[RFC] Should SRA stop producing COMPONENT_REF for non-bit-fields (again)?

[Martin Jambor]

On Fri, Apr 13, 2012 at 01:57:33PM +0200, Rainer Orth wrote:
> Richard Guenther <rguenther@suse.de> writes:
> 
> >> Anyway, the patch I posted previously would risk re-introducing PR
> >> 50386 and PR 50326, even though they are very unlikely with just
> >> bit-fields.  So my current working version is the following, but it
> >> causes failure of libmudflap.c++/pass55-frag.cxx execution test so I'm
> >> not actually proposing it yet (sigh).
> >
> > I would not worry about mudflap tests.  The patch looks good to my
> > eyes.
> 
> Are you sure the failure is new?  At least for 64-bit at -O,
> libmudflap.c++/pass55-frag.cxx already fails right now (cf. PR
> libmudflap/49843).
> 

Well, my patch makes it fail on all other optimization levels too (the
supposedly no-optimization variant is actually compiled with -O2).
But thanks for pointing me towards the bug, I'll use it as an excuse
to commit the patch nevertheless (mainly because I really cannot see
anything wrong going on there, at least with respect to SRA and this
change).

Thanks,

Martin