" temporarily with setlocale() + // as this is not thread-safe. Instead, we try to parse in the current + // locale first. If parsing stops at a '.' character, then this is a + // pretty good hint that we're actually in some other locale in which + // '.' is not the radix character. + + char* temp_endptr; + double result = strtod(text, &temp_endptr); + if (original_endptr != NULL) *original_endptr = temp_endptr; + if (*temp_endptr != '.') return result; + + // Parsing halted on a '.'. Perhaps we're in a different locale? Let's + // try to replace the '.' with a locale-specific radix character and + // try again. + string localized = LocalizeRadix(text, temp_endptr); + const char* localized_cstr = localized.c_str(); + char* localized_endptr; + result = strtod(localized_cstr, &localized_endptr); + if ((localized_endptr - localized_cstr) > + (temp_endptr - text)) { + // This attempt got further, so replacing the decimal must have helped. + // Update original_endptr to point at the right location. + if (original_endptr != NULL) { + // size_diff is non-zero if the localized radix has multiple bytes. + int size_diff = localized.size() - strlen(text); + // const_cast is necessary to match the strtod() interface. + *original_endptr = const_cast<char*>( + text + (localized_endptr - localized_cstr - size_diff)); + } + } + + return result; +} + +} // namespace io +} // namespace protobuf +} // namespace google diff --git a/src/google/protobuf/io/strtod.h b/src/google/protobuf/io/strtod.h new file mode 100644 index 0000000..2be3c43 --- /dev/null +++ b/src/google/protobuf/io/strtod.h @@ -0,0 +1,50 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2008 Google Inc. All rights reserved. +// http://code.google.com/p/protobuf/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +// A locale-independent version of strtod(), used to parse floating +// point default values in .proto files, where the decimal separator +// is always a dot. + +#ifndef GOOGLE_PROTOBUF_IO_STRTOD_H__ +#define GOOGLE_PROTOBUF_IO_STRTOD_H__ + +namespace google { +namespace protobuf { +namespace io { + +// A locale-independent version of the standard strtod(), which always +// uses a dot as the decimal separator. +double NoLocaleStrtod(const char* str, char** endptr); + +} // namespace io +} // namespace protobuf + +} // namespace google +#endif // GOOGLE_PROTOBUF_IO_STRTOD_H__ diff --git a/src/google/protobuf/io/tokenizer.cc b/src/google/protobuf/io/tokenizer.cc index 38fa351..d149305 100644 --- a/src/google/protobuf/io/tokenizer.cc +++ b/src/google/protobuf/io/tokenizer.cc @@ -89,8 +89,12 @@ // exactly pretty. #include <google/protobuf/io/tokenizer.h> +#include <google/protobuf/stubs/common.h> +#include <google/protobuf/stubs/stringprintf.h> +#include <google/protobuf/io/strtod.h> #include <google/protobuf/io/zero_copy_stream.h> #include <google/protobuf/stubs/strutil.h> +#include <google/protobuf/stubs/stl_util.h> namespace google { namespace protobuf { @@ -118,6 +122,8 @@ namespace { CHARACTER_CLASS(Whitespace, c == ' ' || c == '\n' || c == '\t' || c == '\r' || c == '\v' || c == '\f'); +CHARACTER_CLASS(WhitespaceNoNewline, c == ' ' || c == '\t' || + c == '\r' || c == '\v' || c == '\f'); CHARACTER_CLASS(Unprintable, c < ' ' && c > '\0'); @@ -187,12 +193,16 @@ Tokenizer::Tokenizer(ZeroCopyInputStream* input, read_error_(false), line_(0), column_(0), - token_start_(-1), + record_target_(NULL), + record_start_(-1), allow_f_after_float_(false), - comment_style_(CPP_COMMENT_STYLE) { + comment_style_(CPP_COMMENT_STYLE), + require_space_after_number_(true), + allow_multiline_strings_(false) { current_.line = 0; current_.column = 0; + current_.end_column = 0; current_.type = TYPE_START; Refresh(); @@ -237,9 +247,9 @@ void Tokenizer::Refresh() { } // If we're in a token, append the rest of the buffer to it. - if (token_start_ >= 0 && token_start_ < buffer_size_) { - current_.text.append(buffer_ + token_start_, buffer_size_ - token_start_); - token_start_ = 0; + if (record_target_ != NULL && record_start_ < buffer_size_) { + record_target_->append(buffer_ + record_start_, buffer_size_ - record_start_); + record_start_ = 0; } const void* data = NULL; @@ -260,23 +270,34 @@ void Tokenizer::Refresh() { current_char_ = buffer_[0]; } +inline void Tokenizer::RecordTo(string* target) { + record_target_ = target; + record_start_ = buffer_pos_; +} + +inline void Tokenizer::StopRecording() { + // Note: The if() is necessary because some STL implementations crash when + // you call string::append(NULL, 0), presumably because they are trying to + // be helpful by detecting the NULL pointer, even though there's nothing + // wrong with reading zero bytes from NULL. + if (buffer_pos_ != record_start_) { + record_target_->append(buffer_ + record_start_, buffer_pos_ - record_start_); + } + record_target_ = NULL; + record_start_ = -1; +} + inline void Tokenizer::StartToken() { - token_start_ = buffer_pos_; current_.type = TYPE_START; // Just for the sake of initializing it. current_.text.clear(); current_.line = line_; current_.column = column_; + RecordTo(¤t_.text); } inline void Tokenizer::EndToken() { - // Note: The if() is necessary because some STL implementations crash when - // you call string::append(NULL, 0), presumably because they are trying to - // be helpful by detecting the NULL pointer, even though there's nothing - // wrong with reading zero bytes from NULL. - if (buffer_pos_ != token_start_) { - current_.text.append(buffer_ + token_start_, buffer_pos_ - token_start_); - } - token_start_ = -1; + StopRecording(); + current_.end_column = column_; } // ------------------------------------------------------------------- @@ -332,9 +353,16 @@ void Tokenizer::ConsumeString(char delimiter) { while (true) { switch (current_char_) { case '\0': - case '\n': { - AddError("String literals cannot cross line boundaries."); + AddError("Unexpected end of string."); return; + + case '\n': { + if (!allow_multiline_strings_) { + AddError("String literals cannot cross line boundaries."); + return; + } + NextChar(); + break; } case '\\': { @@ -351,6 +379,27 @@ void Tokenizer::ConsumeString(char delimiter) { AddError("Expected hex digits for escape sequence."); } // Possibly followed by another hex digit, but again we don't care. + } else if (TryConsume('u')) { + if (!TryConsumeOne<HexDigit>() || + !TryConsumeOne<HexDigit>() || + !TryConsumeOne<HexDigit>() || + !TryConsumeOne<HexDigit>()) { + AddError("Expected four hex digits for \\u escape sequence."); + } + } else if (TryConsume('U')) { + // We expect 8 hex digits; but only the range up to 0x10ffff is + // legal. + if (!TryConsume('0') || + !TryConsume('0') || + !(TryConsume('0') || TryConsume('1')) || + !TryConsumeOne<HexDigit>() || + !TryConsumeOne<HexDigit>() || + !TryConsumeOne<HexDigit>() || + !TryConsumeOne<HexDigit>() || + !TryConsumeOne<HexDigit>()) { + AddError("Expected eight hex digits up to 10ffff for \\U escape " + "sequence"); + } } else { AddError("Invalid escape sequence in string literal."); } @@ -410,7 +459,7 @@ Tokenizer::TokenType Tokenizer::ConsumeNumber(bool started_with_zero, } } - if (LookingAt<Letter>()) { + if (LookingAt<Letter>() && require_space_after_number_) { AddError("Need space between number and identifier."); } else if (current_char_ == '.') { if (is_float) { @@ -424,26 +473,51 @@ Tokenizer::TokenType Tokenizer::ConsumeNumber(bool started_with_zero, return is_float ? TYPE_FLOAT : TYPE_INTEGER; } -void Tokenizer::ConsumeLineComment() { +void Tokenizer::ConsumeLineComment(string* content) { + if (content != NULL) RecordTo(content); + while (current_char_ != '\0' && current_char_ != '\n') { NextChar(); } TryConsume('\n'); + + if (content != NULL) StopRecording(); } -void Tokenizer::ConsumeBlockComment() { +void Tokenizer::ConsumeBlockComment(string* content) { int start_line = line_; int start_column = column_ - 2; + if (content != NULL) RecordTo(content); + while (true) { while (current_char_ != '\0' && current_char_ != '*' && - current_char_ != '/') { + current_char_ != '/' && + current_char_ != '\n') { NextChar(); } - if (TryConsume('*') && TryConsume('/')) { + if (TryConsume('\n')) { + if (content != NULL) StopRecording(); + + // Consume leading whitespace and asterisk; + ConsumeZeroOrMore<WhitespaceNoNewline>(); + if (TryConsume('*')) { + if (TryConsume('/')) { + // End of comment. + break; + } + } + + if (content != NULL) RecordTo(content); + } else if (TryConsume('*') && TryConsume('/')) { // End of comment. + if (content != NULL) { + StopRecording(); + // Strip trailing "*/". + content->erase(content->size() - 2); + } break; } else if (TryConsume('/') && current_char_ == '*') { // Note: We didn't consume the '*' because if there is a '/' after it @@ -454,42 +528,59 @@ void Tokenizer::ConsumeBlockComment() { AddError("End-of-file inside block comment."); error_collector_->AddError( start_line, start_column, " Comment started here."); + if (content != NULL) StopRecording(); break; } } } +Tokenizer::NextCommentStatus Tokenizer::TryConsumeCommentStart() { + if (comment_style_ == CPP_COMMENT_STYLE && TryConsume('/')) { + if (TryConsume('/')) { + return LINE_COMMENT; + } else if (TryConsume('*')) { + return BLOCK_COMMENT; + } else { + // Oops, it was just a slash. Return it. + current_.type = TYPE_SYMBOL; + current_.text = "/"; + current_.line = line_; + current_.column = column_ - 1; + current_.end_column = column_; + return SLASH_NOT_COMMENT; + } + } else if (comment_style_ == SH_COMMENT_STYLE && TryConsume('#')) { + return LINE_COMMENT; + } else { + return NO_COMMENT; + } +} + // ------------------------------------------------------------------- bool Tokenizer::Next() { - TokenType last_token_type = current_.type; - - // Did we skip any characters after the last token? - bool skipped_stuff = false; + previous_ = current_; while (!read_error_) { - if (TryConsumeOne<Whitespace>()) { - ConsumeZeroOrMore<Whitespace>(); - - } else if (comment_style_ == CPP_COMMENT_STYLE && TryConsume('/')) { - // Starting a comment? - if (TryConsume('/')) { - ConsumeLineComment(); - } else if (TryConsume('*')) { - ConsumeBlockComment(); - } else { - // Oops, it was just a slash. Return it. - current_.type = TYPE_SYMBOL; - current_.text = "/"; - current_.line = line_; - current_.column = column_ - 1; + ConsumeZeroOrMore<Whitespace>(); + + switch (TryConsumeCommentStart()) { + case LINE_COMMENT: + ConsumeLineComment(NULL); + continue; + case BLOCK_COMMENT: + ConsumeBlockComment(NULL); + continue; + case SLASH_NOT_COMMENT: return true; - } + case NO_COMMENT: + break; + } - } else if (comment_style_ == SH_COMMENT_STYLE && TryConsume('#')) { - ConsumeLineComment(); + // Check for EOF before continuing. + if (read_error_) break; - } else if (LookingAt<Unprintable>() || current_char_ == '\0') { + if (LookingAt<Unprintable>() || current_char_ == '\0') { AddError("Invalid control characters encountered in text."); NextChar(); // Skip more unprintable characters, too. But, remember that '\0' is @@ -517,7 +608,9 @@ bool Tokenizer::Next() { if (TryConsumeOne<Digit>()) { // It's a floating-point number. - if (last_token_type == TYPE_IDENTIFIER && !skipped_stuff) { + if (previous_.type == TYPE_IDENTIFIER && + current_.line == previous_.line && + current_.column == previous_.end_column) { // We don't accept syntax like "blah.123". error_collector_->AddError(line_, column_ - 2, "Need space between identifier and decimal point."); @@ -535,6 +628,12 @@ bool Tokenizer::Next() { ConsumeString('\''); current_.type = TYPE_STRING; } else { + // Check if the high order bit is set. + if (current_char_ & 0x80) { + error_collector_->AddError(line_, column_, + StringPrintf("Interpreting non ascii codepoint %d.", + static_cast<unsigned char>(current_char_))); + } NextChar(); current_.type = TYPE_SYMBOL; } @@ -542,8 +641,6 @@ bool Tokenizer::Next() { EndToken(); return true; } - - skipped_stuff = true; } // EOF @@ -551,9 +648,199 @@ bool Tokenizer::Next() { current_.text.clear(); current_.line = line_; current_.column = column_; + current_.end_column = column_; return false; } +namespace { + +// Helper class for collecting comments and putting them in the right places. +// +// This basically just buffers the most recent comment until it can be decided +// exactly where that comment should be placed. When Flush() is called, the +// current comment goes into either prev_trailing_comments or detached_comments. +// When the CommentCollector is destroyed, the last buffered comment goes into +// next_leading_comments. +class CommentCollector { + public: + CommentCollector(string* prev_trailing_comments, + vector<string>* detached_comments, + string* next_leading_comments) + : prev_trailing_comments_(prev_trailing_comments), + detached_comments_(detached_comments), + next_leading_comments_(next_leading_comments), + has_comment_(false), + is_line_comment_(false), + can_attach_to_prev_(true) { + if (prev_trailing_comments != NULL) prev_trailing_comments->clear(); + if (detached_comments != NULL) detached_comments->clear(); + if (next_leading_comments != NULL) next_leading_comments->clear(); + } + + ~CommentCollector() { + // Whatever is in the buffer is a leading comment. + if (next_leading_comments_ != NULL && has_comment_) { + comment_buffer_.swap(*next_leading_comments_); + } + } + + // About to read a line comment. Get the comment buffer pointer in order to + // read into it. + string* GetBufferForLineComment() { + // We want to combine with previous line comments, but not block comments. + if (has_comment_ && !is_line_comment_) { + Flush(); + } + has_comment_ = true; + is_line_comment_ = true; + return &comment_buffer_; + } + + // About to read a block comment. Get the comment buffer pointer in order to + // read into it. + string* GetBufferForBlockComment() { + if (has_comment_) { + Flush(); + } + has_comment_ = true; + is_line_comment_ = false; + return &comment_buffer_; + } + + void ClearBuffer() { + comment_buffer_.clear(); + has_comment_ = false; + } + + // Called once we know that the comment buffer is complete and is *not* + // connected to the next token. + void Flush() { + if (has_comment_) { + if (can_attach_to_prev_) { + if (prev_trailing_comments_ != NULL) { + prev_trailing_comments_->append(comment_buffer_); + } + can_attach_to_prev_ = false; + } else { + if (detached_comments_ != NULL) { + detached_comments_->push_back(comment_buffer_); + } + } + ClearBuffer(); + } + } + + void DetachFromPrev() { + can_attach_to_prev_ = false; + } + + private: + string* prev_trailing_comments_; + vector<string>* detached_comments_; + string* next_leading_comments_; + + string comment_buffer_; + + // True if any comments were read into comment_buffer_. This can be true even + // if comment_buffer_ is empty, namely if the comment was "/**/". + bool has_comment_; + + // Is the comment in the comment buffer a line comment? + bool is_line_comment_; + + // Is it still possible that we could be reading a comment attached to the + // previous token? + bool can_attach_to_prev_; +}; + +} // namespace + +bool Tokenizer::NextWithComments(string* prev_trailing_comments, + vector<string>* detached_comments, + string* next_leading_comments) { + CommentCollector collector(prev_trailing_comments, detached_comments, + next_leading_comments); + + if (current_.type == TYPE_START) { + collector.DetachFromPrev(); + } else { + // A comment appearing on the same line must be attached to the previous + // declaration. + ConsumeZeroOrMore<WhitespaceNoNewline>(); + switch (TryConsumeCommentStart()) { + case LINE_COMMENT: + ConsumeLineComment(collector.GetBufferForLineComment()); + + // Don't allow comments on subsequent lines to be attached to a trailing + // comment. + collector.Flush(); + break; + case BLOCK_COMMENT: + ConsumeBlockComment(collector.GetBufferForBlockComment()); + + ConsumeZeroOrMore<WhitespaceNoNewline>(); + if (!TryConsume('\n')) { + // Oops, the next token is on the same line. If we recorded a comment + // we really have no idea which token it should be attached to. + collector.ClearBuffer(); + return Next(); + } + + // Don't allow comments on subsequent lines to be attached to a trailing + // comment. + collector.Flush(); + break; + case SLASH_NOT_COMMENT: + return true; + case NO_COMMENT: + if (!TryConsume('\n')) { + // The next token is on the same line. There are no comments. + return Next(); + } + break; + } + } + + // OK, we are now on the line *after* the previous token. + while (true) { + ConsumeZeroOrMore<WhitespaceNoNewline>(); + + switch (TryConsumeCommentStart()) { + case LINE_COMMENT: + ConsumeLineComment(collector.GetBufferForLineComment()); + break; + case BLOCK_COMMENT: + ConsumeBlockComment(collector.GetBufferForBlockComment()); + + // Consume the rest of the line so that we don't interpret it as a + // blank line the next time around the loop. + ConsumeZeroOrMore<WhitespaceNoNewline>(); + TryConsume('\n'); + break; + case SLASH_NOT_COMMENT: + return true; + case NO_COMMENT: + if (TryConsume('\n')) { + // Completely blank line. + collector.Flush(); + collector.DetachFromPrev(); + } else { + bool result = Next(); + if (!result || + current_.text == "}" || + current_.text == "]" || + current_.text == ")") { + // It looks like we're at the end of a scope. In this case it + // makes no sense to attach a comment to the following token. + collector.Flush(); + } + return result; + } + break; + } + } +} + // ------------------------------------------------------------------- // Token-parsing helpers. Remember that these don't need to report // errors since any errors should already have been reported while @@ -623,17 +910,138 @@ double Tokenizer::ParseFloat(const string& text) { return result; } +// Helper to append a Unicode code point to a string as UTF8, without bringing +// in any external dependencies. +static void AppendUTF8(uint32 code_point, string* output) { + uint32 tmp = 0; + int len = 0; + if (code_point <= 0x7f) { + tmp = code_point; + len = 1; + } else if (code_point <= 0x07ff) { + tmp = 0x0000c080 | + ((code_point & 0x07c0) << 2) | + (code_point & 0x003f); + len = 2; + } else if (code_point <= 0xffff) { + tmp = 0x00e08080 | + ((code_point & 0xf000) << 4) | + ((code_point & 0x0fc0) << 2) | + (code_point & 0x003f); + len = 3; + } else if (code_point <= 0x1fffff) { + tmp = 0xf0808080 | + ((code_point & 0x1c0000) << 6) | + ((code_point & 0x03f000) << 4) | + ((code_point & 0x000fc0) << 2) | + (code_point & 0x003f); + len = 4; + } else { + // UTF-16 is only defined for code points up to 0x10FFFF, and UTF-8 is + // normally only defined up to there as well. + StringAppendF(output, "\\U%08x", code_point); + return; + } + tmp = ghtonl(tmp); + output->append(reinterpret_cast<const char*>(&tmp) + sizeof(tmp) - len, len); +} + +// Try to read <len> hex digits from ptr, and stuff the numeric result into +// *result. Returns true if that many digits were successfully consumed. +static bool ReadHexDigits(const char* ptr, int len, uint32* result) { + *result = 0; + if (len == 0) return false; + for (const char* end = ptr + len; ptr < end; ++ptr) { + if (*ptr == '\0') return false; + *result = (*result << 4) + DigitValue(*ptr); + } + return true; +} + +// Handling UTF-16 surrogate pairs. UTF-16 encodes code points in the range +// 0x10000...0x10ffff as a pair of numbers, a head surrogate followed by a trail +// surrogate. These numbers are in a reserved range of Unicode code points, so +// if we encounter such a pair we know how to parse it and convert it into a +// single code point. +static const uint32 kMinHeadSurrogate = 0xd800; +static const uint32 kMaxHeadSurrogate = 0xdc00; +static const uint32 kMinTrailSurrogate = 0xdc00; +static const uint32 kMaxTrailSurrogate = 0xe000; + +static inline bool IsHeadSurrogate(uint32 code_point) { + return (code_point >= kMinHeadSurrogate) && (code_point < kMaxHeadSurrogate); +} + +static inline bool IsTrailSurrogate(uint32 code_point) { + return (code_point >= kMinTrailSurrogate) && + (code_point < kMaxTrailSurrogate); +} + +// Combine a head and trail surrogate into a single Unicode code point. +static uint32 AssembleUTF16(uint32 head_surrogate, uint32 trail_surrogate) { + GOOGLE_DCHECK(IsHeadSurrogate(head_surrogate)); + GOOGLE_DCHECK(IsTrailSurrogate(trail_surrogate)); + return 0x10000 + (((head_surrogate - kMinHeadSurrogate) << 10) | + (trail_surrogate - kMinTrailSurrogate)); +} + +// Convert the escape sequence parameter to a number of expected hex digits. +static inline int UnicodeLength(char key) { + if (key == 'u') return 4; + if (key == 'U') return 8; + return 0; +} + +// Given a pointer to the 'u' or 'U' starting a Unicode escape sequence, attempt +// to parse that sequence. On success, returns a pointer to the first char +// beyond that sequence, and fills in *code_point. On failure, returns ptr +// itself. +static const char* FetchUnicodePoint(const char* ptr, uint32* code_point) { + const char* p = ptr; + // Fetch the code point. + const int len = UnicodeLength(*p++); + if (!ReadHexDigits(p, len, code_point)) + return ptr; + p += len; + + // Check if the code point we read is a "head surrogate." If so, then we + // expect it to be immediately followed by another code point which is a valid + // "trail surrogate," and together they form a UTF-16 pair which decodes into + // a single Unicode point. Trail surrogates may only use \u, not \U. + if (IsHeadSurrogate(*code_point) && *p == '\\' && *(p + 1) == 'u') { + uint32 trail_surrogate; + if (ReadHexDigits(p + 2, 4, &trail_surrogate) && + IsTrailSurrogate(trail_surrogate)) { + *code_point = AssembleUTF16(*code_point, trail_surrogate); + p += 6; + } + // If this failed, then we just emit the head surrogate as a code point. + // It's bogus, but so is the string. + } + + return p; +} + +// The text string must begin and end with single or double quote +// characters. void Tokenizer::ParseStringAppend(const string& text, string* output) { - // Reminder: text[0] is always the quote character. (If text is - // empty, it's invalid, so we'll just return.) - if (text.empty()) { + // Reminder: text[0] is always a quote character. (If text is + // empty, it's invalid, so we'll just return). + const size_t text_size = text.size(); + if (text_size == 0) { GOOGLE_LOG(DFATAL) << " Tokenizer::ParseStringAppend() passed text that could not" " have been tokenized as a string: " << CEscape(text); return; } - output->reserve(output->size() + text.size()); + // Reserve room for new string. The branch is necessary because if + // there is already space available the reserve() call might + // downsize the output. + const size_t new_len = text_size + output->size(); + if (new_len > output->capacity()) { + output->reserve(new_len); + } // Loop through the string copying characters to "output" and // interpreting escape sequences. Note that any invalid escape @@ -671,19 +1079,47 @@ void Tokenizer::ParseStringAppend(const string& text, string* output) { } output->push_back(static_cast<char>(code)); + } else if (*ptr == 'u' || *ptr == 'U') { + uint32 unicode; + const char* end = FetchUnicodePoint(ptr, &unicode); + if (end == ptr) { + // Failure: Just dump out what we saw, don't try to parse it. + output->push_back(*ptr); + } else { + AppendUTF8(unicode, output); + ptr = end - 1; // Because we're about to ++ptr. + } } else { // Some other escape code. output->push_back(TranslateEscape(*ptr)); } - } else if (*ptr == text[0]) { - // Ignore quote matching the starting quote. + } else if (*ptr == text[0] && ptr[1] == '\0') { + // Ignore final quote matching the starting quote. } else { output->push_back(*ptr); } } +} - return; +template<typename CharacterClass> +static bool AllInClass(const string& s) { + for (int i = 0; i < s.size(); ++i) { + if (!CharacterClass::InClass(s[i])) + return false; + } + return true; +} + +bool Tokenizer::IsIdentifier(const string& text) { + // Mirrors IDENTIFIER definition in Tokenizer::Next() above. + if (text.size() == 0) + return false; + if (!Letter::InClass(text.at(0))) + return false; + if (!AllInClass<Alphanumeric>(text.substr(1))) + return false; + return true; } } // namespace io diff --git a/src/google/protobuf/io/tokenizer.h b/src/google/protobuf/io/tokenizer.h index d115161..2f07116 100644 --- a/src/google/protobuf/io/tokenizer.h +++ b/src/google/protobuf/io/tokenizer.h @@ -38,6 +38,7 @@ #define GOOGLE_PROTOBUF_IO_TOKENIZER_H__ #include <string> +#include <vector> #include <google/protobuf/stubs/common.h> namespace google { @@ -66,7 +67,8 @@ class LIBPROTOBUF_EXPORT ErrorCollector { // Indicates that there was a warning in the input at the given line and // column numbers. The numbers are zero-based, so you may want to add // 1 to each before printing them. - virtual void AddWarning(int line, int column, const string& message) { } + virtual void AddWarning(int /* line */, int /* column */, + const string& /* message */) { } private: GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(ErrorCollector); @@ -122,16 +124,68 @@ class LIBPROTOBUF_EXPORT Tokenizer { // the token within the input stream. They are zero-based. int line; int column; + int end_column; }; // Get the current token. This is updated when Next() is called. Before // the first call to Next(), current() has type TYPE_START and no contents. const Token& current(); + // Return the previous token -- i.e. what current() returned before the + // previous call to Next(). + const Token& previous(); + // Advance to the next token. Returns false if the end of the input is // reached. bool Next(); + // Like Next(), but also collects comments which appear between the previous + // and next tokens. + // + // Comments which appear to be attached to the previous token are stored + // in *prev_tailing_comments. Comments which appear to be attached to the + // next token are stored in *next_leading_comments. Comments appearing in + // between which do not appear to be attached to either will be added to + // detached_comments. Any of these parameters can be NULL to simply discard + // the comments. + // + // A series of line comments appearing on consecutive lines, with no other + // tokens appearing on those lines, will be treated as a single comment. + // + // Only the comment content is returned; comment markers (e.g. //) are + // stripped out. For block comments, leading whitespace and an asterisk will + // be stripped from the beginning of each line other than the first. Newlines + // are included in the output. + // + // Examples: + // + // optional int32 foo = 1; // Comment attached to foo. + // // Comment attached to bar. + // optional int32 bar = 2; + // + // optional string baz = 3; + // // Comment attached to baz. + // // Another line attached to baz. + // + // // Comment attached to qux. + // // + // // Another line attached to qux. + // optional double qux = 4; + // + // // Detached comment. This is not attached to qux or corge + // // because there are blank lines separating it from both. + // + // optional string corge = 5; + // /* Block comment attached + // * to corge. Leading asterisks + // * will be removed. */ + // /* Block comment attached to + // * grault. */ + // optional int32 grault = 6; + bool NextWithComments(string* prev_trailing_comments, + vector<string>* detached_comments, + string* next_leading_comments); + // Parse helpers --------------------------------------------------- // Parses a TYPE_FLOAT token. This never fails, so long as the text actually @@ -175,11 +229,27 @@ class LIBPROTOBUF_EXPORT Tokenizer { // Sets the comment style. void set_comment_style(CommentStyle style) { comment_style_ = style; } + // Whether to require whitespace between a number and a field name. + // Default is true. Do not use this; for Google-internal cleanup only. + void set_require_space_after_number(bool require) { + require_space_after_number_ = require; + } + + // Whether to allow string literals to span multiple lines. Default is false. + // Do not use this; for Google-internal cleanup only. + void set_allow_multiline_strings(bool allow) { + allow_multiline_strings_ = allow; + } + + // External helper: validate an identifier. + static bool IsIdentifier(const string& text); + // ----------------------------------------------------------------- private: GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(Tokenizer); Token current_; // Returned by current(). + Token previous_; // Returned by previous(). ZeroCopyInputStream* input_; ErrorCollector* error_collector_; @@ -194,15 +264,18 @@ class LIBPROTOBUF_EXPORT Tokenizer { int line_; int column_; - // Position in buffer_ where StartToken() was called. If the token - // started in the previous buffer, this is zero, and current_.text already - // contains the part of the token from the previous buffer. If not - // currently parsing a token, this is -1. - int token_start_; + // String to which text should be appended as we advance through it. + // Call RecordTo(&str) to start recording and StopRecording() to stop. + // E.g. StartToken() calls RecordTo(¤t_.text). record_start_ is the + // position within the current buffer where recording started. + string* record_target_; + int record_start_; // Options. bool allow_f_after_float_; CommentStyle comment_style_; + bool require_space_after_number_; + bool allow_multiline_strings_; // Since we count columns we need to interpret tabs somehow. We'll take // the standard 8-character definition for lack of any way to do better. @@ -217,6 +290,9 @@ class LIBPROTOBUF_EXPORT Tokenizer { // Read a new buffer from the input. void Refresh(); + inline void RecordTo(string* target); + inline void StopRecording(); + // Called when the current character is the first character of a new // token (not including whitespace or comments). inline void StartToken(); @@ -249,9 +325,28 @@ class LIBPROTOBUF_EXPORT Tokenizer { TokenType ConsumeNumber(bool started_with_zero, bool started_with_dot); // Consume the rest of a line. - void ConsumeLineComment(); + void ConsumeLineComment(string* content); // Consume until "*/". - void ConsumeBlockComment(); + void ConsumeBlockComment(string* content); + + enum NextCommentStatus { + // Started a line comment. + LINE_COMMENT, + + // Started a block comment. + BLOCK_COMMENT, + + // Consumed a slash, then realized it wasn't a comment. current_ has + // been filled in with a slash token. The caller should return it. + SLASH_NOT_COMMENT, + + // We do not appear to be starting a comment here. + NO_COMMENT + }; + + // If we're at the start of a new comment, consume it and return what kind + // of comment it is. + NextCommentStatus TryConsumeCommentStart(); // ----------------------------------------------------------------- // These helper methods make the parsing code more readable. The @@ -291,6 +386,10 @@ inline const Tokenizer::Token& Tokenizer::current() { return current_; } +inline const Tokenizer::Token& Tokenizer::previous() { + return previous_; +} + inline void Tokenizer::ParseString(const string& text, string* output) { output->clear(); ParseStringAppend(text, output); diff --git a/src/google/protobuf/io/tokenizer_unittest.cc b/src/google/protobuf/io/tokenizer_unittest.cc index 358ec56..b39279b 100644 --- a/src/google/protobuf/io/tokenizer_unittest.cc +++ b/src/google/protobuf/io/tokenizer_unittest.cc @@ -32,9 +32,10 @@ // Based on original Protocol Buffers design by // Sanjay Ghemawat, Jeff Dean, and others. -#include <vector> -#include <math.h> #include <limits.h> +#include <math.h> + +#include <vector> #include <google/protobuf/io/tokenizer.h> #include <google/protobuf/io/zero_copy_stream_impl.h> @@ -257,6 +258,7 @@ TEST_2D(TokenizerTest, SimpleTokens, kSimpleTokenCases, kBlockSizes) { EXPECT_EQ("", tokenizer.current().text); EXPECT_EQ(0, tokenizer.current().line); EXPECT_EQ(0, tokenizer.current().column); + EXPECT_EQ(0, tokenizer.current().end_column); // Parse the token. ASSERT_TRUE(tokenizer.Next()); @@ -268,6 +270,8 @@ TEST_2D(TokenizerTest, SimpleTokens, kSimpleTokenCases, kBlockSizes) { // Check that it is located at the beginning of the input EXPECT_EQ(0, tokenizer.current().line); EXPECT_EQ(0, tokenizer.current().column); + EXPECT_EQ(kSimpleTokenCases_case.input.size(), + tokenizer.current().end_column); // There should be no more input. EXPECT_FALSE(tokenizer.Next()); @@ -277,6 +281,8 @@ TEST_2D(TokenizerTest, SimpleTokens, kSimpleTokenCases, kBlockSizes) { EXPECT_EQ("", tokenizer.current().text); EXPECT_EQ(0, tokenizer.current().line); EXPECT_EQ(kSimpleTokenCases_case.input.size(), tokenizer.current().column); + EXPECT_EQ(kSimpleTokenCases_case.input.size(), + tokenizer.current().end_column); // There should be no errors. EXPECT_TRUE(error_collector.text_.empty()); @@ -339,76 +345,77 @@ MultiTokenCase kMultiTokenCases[] = { // Test all token types at the same time. { "foo 1 1.2 + 'bar'", { - { Tokenizer::TYPE_IDENTIFIER, "foo" , 0, 0 }, - { Tokenizer::TYPE_INTEGER , "1" , 0, 4 }, - { Tokenizer::TYPE_FLOAT , "1.2" , 0, 6 }, - { Tokenizer::TYPE_SYMBOL , "+" , 0, 10 }, - { Tokenizer::TYPE_STRING , "'bar'", 0, 12 }, - { Tokenizer::TYPE_END , "" , 0, 17 }, + { Tokenizer::TYPE_IDENTIFIER, "foo" , 0, 0, 3 }, + { Tokenizer::TYPE_INTEGER , "1" , 0, 4, 5 }, + { Tokenizer::TYPE_FLOAT , "1.2" , 0, 6, 9 }, + { Tokenizer::TYPE_SYMBOL , "+" , 0, 10, 11 }, + { Tokenizer::TYPE_STRING , "'bar'", 0, 12, 17 }, + { Tokenizer::TYPE_END , "" , 0, 17, 17 }, }}, // Test that consecutive symbols are parsed as separate tokens. { "!@+%", { - { Tokenizer::TYPE_SYMBOL , "!" , 0, 0 }, - { Tokenizer::TYPE_SYMBOL , "@" , 0, 1 }, - { Tokenizer::TYPE_SYMBOL , "+" , 0, 2 }, - { Tokenizer::TYPE_SYMBOL , "%" , 0, 3 }, - { Tokenizer::TYPE_END , "" , 0, 4 }, + { Tokenizer::TYPE_SYMBOL , "!" , 0, 0, 1 }, + { Tokenizer::TYPE_SYMBOL , "@" , 0, 1, 2 }, + { Tokenizer::TYPE_SYMBOL , "+" , 0, 2, 3 }, + { Tokenizer::TYPE_SYMBOL , "%" , 0, 3, 4 }, + { Tokenizer::TYPE_END , "" , 0, 4, 4 }, }}, // Test that newlines affect line numbers correctly. { "foo bar\nrab oof", { - { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0 }, - { Tokenizer::TYPE_IDENTIFIER, "bar", 0, 4 }, - { Tokenizer::TYPE_IDENTIFIER, "rab", 1, 0 }, - { Tokenizer::TYPE_IDENTIFIER, "oof", 1, 4 }, - { Tokenizer::TYPE_END , "" , 1, 7 }, + { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0, 3 }, + { Tokenizer::TYPE_IDENTIFIER, "bar", 0, 4, 7 }, + { Tokenizer::TYPE_IDENTIFIER, "rab", 1, 0, 3 }, + { Tokenizer::TYPE_IDENTIFIER, "oof", 1, 4, 7 }, + { Tokenizer::TYPE_END , "" , 1, 7, 7 }, }}, // Test that tabs affect column numbers correctly. { "foo\tbar \tbaz", { - { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0 }, - { Tokenizer::TYPE_IDENTIFIER, "bar", 0, 8 }, - { Tokenizer::TYPE_IDENTIFIER, "baz", 0, 16 }, - { Tokenizer::TYPE_END , "" , 0, 19 }, + { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0, 3 }, + { Tokenizer::TYPE_IDENTIFIER, "bar", 0, 8, 11 }, + { Tokenizer::TYPE_IDENTIFIER, "baz", 0, 16, 19 }, + { Tokenizer::TYPE_END , "" , 0, 19, 19 }, + }}, + + // Test that tabs in string literals affect column numbers correctly. + { "\"foo\tbar\" baz", { + { Tokenizer::TYPE_STRING , "\"foo\tbar\"", 0, 0, 12 }, + { Tokenizer::TYPE_IDENTIFIER, "baz" , 0, 13, 16 }, + { Tokenizer::TYPE_END , "" , 0, 16, 16 }, }}, // Test that line comments are ignored. { "foo // This is a comment\n" "bar // This is another comment", { - { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0 }, - { Tokenizer::TYPE_IDENTIFIER, "bar", 1, 0 }, - { Tokenizer::TYPE_END , "" , 1, 30 }, + { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0, 3 }, + { Tokenizer::TYPE_IDENTIFIER, "bar", 1, 0, 3 }, + { Tokenizer::TYPE_END , "" , 1, 30, 30 }, }}, // Test that block comments are ignored. { "foo /* This is a block comment */ bar", { - { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0 }, - { Tokenizer::TYPE_IDENTIFIER, "bar", 0, 34 }, - { Tokenizer::TYPE_END , "" , 0, 37 }, + { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0, 3 }, + { Tokenizer::TYPE_IDENTIFIER, "bar", 0, 34, 37 }, + { Tokenizer::TYPE_END , "" , 0, 37, 37 }, }}, // Test that sh-style comments are not ignored by default. { "foo # bar\n" "baz", { - { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0 }, - { Tokenizer::TYPE_SYMBOL , "#" , 0, 4 }, - { Tokenizer::TYPE_IDENTIFIER, "bar", 0, 6 }, - { Tokenizer::TYPE_IDENTIFIER, "baz", 1, 0 }, - { Tokenizer::TYPE_END , "" , 1, 3 }, - }}, - - // Bytes with the high-order bit set should not be seen as control characters. - { "\300", { - { Tokenizer::TYPE_SYMBOL, "\300", 0, 0 }, - { Tokenizer::TYPE_END , "" , 0, 1 }, + { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0, 3 }, + { Tokenizer::TYPE_SYMBOL , "#" , 0, 4, 5 }, + { Tokenizer::TYPE_IDENTIFIER, "bar", 0, 6, 9 }, + { Tokenizer::TYPE_IDENTIFIER, "baz", 1, 0, 3 }, + { Tokenizer::TYPE_END , "" , 1, 3, 3 }, }}, // Test all whitespace chars { "foo\n\t\r\v\fbar", { - { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0 }, - { Tokenizer::TYPE_IDENTIFIER, "bar", 1, 11 }, - { Tokenizer::TYPE_END , "" , 1, 14 }, + { Tokenizer::TYPE_IDENTIFIER, "foo", 0, 0, 3 }, + { Tokenizer::TYPE_IDENTIFIER, "bar", 1, 11, 14 }, + { Tokenizer::TYPE_END , "" , 1, 14, 14 }, }}, }; @@ -425,6 +432,7 @@ TEST_2D(TokenizerTest, MultipleTokens, kMultiTokenCases, kBlockSizes) { EXPECT_EQ("", tokenizer.current().text); EXPECT_EQ(0, tokenizer.current().line); EXPECT_EQ(0, tokenizer.current().column); + EXPECT_EQ(0, tokenizer.current().end_column); // Loop through all expected tokens. int i = 0; @@ -434,6 +442,8 @@ TEST_2D(TokenizerTest, MultipleTokens, kMultiTokenCases, kBlockSizes) { SCOPED_TRACE(testing::Message() << "Token #" << i << ": " << token.text); + Tokenizer::Token previous = tokenizer.current(); + // Next() should only return false when it hits the end token. if (token.type != Tokenizer::TYPE_END) { ASSERT_TRUE(tokenizer.Next()); @@ -441,11 +451,19 @@ TEST_2D(TokenizerTest, MultipleTokens, kMultiTokenCases, kBlockSizes) { ASSERT_FALSE(tokenizer.Next()); } + // Check that the previous token is set correctly. + EXPECT_EQ(previous.type, tokenizer.previous().type); + EXPECT_EQ(previous.text, tokenizer.previous().text); + EXPECT_EQ(previous.line, tokenizer.previous().line); + EXPECT_EQ(previous.column, tokenizer.previous().column); + EXPECT_EQ(previous.end_column, tokenizer.previous().end_column); + // Check that the token matches the expected one. EXPECT_EQ(token.type, tokenizer.current().type); EXPECT_EQ(token.text, tokenizer.current().text); EXPECT_EQ(token.line, tokenizer.current().line); EXPECT_EQ(token.column, tokenizer.current().column); + EXPECT_EQ(token.end_column, tokenizer.current().end_column); } while (token.type != Tokenizer::TYPE_END); @@ -491,6 +509,217 @@ TEST_1D(TokenizerTest, ShCommentStyle, kBlockSizes) { // ------------------------------------------------------------------- +// In each case, the input is expected to have two tokens named "prev" and +// "next" with comments in between. +struct DocCommentCase { + string input; + + const char* prev_trailing_comments; + const char* detached_comments[10]; + const char* next_leading_comments; +}; + +inline ostream& operator<<(ostream& out, + const DocCommentCase& test_case) { + return out << CEscape(test_case.input); +} + +DocCommentCase kDocCommentCases[] = { + { + "prev next", + + "", + {}, + "" + }, + + { + "prev /* ignored */ next", + + "", + {}, + "" + }, + + { + "prev // trailing comment\n" + "next", + + " trailing comment\n", + {}, + "" + }, + + { + "prev\n" + "// leading comment\n" + "// line 2\n" + "next", + + "", + {}, + " leading comment\n" + " line 2\n" + }, + + { + "prev\n" + "// trailing comment\n" + "// line 2\n" + "\n" + "next", + + " trailing comment\n" + " line 2\n", + {}, + "" + }, + + { + "prev // trailing comment\n" + "// leading comment\n" + "// line 2\n" + "next", + + " trailing comment\n", + {}, + " leading comment\n" + " line 2\n" + }, + + { + "prev /* trailing block comment */\n" + "/* leading block comment\n" + " * line 2\n" + " * line 3 */" + "next", + + " trailing block comment ", + {}, + " leading block comment\n" + " line 2\n" + " line 3 " + }, + + { + "prev\n" + "/* trailing block comment\n" + " * line 2\n" + " * line 3\n" + " */\n" + "/* leading block comment\n" + " * line 2\n" + " * line 3 */" + "next", + + " trailing block comment\n" + " line 2\n" + " line 3\n", + {}, + " leading block comment\n" + " line 2\n" + " line 3 " + }, + + { + "prev\n" + "// trailing comment\n" + "\n" + "// detached comment\n" + "// line 2\n" + "\n" + "// second detached comment\n" + "/* third detached comment\n" + " * line 2 */\n" + "// leading comment\n" + "next", + + " trailing comment\n", + { + " detached comment\n" + " line 2\n", + " second detached comment\n", + " third detached comment\n" + " line 2 " + }, + " leading comment\n" + }, + + { + "prev /**/\n" + "\n" + "// detached comment\n" + "\n" + "// leading comment\n" + "next", + + "", + { + " detached comment\n" + }, + " leading comment\n" + }, + + { + "prev /**/\n" + "// leading comment\n" + "next", + + "", + {}, + " leading comment\n" + }, + }; + +TEST_2D(TokenizerTest, DocComments, kDocCommentCases, kBlockSizes) { + // Set up the tokenizer. + TestInputStream input(kDocCommentCases_case.input.data(), + kDocCommentCases_case.input.size(), + kBlockSizes_case); + TestErrorCollector error_collector; + Tokenizer tokenizer(&input, &error_collector); + + // Set up a second tokenizer where we'll pass all NULLs to NextWithComments(). + TestInputStream input2(kDocCommentCases_case.input.data(), + kDocCommentCases_case.input.size(), + kBlockSizes_case); + Tokenizer tokenizer2(&input2, &error_collector); + + tokenizer.Next(); + tokenizer2.Next(); + + EXPECT_EQ("prev", tokenizer.current().text); + EXPECT_EQ("prev", tokenizer2.current().text); + + string prev_trailing_comments; + vector<string> detached_comments; + string next_leading_comments; + tokenizer.NextWithComments(&prev_trailing_comments, &detached_comments, + &next_leading_comments); + tokenizer2.NextWithComments(NULL, NULL, NULL); + EXPECT_EQ("next", tokenizer.current().text); + EXPECT_EQ("next", tokenizer2.current().text); + + EXPECT_EQ(kDocCommentCases_case.prev_trailing_comments, + prev_trailing_comments); + + for (int i = 0; i < detached_comments.size(); i++) { + ASSERT_LT(i, GOOGLE_ARRAYSIZE(kDocCommentCases)); + ASSERT_TRUE(kDocCommentCases_case.detached_comments[i] != NULL); + EXPECT_EQ(kDocCommentCases_case.detached_comments[i], + detached_comments[i]); + } + + // Verify that we matched all the detached comments. + EXPECT_EQ(NULL, + kDocCommentCases_case.detached_comments[detached_comments.size()]); + + EXPECT_EQ(kDocCommentCases_case.next_leading_comments, + next_leading_comments); +} + +// ------------------------------------------------------------------- + // Test parse helpers. It's not really worth setting up a full data-driven // test here. TEST_F(TokenizerTest, ParseInteger) { @@ -506,7 +735,7 @@ TEST_F(TokenizerTest, ParseInteger) { EXPECT_EQ(0, ParseInteger("0x")); uint64 i; -#ifdef GTEST_HAS_DEATH_TEST // death tests do not work on Windows yet +#ifdef PROTOBUF_HAS_DEATH_TEST // death tests do not work on Windows yet // Test invalid integers that will never be tokenized as integers. EXPECT_DEBUG_DEATH(Tokenizer::ParseInteger("zxy", kuint64max, &i), "passed text that could not have been tokenized as an integer"); @@ -518,7 +747,7 @@ TEST_F(TokenizerTest, ParseInteger) { "passed text that could not have been tokenized as an integer"); EXPECT_DEBUG_DEATH(Tokenizer::ParseInteger("-1", kuint64max, &i), "passed text that could not have been tokenized as an integer"); -#endif // GTEST_HAS_DEATH_TEST +#endif // PROTOBUF_HAS_DEATH_TEST // Test overflows. EXPECT_TRUE (Tokenizer::ParseInteger("0", 0, &i)); @@ -561,7 +790,7 @@ TEST_F(TokenizerTest, ParseFloat) { EXPECT_EQ( 0.0, Tokenizer::ParseFloat("1e-9999999999999999999999999999")); EXPECT_EQ(HUGE_VAL, Tokenizer::ParseFloat("1e+9999999999999999999999999999")); -#ifdef GTEST_HAS_DEATH_TEST // death tests do not work on Windows yet +#ifdef PROTOBUF_HAS_DEATH_TEST // death tests do not work on Windows yet // Test invalid integers that will never be tokenized as integers. EXPECT_DEBUG_DEATH(Tokenizer::ParseFloat("zxy"), "passed text that could not have been tokenized as a float"); @@ -569,7 +798,7 @@ TEST_F(TokenizerTest, ParseFloat) { "passed text that could not have been tokenized as a float"); EXPECT_DEBUG_DEATH(Tokenizer::ParseFloat("-1.0"), "passed text that could not have been tokenized as a float"); -#endif // GTEST_HAS_DEATH_TEST +#endif // PROTOBUF_HAS_DEATH_TEST } TEST_F(TokenizerTest, ParseString) { @@ -591,11 +820,27 @@ TEST_F(TokenizerTest, ParseString) { Tokenizer::ParseString("'\\", &output); EXPECT_EQ("\\", output); + // Experiment with Unicode escapes. Here are one-, two- and three-byte Unicode + // characters. + Tokenizer::ParseString("'\\u0024\\u00a2\\u20ac\\U00024b62XX'", &output); + EXPECT_EQ("$¢€𤭢XX", output); + // Same thing encoded using UTF16. + Tokenizer::ParseString("'\\u0024\\u00a2\\u20ac\\ud852\\udf62XX'", &output); + EXPECT_EQ("$¢€𤭢XX", output); + // Here's some broken UTF16; there's a head surrogate with no tail surrogate. + // We just output this as if it were UTF8; it's not a defined code point, but + // it has a defined encoding. + Tokenizer::ParseString("'\\ud852XX'", &output); + EXPECT_EQ("\xed\xa1\x92XX", output); + // Malformed escape: Demons may fly out of the nose. + Tokenizer::ParseString("\\u0", &output); + EXPECT_EQ("u0", output); + // Test invalid strings that will never be tokenized as strings. -#ifdef GTEST_HAS_DEATH_TEST // death tests do not work on Windows yet +#ifdef PROTOBUF_HAS_DEATH_TEST // death tests do not work on Windows yet EXPECT_DEBUG_DEATH(Tokenizer::ParseString("", &output), "passed text that could not have been tokenized as a string"); -#endif // GTEST_HAS_DEATH_TEST +#endif // PROTOBUF_HAS_DEATH_TEST } TEST_F(TokenizerTest, ParseStringAppend) { @@ -632,9 +877,15 @@ ErrorCase kErrorCases[] = { { "'\\x' foo", true, "0:3: Expected hex digits for escape sequence.\n" }, { "'foo", false, - "0:4: String literals cannot cross line boundaries.\n" }, + "0:4: Unexpected end of string.\n" }, { "'bar\nfoo", true, "0:4: String literals cannot cross line boundaries.\n" }, + { "'\\u01' foo", true, + "0:5: Expected four hex digits for \\u escape sequence.\n" }, + { "'\\u01' foo", true, + "0:5: Expected four hex digits for \\u escape sequence.\n" }, + { "'\\uXYZ' foo", true, + "0:3: Expected four hex digits for \\u escape sequence.\n" }, // Integer errors. { "123foo", true, @@ -694,6 +945,10 @@ ErrorCase kErrorCases[] = { "0:0: Invalid control characters encountered in text.\n" }, { string("\0\0foo", 5), true, "0:0: Invalid control characters encountered in text.\n" }, + + // Check error from high order bits set + { "\300foo", true, + "0:0: Interpreting non ascii codepoint 192.\n" }, }; TEST_2D(TokenizerTest, Errors, kErrorCases, kBlockSizes) { @@ -711,7 +966,7 @@ TEST_2D(TokenizerTest, Errors, kErrorCases, kBlockSizes) { } // Check that the errors match what was expected. - EXPECT_EQ(error_collector.text_, kErrorCases_case.errors); + EXPECT_EQ(kErrorCases_case.errors, error_collector.text_); // If the error was recoverable, make sure we saw "foo" after it. if (kErrorCases_case.recoverable) { @@ -737,6 +992,7 @@ TEST_1D(TokenizerTest, BackUpOnDestruction, kBlockSizes) { EXPECT_EQ(strlen("foo"), input.ByteCount()); } + } // namespace } // namespace io } // namespace protobuf diff --git a/src/google/protobuf/io/zero_copy_stream.cc b/src/google/protobuf/io/zero_copy_stream.cc index dad6ff1..4d53f29 100644 --- a/src/google/protobuf/io/zero_copy_stream.cc +++ b/src/google/protobuf/io/zero_copy_stream.cc @@ -34,6 +34,7 @@ #include <google/protobuf/io/zero_copy_stream.h> +#include <google/protobuf/stubs/common.h> namespace google { namespace protobuf { @@ -43,6 +44,14 @@ ZeroCopyInputStream::~ZeroCopyInputStream() {} ZeroCopyOutputStream::~ZeroCopyOutputStream() {} +bool ZeroCopyOutputStream::WriteAliasedRaw(const void* /* data */, + int /* size */) { + GOOGLE_LOG(FATAL) << "This ZeroCopyOutputStream doesn't support aliasing. " + "Reaching here usually means a ZeroCopyOutputStream " + "implementation bug."; + return false; +} + } // namespace io } // namespace protobuf } // namespace google diff --git a/src/google/protobuf/io/zero_copy_stream.h b/src/google/protobuf/io/zero_copy_stream.h index db5326f..f892122 100644 --- a/src/google/protobuf/io/zero_copy_stream.h +++ b/src/google/protobuf/io/zero_copy_stream.h @@ -226,6 +226,16 @@ class LIBPROTOBUF_EXPORT ZeroCopyOutputStream { // Returns the total number of bytes written since this object was created. virtual int64 ByteCount() const = 0; + // Write a given chunk of data to the output. Some output streams may + // implement this in a way that avoids copying. Check AllowsAliasing() before + // calling WriteAliasedRaw(). It will GOOGLE_CHECK fail if WriteAliasedRaw() is + // called on a stream that does not allow aliasing. + // + // NOTE: It is caller's responsibility to ensure that the chunk of memory + // remains live until all of the data has been consumed from the stream. + virtual bool WriteAliasedRaw(const void* data, int size); + virtual bool AllowsAliasing() const { return false; } + private: GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(ZeroCopyOutputStream); diff --git a/src/google/protobuf/io/zero_copy_stream_impl.cc b/src/google/protobuf/io/zero_copy_stream_impl.cc index 1384c74..7829a29 100644 --- a/src/google/protobuf/io/zero_copy_stream_impl.cc +++ b/src/google/protobuf/io/zero_copy_stream_impl.cc @@ -46,7 +46,8 @@ #include <google/protobuf/io/zero_copy_stream_impl.h> #include <google/protobuf/stubs/common.h> -#include <google/protobuf/stubs/stl_util-inl.h> +#include <google/protobuf/stubs/stl_util.h> + namespace google { namespace protobuf { @@ -412,7 +413,9 @@ int64 ConcatenatingInputStream::ByteCount() const { LimitingInputStream::LimitingInputStream(ZeroCopyInputStream* input, int64 limit) - : input_(input), limit_(limit) {} + : input_(input), limit_(limit) { + prior_bytes_read_ = input_->ByteCount(); +} LimitingInputStream::~LimitingInputStream() { // If we overshot the limit, back up. @@ -456,9 +459,9 @@ bool LimitingInputStream::Skip(int count) { int64 LimitingInputStream::ByteCount() const { if (limit_ < 0) { - return input_->ByteCount() + limit_; + return input_->ByteCount() + limit_ - prior_bytes_read_; } else { - return input_->ByteCount(); + return input_->ByteCount() - prior_bytes_read_; } } diff --git a/src/google/protobuf/io/zero_copy_stream_impl.h b/src/google/protobuf/io/zero_copy_stream_impl.h index 9fedb00..8382709 100644 --- a/src/google/protobuf/io/zero_copy_stream_impl.h +++ b/src/google/protobuf/io/zero_copy_stream_impl.h @@ -344,6 +344,7 @@ class LIBPROTOBUF_EXPORT LimitingInputStream : public ZeroCopyInputStream { private: ZeroCopyInputStream* input_; int64 limit_; // Decreases as we go, becomes negative if we overshoot. + int64 prior_bytes_read_; // Bytes read on underlying stream at construction GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(LimitingInputStream); }; diff --git a/src/google/protobuf/io/zero_copy_stream_impl_lite.cc b/src/google/protobuf/io/zero_copy_stream_impl_lite.cc index e801251..d186a98 100644 --- a/src/google/protobuf/io/zero_copy_stream_impl_lite.cc +++ b/src/google/protobuf/io/zero_copy_stream_impl_lite.cc @@ -32,9 +32,13 @@ // Based on original Protocol Buffers design by // Sanjay Ghemawat, Jeff Dean, and others. -#include <google/protobuf/io/zero_copy_stream_impl.h> +#include <google/protobuf/io/zero_copy_stream_impl_lite.h> + +#include <algorithm> +#include <limits> + #include <google/protobuf/stubs/common.h> -#include <google/protobuf/stubs/stl_util-inl.h> +#include <google/protobuf/stubs/stl_util.h> namespace google { namespace protobuf { @@ -159,15 +163,23 @@ bool StringOutputStream::Next(void** data, int* size) { // without a memory allocation this way. STLStringResizeUninitialized(target_, target_->capacity()); } else { - // Size has reached capacity, so double the size. Also make sure - // that the new size is at least kMinimumSize. + // Size has reached capacity, try to double the size. + if (old_size > std::numeric_limits<int>::max() / 2) { + // Can not double the size otherwise it is going to cause integer + // overflow in the expression below: old_size * 2 "; + GOOGLE_LOG(ERROR) << "Cannot allocate buffer larger than kint32max for " + << "StringOutputStream."; + return false; + } + // Double the size, also make sure that the new size is at least + // kMinimumSize. STLStringResizeUninitialized( target_, max(old_size * 2, kMinimumSize + 0)); // "+ 0" works around GCC4 weirdness. } - *data = string_as_array(target_) + old_size; + *data = mutable_string_data(target_) + old_size; *size = target_->size() - old_size; return true; } diff --git a/src/google/protobuf/io/zero_copy_stream_impl_lite.h b/src/google/protobuf/io/zero_copy_stream_impl_lite.h index 153f543..b980143 100644 --- a/src/google/protobuf/io/zero_copy_stream_impl_lite.h +++ b/src/google/protobuf/io/zero_copy_stream_impl_lite.h @@ -48,6 +48,7 @@ #include <iosfwd> #include <google/protobuf/io/zero_copy_stream.h> #include <google/protobuf/stubs/common.h> +#include <google/protobuf/stubs/stl_util.h> namespace google { @@ -333,6 +334,19 @@ class LIBPROTOBUF_EXPORT CopyingOutputStreamAdaptor : public ZeroCopyOutputStrea // =================================================================== +// Return a pointer to mutable characters underlying the given string. The +// return value is valid until the next time the string is resized. We +// trust the caller to treat the return value as an array of length s->size(). +inline char* mutable_string_data(string* s) { +#ifdef LANG_CXX11 + // This should be simpler & faster than string_as_array() because the latter + // is guaranteed to return NULL when *s is empty, so it has to check for that. + return &(*s)[0]; +#else + return string_as_array(s); +#endif +} + } // namespace io } // namespace protobuf diff --git a/src/google/protobuf/io/zero_copy_stream_unittest.cc b/src/google/protobuf/io/zero_copy_stream_unittest.cc index 8229ee6..75eb2a4 100644 --- a/src/google/protobuf/io/zero_copy_stream_unittest.cc +++ b/src/google/protobuf/io/zero_copy_stream_unittest.cc @@ -61,6 +61,7 @@ #include <sstream> #include <google/protobuf/io/zero_copy_stream_impl.h> +#include <google/protobuf/io/coded_stream.h> #if HAVE_ZLIB #include <google/protobuf/io/gzip_stream.h> @@ -285,6 +286,57 @@ TEST_F(IoTest, ArrayIo) { } } +TEST_F(IoTest, TwoSessionWrite) { + // Test that two concatenated write sessions read correctly + + static const char* strA = "0123456789"; + static const char* strB = "WhirledPeas"; + const int kBufferSize = 2*1024; + uint8* buffer = new uint8[kBufferSize]; + char* temp_buffer = new char[40]; + + for (int i = 0; i < kBlockSizeCount; i++) { + for (int j = 0; j < kBlockSizeCount; j++) { + ArrayOutputStream* output = + new ArrayOutputStream(buffer, kBufferSize, kBlockSizes[i]); + CodedOutputStream* coded_output = new CodedOutputStream(output); + coded_output->WriteVarint32(strlen(strA)); + coded_output->WriteRaw(strA, strlen(strA)); + delete coded_output; // flush + int64 pos = output->ByteCount(); + delete output; + output = new ArrayOutputStream( + buffer + pos, kBufferSize - pos, kBlockSizes[i]); + coded_output = new CodedOutputStream(output); + coded_output->WriteVarint32(strlen(strB)); + coded_output->WriteRaw(strB, strlen(strB)); + delete coded_output; // flush + int64 size = pos + output->ByteCount(); + delete output; + + ArrayInputStream* input = + new ArrayInputStream(buffer, size, kBlockSizes[j]); + CodedInputStream* coded_input = new CodedInputStream(input); + uint32 insize; + EXPECT_TRUE(coded_input->ReadVarint32(&insize)); + EXPECT_EQ(strlen(strA), insize); + EXPECT_TRUE(coded_input->ReadRaw(temp_buffer, insize)); + EXPECT_EQ(0, memcmp(temp_buffer, strA, insize)); + + EXPECT_TRUE(coded_input->ReadVarint32(&insize)); + EXPECT_EQ(strlen(strB), insize); + EXPECT_TRUE(coded_input->ReadRaw(temp_buffer, insize)); + EXPECT_EQ(0, memcmp(temp_buffer, strB, insize)); + + delete coded_input; + delete input; + } + } + + delete [] temp_buffer; + delete [] buffer; +} + #if HAVE_ZLIB TEST_F(IoTest, GzipIo) { const int kBufferSize = 2*1024; @@ -296,9 +348,49 @@ TEST_F(IoTest, GzipIo) { int size; { ArrayOutputStream output(buffer, kBufferSize, kBlockSizes[i]); - GzipOutputStream gzout( - &output, GzipOutputStream::GZIP, gzip_buffer_size); + GzipOutputStream::Options options; + options.format = GzipOutputStream::GZIP; + if (gzip_buffer_size != -1) { + options.buffer_size = gzip_buffer_size; + } + GzipOutputStream gzout(&output, options); + WriteStuff(&gzout); + gzout.Close(); + size = output.ByteCount(); + } + { + ArrayInputStream input(buffer, size, kBlockSizes[j]); + GzipInputStream gzin( + &input, GzipInputStream::GZIP, gzip_buffer_size); + ReadStuff(&gzin); + } + } + } + } + delete [] buffer; +} + +TEST_F(IoTest, GzipIoWithFlush) { + const int kBufferSize = 2*1024; + uint8* buffer = new uint8[kBufferSize]; + // We start with i = 4 as we want a block size > 6. With block size <= 6 + // Flush() fills up the entire 2K buffer with flush markers and the test + // fails. See documentation for Flush() for more detail. + for (int i = 4; i < kBlockSizeCount; i++) { + for (int j = 0; j < kBlockSizeCount; j++) { + for (int z = 0; z < kBlockSizeCount; z++) { + int gzip_buffer_size = kBlockSizes[z]; + int size; + { + ArrayOutputStream output(buffer, kBufferSize, kBlockSizes[i]); + GzipOutputStream::Options options; + options.format = GzipOutputStream::GZIP; + if (gzip_buffer_size != -1) { + options.buffer_size = gzip_buffer_size; + } + GzipOutputStream gzout(&output, options); WriteStuff(&gzout); + EXPECT_TRUE(gzout.Flush()); gzout.Close(); size = output.ByteCount(); } @@ -314,6 +406,64 @@ TEST_F(IoTest, GzipIo) { delete [] buffer; } +TEST_F(IoTest, GzipIoContiguousFlushes) { + const int kBufferSize = 2*1024; + uint8* buffer = new uint8[kBufferSize]; + + int block_size = kBlockSizes[4]; + int gzip_buffer_size = block_size; + int size; + + ArrayOutputStream output(buffer, kBufferSize, block_size); + GzipOutputStream::Options options; + options.format = GzipOutputStream::GZIP; + if (gzip_buffer_size != -1) { + options.buffer_size = gzip_buffer_size; + } + GzipOutputStream gzout(&output, options); + WriteStuff(&gzout); + EXPECT_TRUE(gzout.Flush()); + EXPECT_TRUE(gzout.Flush()); + gzout.Close(); + size = output.ByteCount(); + + ArrayInputStream input(buffer, size, block_size); + GzipInputStream gzin( + &input, GzipInputStream::GZIP, gzip_buffer_size); + ReadStuff(&gzin); + + delete [] buffer; +} + +TEST_F(IoTest, GzipIoReadAfterFlush) { + const int kBufferSize = 2*1024; + uint8* buffer = new uint8[kBufferSize]; + + int block_size = kBlockSizes[4]; + int gzip_buffer_size = block_size; + int size; + ArrayOutputStream output(buffer, kBufferSize, block_size); + GzipOutputStream::Options options; + options.format = GzipOutputStream::GZIP; + if (gzip_buffer_size != -1) { + options.buffer_size = gzip_buffer_size; + } + + GzipOutputStream gzout(&output, options); + WriteStuff(&gzout); + EXPECT_TRUE(gzout.Flush()); + size = output.ByteCount(); + + ArrayInputStream input(buffer, size, block_size); + GzipInputStream gzin( + &input, GzipInputStream::GZIP, gzip_buffer_size); + ReadStuff(&gzin); + + gzout.Close(); + + delete [] buffer; +} + TEST_F(IoTest, ZlibIo) { const int kBufferSize = 2*1024; uint8* buffer = new uint8[kBufferSize]; @@ -324,8 +474,12 @@ TEST_F(IoTest, ZlibIo) { int size; { ArrayOutputStream output(buffer, kBufferSize, kBlockSizes[i]); - GzipOutputStream gzout( - &output, GzipOutputStream::ZLIB, gzip_buffer_size); + GzipOutputStream::Options options; + options.format = GzipOutputStream::ZLIB; + if (gzip_buffer_size != -1) { + options.buffer_size = gzip_buffer_size; + } + GzipOutputStream gzout(&output, options); WriteStuff(&gzout); gzout.Close(); size = output.ByteCount(); @@ -348,7 +502,9 @@ TEST_F(IoTest, ZlibIoInputAutodetect) { int size; { ArrayOutputStream output(buffer, kBufferSize); - GzipOutputStream gzout(&output, GzipOutputStream::ZLIB); + GzipOutputStream::Options options; + options.format = GzipOutputStream::ZLIB; + GzipOutputStream gzout(&output, options); WriteStuff(&gzout); gzout.Close(); size = output.ByteCount(); @@ -360,7 +516,9 @@ TEST_F(IoTest, ZlibIoInputAutodetect) { } { ArrayOutputStream output(buffer, kBufferSize); - GzipOutputStream gzout(&output, GzipOutputStream::GZIP); + GzipOutputStream::Options options; + options.format = GzipOutputStream::GZIP; + GzipOutputStream gzout(&output, options); WriteStuff(&gzout); gzout.Close(); size = output.ByteCount(); @@ -402,9 +560,10 @@ TEST_F(IoTest, CompressionOptions) { // Some ad-hoc testing of compression options. string golden; - File::ReadFileToStringOrDie( - TestSourceDir() + "/google/protobuf/testdata/golden_message", - &golden); + GOOGLE_CHECK_OK(File::GetContents( + TestSourceDir() + + "/google/protobuf/testdata/golden_message", + &golden, true)); GzipOutputStream::Options options; string gzip_compressed = Compress(golden, options); @@ -432,6 +591,71 @@ TEST_F(IoTest, CompressionOptions) { EXPECT_TRUE(Uncompress(gzip_compressed) == golden); EXPECT_TRUE(Uncompress(zlib_compressed) == golden); } + +TEST_F(IoTest, TwoSessionWriteGzip) { + // Test that two concatenated gzip streams can be read correctly + + static const char* strA = "0123456789"; + static const char* strB = "QuickBrownFox"; + const int kBufferSize = 2*1024; + uint8* buffer = new uint8[kBufferSize]; + char* temp_buffer = new char[40]; + + for (int i = 0; i < kBlockSizeCount; i++) { + for (int j = 0; j < kBlockSizeCount; j++) { + ArrayOutputStream* output = + new ArrayOutputStream(buffer, kBufferSize, kBlockSizes[i]); + GzipOutputStream* gzout = new GzipOutputStream(output); + CodedOutputStream* coded_output = new CodedOutputStream(gzout); + int32 outlen = strlen(strA) + 1; + coded_output->WriteVarint32(outlen); + coded_output->WriteRaw(strA, outlen); + delete coded_output; // flush + delete gzout; // flush + int64 pos = output->ByteCount(); + delete output; + output = new ArrayOutputStream( + buffer + pos, kBufferSize - pos, kBlockSizes[i]); + gzout = new GzipOutputStream(output); + coded_output = new CodedOutputStream(gzout); + outlen = strlen(strB) + 1; + coded_output->WriteVarint32(outlen); + coded_output->WriteRaw(strB, outlen); + delete coded_output; // flush + delete gzout; // flush + int64 size = pos + output->ByteCount(); + delete output; + + ArrayInputStream* input = + new ArrayInputStream(buffer, size, kBlockSizes[j]); + GzipInputStream* gzin = new GzipInputStream(input); + CodedInputStream* coded_input = new CodedInputStream(gzin); + uint32 insize; + EXPECT_TRUE(coded_input->ReadVarint32(&insize)); + EXPECT_EQ(strlen(strA) + 1, insize); + EXPECT_TRUE(coded_input->ReadRaw(temp_buffer, insize)); + EXPECT_EQ(0, memcmp(temp_buffer, strA, insize)) + << "strA=" << strA << " in=" << temp_buffer; + + EXPECT_TRUE(coded_input->ReadVarint32(&insize)); + EXPECT_EQ(strlen(strB) + 1, insize); + EXPECT_TRUE(coded_input->ReadRaw(temp_buffer, insize)); + EXPECT_EQ(0, memcmp(temp_buffer, strB, insize)) + << " out_block_size=" << kBlockSizes[i] + << " in_block_size=" << kBlockSizes[j] + << " pos=" << pos + << " size=" << size + << " strB=" << strB << " in=" << temp_buffer; + + delete coded_input; + delete gzin; + delete input; + } + } + + delete [] temp_buffer; + delete [] buffer; +} #endif // There is no string input, only string output. Also, it doesn't support @@ -700,6 +924,26 @@ TEST_F(IoTest, LimitingInputStream) { ReadStuff(&input); } +// Checks that ByteCount works correctly for LimitingInputStreams where the +// underlying stream has already been read. +TEST_F(IoTest, LimitingInputStreamByteCount) { + const int kHalfBufferSize = 128; + const int kBufferSize = kHalfBufferSize * 2; + uint8 buffer[kBufferSize]; + + // Set up input. Only allow half to be read at once. + ArrayInputStream array_input(buffer, kBufferSize, kHalfBufferSize); + const void* data; + int size; + EXPECT_TRUE(array_input.Next(&data, &size)); + EXPECT_EQ(kHalfBufferSize, array_input.ByteCount()); + // kHalfBufferSize - 1 to test limiting logic as well. + LimitingInputStream input(&array_input, kHalfBufferSize - 1); + EXPECT_EQ(0, input.ByteCount()); + EXPECT_TRUE(input.Next(&data, &size)); + EXPECT_EQ(kHalfBufferSize - 1 , input.ByteCount()); +} + // Check that a zero-size array doesn't confuse the code. TEST(ZeroSizeArray, Input) { ArrayInputStream input(NULL, 0); |