Re: Pre-inline optimization, version 3

[Jan Hubicka <jh at suse dot cz>]

> > WHat I saw in the file producing segfault, was VAR_DECL pointing to
> > ARRAY_REF via the language specific pointer 't'. The ARRAY_REF further
> > pointed to SSA_NAMES that was however already optimized out.  
> 
> Actually, if it is DECL_RENAMED_OBJECT, it's not GIMPLE, but GENERIC, since
> there's no attempt to gimplify it.  So there's definitely a bug here: if
> there *is* some use for it outside the front-end (and I don't see how there
> could be), then there needs to be a VAR_DECL case in gimplify_gnat_expr to
> gimplify it.  If it isn't needed (as I suspect), that case should clear it out.

The variable itself was available from BIND_EXPR, perhaps Ada frontend
didn't decided to gimplify the BIND_EXPR in question.  Is it posssible that
DECL_RENAME_OBJECT in BIND_EXPR of one function point to ARRAY_REF used
in statement of other function?

Honza