ports/72014: Eclipse doesn't work (SigBus 10) if it has been built with GTK

[Panagiotis Astithas]

I'm forwarding this to the people who are responsible for this.

Daniel Roethlisberger wrote:
> I have just double checked on my pentium-m box, I can confirm that
> adding -mno-sse2 does solve the problem for me (at least as far as I can
> tell, as there is no definitive way to reproduce the problem).
> 
> Compiling x11-toolkits/gtk20 with CPUTYPE?=pentium-m and CFLAGS=-O -pipe
> results in an unstable eclipse, while compiling gtk20 with the same
> CPUTYPE but CFLAGS=-O -pipe -mno-sse2 saves the day, giving me a stable
> eclipse.
> 
> Can we add some kind of -mno-sse2 hack to the GTK port for the time
> being, or is there a better solution?


Some background:

In PR ports/72014 it was reported a long time ago that eclipse/gtk 
crashes and this was tracked down to non-standard compiler optimizations 
when building gtk. There have been similar bug reports filed against the 
eclipse and gcc bug databases. Some gcc people have tracked it down to 
incorrect stack alignment (see ports/72014 for the pointers) and have 
received reports that using -mno-sse2 fixed it. Daniel has verified that 
this seems to help our case also.

Therefore the question is, should we force -mno-sse2 to the compiler 
flags for the gtk ports? I would expect less bug reports from people who 
compile gtk apps with non-standard CFLAGS, at least.

Cheers,

Panagiotis