On Mon, Jan 11, 2021 at 01:05:37PM -0800, Junio C Hamano wrote: > Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx> writes: > > > What do you think about just removing it? I.e. make setting it a noop? > > I have been seeing occasional CI job failures from new tests that > forget to use test_i18ngrep. I actually think marking such a grep > as "this is looking for a string that is meant for humans" a good > way to document the interface and expected end-user interaction, > so I am not sure about just removing it. > > So after all, test_i18ngrep may make more sense than setting > GIT_TEST_GETTEXT_POISON to false. I dunno. I agree that test_i18ngrep is preferable to setting GIT_TEST_GETTEXT_POISON. Since it's tied to the comparison itself, I think it's easier to see which messages we're expecting as human-readable. It's also shorter and more readable IMHO (you'd already be calling grep or test_cmp, so it is only a little longer). I am on the fence on whether the presence of something like test_i18ngrep in the test suite is really serving as a useful indication of what was meant to be translated and what wasn't. It's such an incomplete coverage of the total set of messages we generate that I wouldn't trust it. I'd be more likely to look at the source to see if a message is actually translated, or just assume we follow some basic rules (in general, things to stderr are translatable; I think the plumbing outputs from unpack_trees() were really the exception). So I'm open to the idea that the whole poison mechanism has just outlived its usefulness. -Peff
