> > I have repeated the problem with CVS head on XP SP2. It
> *does* create
> > it there (or rather, it tries to).
>
> > tmpnam() returns a file in the current dir per documentation, but I
> > see it generating one in the root instead.
> > tempnam() uses TMP environment variable.
>
> > tmpfile() and tmpnam() both use the same function to generate the
> > filename.
>
> Hm. I guess I concur with Peter's conclusion: the cleanest
> fix is to put our own implementation of tmpfile() into src/port/.

Ok. Should be easy enough once the code is fine - can you comment on the
patch as sent, if the code itself looks right provided i wrap it up in a
function in port/?

//Magnus

---------------------------(end of broadcast)---------------------------
TIP 3: Have you checked our extensive FAQ?

http://www.postgresql.org/docs/faq

"Magnus Hagander" <mha@sollentuna.net> writes:
> Ok. Should be easy enough once the code is fine - can you comment on the
> patch as sent, if the code itself looks right provided i wrap it up in a
> function in port/?

Not sure if the error handling is adequate --- are there any cases
besides EEXIST that should loop?

A look at
http://msdn.microsoft.com/library/de..._wtmpna m.asp

suggests that tempnam() is also pretty fragile, esp. if you're passing
NULLs. Apparently it won't work at all if TMP isn't set?

regards, tom lane

---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster