Re: [Valgrind-developers] valgrind: r10938 - trunk/coregrind

[Bart V. A. <bar...@gm...>]

On Thu, Nov 12, 2009 at 2:19 PM,  <sv...@va...> wrote:
> Author: tom
> Date: 2009-11-12 13:19:41 +0000 (Thu, 12 Nov 2009)
> New Revision: 10938
>
> Log:
> Fix stack unwinding on PPC to store the correct frame pointer for
> frames other than the first one found.
>
> This is taken from Jakub Jelinek's second patch on bug #210479.
>
>
> Modified:
>   trunk/coregrind/m_stacktrace.c
>
>
> Modified: trunk/coregrind/m_stacktrace.c
> ===================================================================
> --- trunk/coregrind/m_stacktrace.c      2009-11-11 19:22:05 UTC (rev 10937)
> +++ trunk/coregrind/m_stacktrace.c      2009-11-12 13:19:41 UTC (rev 10938)
> @@ -430,9 +430,9 @@
>  #           endif
>
>             if (0 == ip || 1 == ip) break;
> -            fp = (((UWord*)fp)[0]);
>             if (sps) sps[i] = fp; /* NB. not sp */
>             if (fps) fps[i] = fp;
> +            fp = (((UWord*)fp)[0]);
>             ips[i++] = ip - 1; /* -1: refer to calling insn, not the RA */
>             if (debug)
>                VG_(printf)("     ipsF[%d]=%#08lx\n", i-1, ips[i-1]);
>

Hello Tom,

This patch looks like an improvement -- e.g.
memcheck/tests/varinfo2.stderr.diff and
memcheck/tests/varinfo4.stderr.diff became smaller thanks to this
patch. But as you can see in e.g. memcheck/tests/varinfo1.stderr.diff,
unfortunately the off-by-one line number problem in the call stacks
reported on PPC still exists.

Was this the intended effect of this patch ?

Bart.