Re: ERRORDATA_STACK_SIZE panic crashes on Windows

Tom Lane wrote:
> Magnus Hagander <magnus(at)hagander(dot)net> writes:
> > Tom Lane wrote:
> >> We can either add a configure test or say that we don't support
> >> such old versions of gettext ...
>
> > Or we could just #ifdef the whole thing to win32, since it's not
> > really needed on other platforms, pushing that decision to later...
> > (when that version of gettext will be even more obsolete)
>
> That would work for the moment, but we're almost certainly going to
> have to insist on bind_textdomain_codeset being available eventually;
> AFAICS there's no hope of multi-locale/multi-encoding support without
> it.

Yes, that's why I said it would only push the decision to the future.

Perhaps doing this #ifdef would be a good idea for back-branches, and
then we look at one of the other solutions for 8.4?

> I was considering either:
>
> 1. Add a probe for bind_textdomain_codeset to configure, and
> conditionalize the new patch on HAVE_BIND_TEXTDOMAIN_CODESET.
>
> 2. Adjust the AC_SEARCH_LIBS call to probe for
> bind_textdomain_codeset() instead of gettext() as it does now. This
> would have the effect of rejecting pre-0.10.36 versions of the
> library.

Depending on the buildfarm issue it may be that the software is antique
enough that almost only Bruce runs such an old version. If so, I think
#2 is just fine (except in back branches, of course)

> Magnus' suggestion gives a third possibility.
>
> I notice that the PGAC_CHECK_GETTEXT macro already contains the
> comment dnl FIXME: We should probably check for version >=0.10.36.
> So depending on what that's about, there might be some other good
> reasons to go with choice #2. Peter, it appears you put that comment
> in when you first added the macro, on 2001-06-02. Do you remember
> why?

Could it possibly be for this very reason?

//Magnus