diff options
| author | Neil Fuller <nfuller@google.com> | 2014-04-09 10:58:08 +0100 |
|---|---|---|
| committer | Neil Fuller <nfuller@google.com> | 2014-04-15 17:42:16 +0100 |
| commit | c96651d01dc87a20aa718d60feaff6611451852e (patch) | |
| tree | ce98bb104e94dfdb9c066fced7cab12ca0760a5b /expectations | |
| parent | d1c5e1ffb8bb8a351300d96e5edb9eda2c81741f (diff) | |
| download | libcore-c96651d01dc87a20aa718d60feaff6611451852e.zip libcore-c96651d01dc87a20aa718d60feaff6611451852e.tar.gz libcore-c96651d01dc87a20aa718d60feaff6611451852e.tar.bz2 | |
Change DecimalFormat.setRoundingMode() behavior to stay in spec
The serialization tests revealed that some DecimalFormat instances
could not be round-tripped. The reason is that the setRoundingMode()
method was setting a non-zero rounding increment, which cannot
be persisted, and also creates patterns that are outside of the
DecimalFormat spec.
Addressed the fact that ICU4C appears to support
RoundingMode.UNNECESSARY now by deleting special-case code.
Changed SerializationStressTest to use canonical DecimalFormat
patterns:
The ones chosen had a couple things that were not ideal:
1) Putting '-' in quotes causes ICU (for non-obvious reasons)
treat the negative pattern prefix as not being default. Default
in this case means "the positive prefix with a preceding '-'.
Removing the quotes meant it would recognize the pattern as
not needing everything after ;., so the pattern would be shrunk.
2) The positive pattern was also not one ICU could deal with
fully. ICU changed "#.#" to "#0.#" at parse time (which can be
corrected with a setMinimumIntegerDigits(0)). It's not clear why
ICU cannot deal with parsing "#.#" when it can produce it, but
there is a long comment in decimfmt.cc suggesting it is
intentional and to avoid format producing empty strings.
The pattern has been changed to one that ICU can handle and would
choose when the various serialized fields are read back. This is
necessary because NativeDecimalFormat compares the patterns and
not just the serialized fields. Leaving the pattern check part of
equals() means that two non-spec patterns will not be considered
equal(), which is important while Android allows non-spec patterns.
Restricting support in DecimalFormat to the documented subset of
characters is not worth the effort and may cause problems
elsewhere.
Change-Id: I2de588458b86619733c0dc1692d526f179059910
Bug: 11668227
Diffstat (limited to 'expectations')
| -rw-r--r-- | expectations/knownfailures.txt | 2 |
1 files changed, 0 insertions, 2 deletions
diff --git a/expectations/knownfailures.txt b/expectations/knownfailures.txt index eabadfd..cf79a53 100644 --- a/expectations/knownfailures.txt +++ b/expectations/knownfailures.txt @@ -1700,8 +1700,6 @@ description: "Known failures in SerializationStressTest", bug: 11668227, names: [ - "org.apache.harmony.tests.java.io.SerializationStressTest3#test_18_113_writeObject", - "org.apache.harmony.tests.java.io.SerializationStressTest3#test_18_115_writeObject", "org.apache.harmony.tests.java.io.SerializationStressTest4#test_writeObject_Collections_UnmodifiableMap_UnmodifiableEntrySet" ] }, |