optimization issues on an SGI system

[Tim Peters]

[Mary Haley, running on an SGI box]
> ...
> If I didn't heed this warning, then the simple line of code:
>
>    p = 0 + 1.
>
> would cause python to core dump.  Turning off optimization fixed the
> problem.
> ...
> This partially optimized version was created by narrowing down the
> problem to a single file, and only turning off optimization for it.
> The file is "$PYTHON/Objects/floatobject.c".

Robert Minsk did an excellent analysis of what's probably the same problem,
and submitted a short floatobject.c patch to worm around it:

http://sf.net/tracker/?group_id=5470&atid=305470&func=detail&aid=436193

Nothing wrong with the patch, but we haven't checked it in and don't intend
to (see the patch and associated bug report for details).

It would be nice if the SGI community stepped up to the plate and maintained
a matrix of SGI patches (Python release vs compiler version) needed to worm
around the never-ending stream of SGI -O bugs.

It would also be nice if SGI added Python to their compiler regression
suite, since, IIRC. no release of Python has ever worked 100% right on an
SGI box due to -O bugs (which vary across both Python and SGI releases).

BTW, I don't mean to impugn the SGI compilers:  they have more -O bugs than
most because their optimization is more aggressive than most.  OTOH, it's
not my job to fix their bugs either <wink>.