[Bug middle-end/51994] [4.6/4.7 Regression] git-1.7.8.3 miscompiled due to negative bitpos from get_inner_reference

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51994

--- Comment #22 from Uros Bizjak <ubizjak at gmail dot com> 2012-01-26 18:41:57 UTC ---
(In reply to comment #20)

> I agree that making get_inner_reference artificially return a non-zero poffset
> would most certainly be problematic as this would change the semantics.  But
> it's also clear that the lower level bit-field manipulation routines aren't
> really prepared to deal with negative stuff.  So I think that we shouldn't
> change the prototypes of these routines, but instead patch up callers that
> forward the values returned by get_inner_reference to these routines.
> 
> Adding assertions in these routines could indeed help.

I have added these. But.. could you please take the fix to this problem
further?