SPECS: ruby.spec - document ia64 bug better, so it can be examined...

Jakub Bogusz qboosh at pld-linux.org
Mon Jul 26 20:11:21 CEST 2004


On Sun, Jul 25, 2004 at 11:48:42PM -0600, Aredridel wrote:
> >  # bleh, some nasty (gcc or ruby) bug still not fixed
> > +# (SEGV or "unexpected break" on miniruby run during build)
> >  %define		specflags_ia64	-O0
> 
> Asked about on ruby-talk at ruby-lang.org. I'll add a patch if someone has
> information.

It's something strange, probably some bug causing memory corruption.
In gc.c there is some "magic stack limit" applied on ia64 (which is too
low to build ruby now anyway), but it doesn't change anything when code
is compiled with -O1 or -O2.
With -O2 miniruby fails with "unexpected break".
With -O1 catches SIGSEGV.
Adding -fno-strict-aliasing doesn't help.

Builds only with -O0 and "magic stack limit" disabled.

I've googled a bit and found that:
- RH/Fedora and SuSE use -O0 and have ruby on ia64,
- Gentoo and some BSD marked ia64 port as broken.


-- 
Jakub Bogusz    http://cyber.cs.net.pl/~qboosh/




More information about the pld-devel-en mailing list