Re: ab/i18n (What's cooking in git.git (Feb 2011, #05; Wed, 23))

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Feb 24, 2011, 02:55 UTC
Message-ID: <AANLkTinY4pt5DLokKkuCNnC1yi7nQBHcZv722x9CKvCh@mail.gmail.com>
In-Reply-To: <7vd3mi44gx.fsf@alter.siamese.dyndns.org>

On Thu, Feb 24, 2011 at 02:16, Junio C Hamano <gitster@pobox.com> wrote:

Show 14 quoted lines

> Jonathan Nieder <jrnieder@gmail.com> writes:
>
>> The commit message for the second one seems to have been mangled:
>>
>>       i18n: add GETTEXT_POISON to simulate translated messages unfriendly translator
>
> Yeah, I think I was fooled by the header folding while fixing things up
> inside the mailbox.
>
>> I still don't like the #-sign business in this commit.  Couldn't it
>> be split into a separate patch, not to be applied until just before
>> the strings in commit/tag/wt-status are marked for translation?
>
> That might be a sensible thing to do.   Ævar what do you think?

It could, but I don't want to spend time on it. I think it's fine that the choice of string is documented in the commit that introduces it.

Show 6 quoted lines

>> There is also a patch out there to make this use rot13, which I am
>> somewhat fond of.  Unfortunately, it leaks (because it is not clear
>> how long translated strings are supposed to last).
>
> Yeah I would imagine it would leak.  Also blindly running rot13 to turn %d
> into %q is probably not what you want.

Gettext gives you a pointer to the string inside a mmaped .mo file, effectively, so the caller doesn't have to care about it. So you can't write something that doesn't leak memory unless you preserve that characteristic.

But it's a debugging mode, it doesn't matter IMO if it leaks memory.

Anyway, to be blunt I really don't see the point of fiddling around with this bit so much. Whether it's a `"GETTEXT POISON"` constant or `rot13(msgid)` the same tests will fail. So it's not functionally changing what the feature was for in the first place, just adding more complexity.

Of course there might be cases where a test will fail because it's supposed to end in \n but the poison string doesn't, but since none of them did I didn't worry about that.

And even if one did I'd probably just mark it as C_LOCALE_ONLY anyway. Since the intent is to make the person doing the i18n work not break plumbing, not to perfectly annotate our test suite. I'd much rather have a few more tests skipped under this poison mode than 20 extra lines of C code that effectively give us nothing to maintain.

If someone else wants to do the work I don't care. I just think there's more important things to worry about.