On Mon, Mar 5, 2012 at 21:39, Junio C Hamano <gitster@xxxxxxxxx> wrote: > I do not see any reason to insist that their fallback behaviour must > be to fill them with question marks. In other words, I simply do > not see the point of this test and the original "pass runs of > question" to begin with, and this patch makes things even worse. > > Why not get rid of this single test that only tries to check the > output for an impossible request? > > Ævar, am I missing something? Sorry about the late reply, I think if anything it needs to be expanded beyond its current scope. Our gettext implementation is peculiar in two ways: * We support 3 different languages, one of which (Perl) is supported through a non-GNU package. * We have this fantastic hack in gettext.c's init_gettext_charset to deal with a glibc bug. I think in particular re-encoding works completely differently under the Perl i18n library than it does with GNU gettext (and we don't have tests for this) and instead of just dropping unknown characters under non-UTF-8 locales on some platforms everything may turn into complete gibberish. When I added this test this was what I was planning to smoke out. Different behaviors on different platforms. In this case the OSX behavior seems benign, but I'd much rather patch the test to detect OSX rather than removing it entirely, maybe other obscure operating systems (Aix, HP/UX etc.) will behave differently. We still don't really have any use of the i18n feature, i.e. we haven't had translations submitted yet, and thus don't have any users. I think it's important to be pro-active about trying to spot any issues that might affect end users before they happen. -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html