[LTO] Committed merge feedback to the branch

[Diego Novillo]

On Fri, Oct 2, 2009 at 06:45, Richard Guenther <rguenther@suse.de> wrote:

> How would that happen?  I assumed we only ever emit things from
> the cgraph and varpool...

These are things reachable from BINFO_VIRTUALS, DECL_ABSTRACT_ORIGIN,
etc.  We then try to merge to share them again on the reader side, but
fail to find an assembler name for them (which is the only thing we
can use to merge them).  One could argue that these should be cleared
out, though BINFO_VIRTUALS are used for obj_type_ref folding.


> It might be caused by a local patch that enables AC_SYS_LARGEFILE
> throughout the build and our libelf _not_ using LFS.  I'm double-checking
> with the AC_SYS_LARGEFILE removed.  A quick check on my native i586
> build w/o LFS doesn't reproduce the segfaults at least.

OK.


Diego.