Re: ERRORDATA_STACK_SIZE panic crashes on Windows

Bruce Momjian <bruce(at)momjian(dot)us> writes:
> Tom Lane wrote:
>> Peter Eisentraut <peter_e(at)gmx(dot)net> writes:
>>> I think that was the first version that worked sanely in general.
>>
>> Hmm. Bruce, what gettext version are you running exactly, and how much
>> have you ever tested the localization behavior with it?

> Uh, I can't seem to find the libintl version --- my bet is that it was
> installed as part of another package but I am not sure which one.
> I have never used it to test localization support so turning it off is
> fine --- I just enabled it to catch compile failures.

I dug through the gettext changelogs a bit more. There were *three
years* and a lot of fixes between 0.10.35 and 0.10.36; the one that
is most directly relevant to us is

* Locales which differ only in the character encoding, for example ja_JP and
ja_JP.UTF-8, can now share the same message catalogs. gettext converts
the messages to the appropriate character encoding on the fly.

Since we supply only one message catalog per language, it's probably
fair to say that gettext versions without this ability are broken for
our purposes.

So I'm thinking that the correct fix is to require
bind_textdomain_codeset to be present when --enable-nls is requested.
I will go make that happen in the versions that have the new patch
(ie, 8.2 and up).

regards, tom lane