Andres Freund <andres@2ndquadrant.com> writes:
> On 2014-01-17 13:50:08 -0500, Tom Lane wrote:
>> I think a better solution approach is to teach our src/port/snprintf.c
>> about the z flag, then extend configure's checking to force use of our
>> snprintf if the platform's version doesn't handle z.
> Hm. I had thought about that, but dismissed it because I thought people
> would argue about it being too invasive...
How so? It'd be a lot less invasive than what we'd have to do to use
'z' flags the other way.
> If we're going there, we should just eliminate expand_fmt_string() from
> elog.c and test for it in configure too, right?
If you mean "let's rely on glibc for %m", the answer is "not bloody
likely". See useful_strerror(), which is functionality we'd lose if
we short-circuit that.
>> You suggest below that we could invent some additional
>> macros to support that; but since the "z" flag is in C99, there really
>> ought to be only a small minority of platforms where it doesn't work.
> Well, maybe just a minority numberwise, but one of them being windows
> surely makes it count in number of installations...
Agreed, but I believe we're already using src/port/snprintf.c on Windows
because of lack of %n$ support (which is also required by C99).
regards, tom lane