Web lists-archives.com

Re: What's cooking in git.git (Nov 2018, #04; Tue, 13)




Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx> writes:

>> * ab/dynamic-gettext-poison (2018-11-09) 2 commits
>>  - Makefile: ease dynamic-gettext-poison transition
>>  - i18n: make GETTEXT_POISON a runtime option
>>
>>  On hold.
>>  The tip one may be controversial, but at least it would get me going.
>>  cf. <xmqqpnvg8d5z.fsf@xxxxxxxxxxxxxxxxxxxxxxxxx>
>
> So just to clarify the state of this. I sent a v4 with the patch you're
> referring to included,...

Clearly I failed to update the comments when I replaced the patch
(v3 plus SQUASH fix) with what is queued which is v4.  Thanks for
noticing.  I think this one is ready to move forward, if it is
deemed worth doing.

> I'm a bit surprised that people aren't excited about this whole "you
> don't need to recompile git just to run this one special test mode",

I actually do not think it is surprising at all.  Only a few people
care about i18n breakages so they are not daily running the poison
test in the first place.  I myself only "see" its result in Travis
not because I am particularly interested in the poison test, but
because I cannot selectively not seeing it when visiting the summary
page like https://travis-ci.org/git/git/builds/454503815 (I do look
at its log if it fails, though).

So a true improvement for most of the developers is to somehow make
it unnecessary to even worry about poison test.  How poison test is
done is secondary to them and I think that is why we see nobody is
enthused when the way it is run is improved.  Having to carry code
that is only used while running poison test in the production binary
is an annoyance, not an improvement, to them.

As I do not immediately see a way to satisfy the pie-in-the-sky wish
"the world would be a great place if it were impossible to introduce
a bug that marks plumbing messages with _() so that we did not even
have to do a poison test", I am neutral, not negative, on this topic.
I am also hoping that the dynamic-poisoning may allow us to do a
"readable but still poisoned" fake translation more easily than the
separate build method we have used so far, so that adds a slight
positive to the topic.

Thanks.