Bugzilla Bug 149008

I can get to the menu screen, but when I try to load any of the original quake3
levels, it crashes.  Interestingly, a couple of the levels that come with the
Reaction Quake 3 mod (420main and feud) will load.  The first is completely
playable (the elevator sounds aren't being played correctly, though); the
latter will load, but as soon as you start to interact with anything, it
crashes like the rest.

After upgrading to gcc and glibc to current versions a month ago, I rebuilt
everything, and many things broke.  I went around fixing them, and have only
now noticed that quake3 has stopped working, so I'm not sure exactly what has
broken it.  I do remember upgrading to openal-0.0.8 happened right before it,
but downgrading to older versions doesn't fix it.  I recompiled openal with the
debug flag, and even applied the patches described in <a
href="http://bugs.gentoo.org/show_bug.cgi?id=145245">Bug 145245</a>, to no
avail.  The output of the crash and my emerge --info are attached.

Since I haven't found any other bugs that sound like this, I'm guessing it's
going to be unreproducible on other machines, just a peculiarity of my
hardware.

Created an attachment (id=97991) [details]
Output of quake3 crash

Created an attachment (id=97992) [details]
emerge --info output

Can you see if it still crashed with openal-0.0.8-r1?

Yes, it still crashes with openal-0.0.8-r1.  No change at all.

Recompile quake3 w/o mfpmath=sse. It won't work with quake's i387 asm.

(In reply to comment #5)
> Recompile quake3 w/o mfpmath=sse. It won't work with quake's i387 asm.

This fixed it.  Thanks!  I would have never figured this out myself.  Is using
this CFLAG a bad idea in general, or is this just a quirk of quake3?

As long as your package doesn't do explicit i387 operation, the flag is safe,
but if it does (as quake3) it breaks. But there should be the possibility to
disable i387 asm, as quake3 works on x86_64 as well, where newer gcc's uses
mfpmath=sse by default.

BTW, I don't think this bug shou be marked "fixed", as the ebuiold doesn't
filter the offending cflag out on x86. The bug should rather be marked
INVALID...

Actually, I just updated CVS, so we can leave this as FIXED.