[Bug middle-end/60429] [4.7 Regression] Miscompilation (aliasing) with -finline-functions

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60429

--- Comment #31 from rguenther at suse dot de <rguenther at suse dot de> ---
On Sat, 22 Mar 2014, mikpelinux at gmail dot com wrote:

> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60429
> 
> Mikael Pettersson <mikpelinux at gmail dot com> changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |mikpelinux at gmail dot com
> 
> --- Comment #29 from Mikael Pettersson <mikpelinux at gmail dot com> ---
> Richard, your 4.8 backport performs the same call twice in a row on lines 3017
> and 3018 in tree-ssa-structalias.c; is that really intentional?
> 
> +      /* We have to include all fields that overlap the current
> +         field shifted by rhsoffset.  And we include at least
> +         the last or the first field of the variable to represent
> +         reachability of off-bound addresses, in particular &object + 1,
> +         conservatively correct.  */
> +      temp = first_or_preceding_vi_for_offset (curr, offset);
> +      temp = first_or_preceding_vi_for_offset (curr, offset);

No, fixed.