On 27/11/14 08:47, Daiki Ueno wrote: > Hello, > > When trying to compile an older gettext tarball with libunistring > 0.9.5-alpha3 (with an update to Unicode 7.0.0), I encountered a test > failure caused by a data mismatch: > > FAIL: uniname/test-uninames.sh > ============================== > > \u037F name lookup returned wrong name: GREEK CAPITAL LETTER YOT > > where \u037F is not defined in the test data bundled in the gettext > tarball (based on Unicode 5.0.0), but in the system's libunistring. For > the meantime I plan to go with the attached a patch, which skips the test > when system's libunistring is used. However, I'm not really sure of the > relationship between Gnulib modules and system's libunistring. > > For example, modules/unistr/u32-mbtouc-tests has a line: > > test_u32_mbtouc_LDADD = $(LDADD) $(LIBUNISTRING) > > This prefers system's libunistring to the corresponding Gnulib module. > I think Gnulib tests should rather prefer Gnulib modules. Does anyone > know what's the rationale behind this?
The tests generally test the combination of system and gnulib bits. Ideally the test would be split based on features per libunistring version, but if that's awkward then skipping is fine. thanks, Pádraig
