* Tom Lane <tgl@sss.pgh.pa.us> [001128 20:48] wrote:
> Adam Haberlach <adam@newsnipple.com> writes:
> > On Mon, Nov 27, 2000 at 04:09:46PM -0500, Tom Lane wrote:
> >> Somewhere right around here is where I am going to ask why we are
> >> entertaining the idea of a BeOS port in the first place... it's
> >> evidently not Unix or even trying hard to be close to Unix.
>
> > You've asked this before.
>
> > How does Windows manage to work?
>
> Objection! Point not in evidence!
>
> ;-)
>
> Seriously, we do not pretend to run on Windows. It does seem to be
> possible to run Postgres atop Cygwin's Unix emulation atop Windows.
> However, that's only because of some superhuman efforts from the
> Cygwin team, not because Windows is a Postgres-compatible platform.
>
> As far as the original question goes, I suspect that a rename() would
> work just as well as the link()/unlink() combo that's in that code now.
> I would have no objection to a submitted patch along that line. But the
> target audience for Postgres is POSIX-compatible platforms, and I do not
> think that the core group of developers should be spending much time on
> hacking the code to work on platforms that can't meet the POSIX spec.
> If anyone else wants to make that happen, we'll accept patches ... but
> don't expect us to supply solutions, OK?
Afaik the atomicity of rename() (the same as a link()/unlink() pair)
is specified by POSIX.
Sorry for jumping in late in the thread, but rename() sure sounds a
lot better than a link()/unlink() pair, but I'm probably taking it
out of context.
--
-Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org]
"I have the heart of a child; I keep it in a jar on my desk."
