[FFmpeg-devel] [RFC] make sure $TMPE is executable

[Reimar Döffinger]

On Thu, Sep 10, 2009 at 06:28:51PM +0100, M?ns Rullg?rd wrote:
> Reimar D?ffinger <Reimar.Doeffinger at gmx.de> writes:
> > On Thu, Sep 10, 2009 at 02:29:56PM +0100, M?ns Rullg?rd wrote:
> >> Reimar D?ffinger <Reimar.Doeffinger at gmx.de> writes:
> >> > it seems that "GNU gold (GNU Binutils 2.19.1) 1.7" has a strange bug
> >> 
> >> What's that?
> >
> > The "new", supposedly much better and faster (IIRC ELF-only?) linker...
> > Or to put it simpler, that's what you get when you do experiments and
> > set in Gentoo the "gold" use flag for binutils (and no Gentoo flaming,
> > I knew what I was getting myself into, allowing people to shoot
> > themselves in the foot is not generally a bit thing, if nothing else
> > it helps evolution along - though mostly I was hoping that maybe then
> > compiling C++ programs would no longer be about 10 times slower than
> > plain C).
> 
> So apart form this little quirk, is it any better?  Not that I'm
> concerned about linking times myself.

Haven't noticed a difference yet. Though I might when the next Qt update
or something comes up. I'll probably notice when those libtool processes
with several minutes of CPU time finally don't appear anymore (though I
doubt that will happen...).

> > So, should I apply this even if it is a hack?
> > Index: configure
> > ===================================================================
> > --- configure   (revision 19808)
> > +++ configure   (working copy)
> > @@ -1493,6 +1493,7 @@
> >  tmpfile TMPO  .o
> >  tmpfile TMPS  .S
> >  tmpfile TMPSH .sh
> > +chmod +x $TMPE
> >
> >  unset -f mktemp
> 
> Ping me tomorrow if I haven't replied or committed something.

Sure, no hurry. I mostly brought it up because with some stretching it
could be considered "correct" behaviour, as IIRC cp also behaves that
way (keeping permissions of already existing files), though I still tend
to consider it a linker bug.

> > The only other check_exec IIRC is because ICC silently generates
> > broken code when ebp clobber is used, I can't think of a good way to
> > check that differently, even though I admit this might cause a
> > slight performance degradation when cross-compiling for x86 with
> > gcc.
> 
> This rings a bell.  Wasn't there something about it behaving
> differently in main() than in other functions?  I'll have to dig up
> that old thread again.

Now that you mention it, I think so yes. But I obviously remember even
fewer details than you do...