Re: [PATCH] gcc cannot deal with full /tmp

[Denys Vlasenko <dvlasenk@redhat.com>]

On Mon, 2008-08-18 at 14:09 +0200, Andreas Schwab wrote:
> >> > ELOOP is a good counter-example. "ln -s bogus /tmp/bogus" is not such
> >> > an implausible operation. Any user can do that.
> >> 
> >> Can you find a single OS which gives you ELOOP for O_EXCL creat?
> >
> > No.
> >
> > But I cannot find an OS which gives me EISDIR either,
> > thus the same argument works against EISDIR all the same.
> >
> > My argument is that it's better to proactively assume that there can be
> > some OSes which do it, or will do it in the future, and code around it
> > so that we won't need to revisit this issue.
> 
> If you can't create a file in /tmp for a reason other than `file exists'
> trying a different name is unlikely to succeed either, so in practice,
> EEXIST is the only interesting error.

Did you read the whole thread? Specifically, the below mail?
It's not *me* who said that we should check other errno's.

Moreover, at https://bugzilla.redhat.com/show_bug.cgi?id=203231
I *did* attach a patch which checks only for EEXIST.


On Fri, 2008-08-01 at 14:51 -0700, Zack Weinberg wrote:
On Fri, Aug 1, 2008 at 3:57 AM, Denys Vlasenko <dvlasenk@redhat.com>
wrote:
> [...]
> > My original approach was to reduce number of loops
> > instead of bailing out _immediately_ on errno != EEXIST.
> > I feared that otherwise I'll have to deal with
> > "obscure system XYZ throws weird errno at me" syndrome.
> >
> > This is it coming true.
> 
> I'm quite confident that EEXIST and EISDIR are the only two errno
> codes that should trigger looping.


--
vda