This applies to all versions of valgrind, actually. If I start valgrind in "attach gdb to process when error encountered" mode (aka --gdb-attach=yes), when an error IS encountered and gdb has been started and attached, trying to run the backtrace command inside gdb produces a gobbledegook stack-trace. If I configure, build and install valgrind 2.0.0 myself, then everything works fine. After much diffing, finally worked out that this is because the valgrind binaries and shared-libraries are stripped during the emerge, and this loss of symbolic information is quite crucial in valgrind's case! Simple fix is to add the following lines to ALL of the valgrind ebuild files: # Do _NOT_ strip symbols in the build! RESTRICT="nostrip" I added them AFTER the following line in the 2.0.0 ebuild file: inherit flag-o-matic I'm attaching a modified ebuild for a proposed 2.0.0-r1 version. This is the same as the existing 2.0.0 except for the added "nostrip" lines. However, I have also rejigged the ebuild to bring it more inline with those for 2.1.0 and 2.1.1. I repeat, similar fixes (i.e. adding the "nostrip" line) are needed in the 2.1.0 and 2.1.1 ebuilds. However, I can't be bothered to do this since those builds are currently broken for other reasons (see http://bugs.gentoo.org/show_bug.cgi?id=49084). Reproducible: Always Steps to Reproduce: 1. 2. 3.
Created attachment 30140 [details] Fixed ebuild for valgrind 2.0.0
reattach the patch as text-plain please.
Created attachment 30987 [details] Fixed ebuild for valgrind 2.0.0 Re-attached new ebuild as plain text. Now identical to 2.1.1 ebuild, bar my changes and addition of eutils inheritance (which was added to 2.0.0 ebuild by Aron Griffis but not to 2.1.0 or 2.1.1).
added to valgrind ebuilds - thanks for the report.