https://bugs.kde.org/show_bug.cgi?id=345307
--- Comment #34 from Philippe Waroquiers <philippe.waroqui...@skynet.be> --- (In reply to Philippe Waroquiers from comment #32) > As I understand, the problem is 'cleanly' fixed in a newer version of glibc, > by having valgrind calling __gnu_cxx::__freeres, but which is only present > in newer library version. Humph, in fact, no :). The problem of having a reachable at the end of the execution is solved, but even when the __gnu_cxx::__freeres is there, during the run, the test produces different output. So, yes, we need to do something to have the test succeeding even with a new gcc and a new lib. I am however wondering if we can do something cleaner that new .exp files, e.g. by rather having some filter script on the output, to make the filtered output identical on various platforms and/or combine a filter with a suppression ntry. I think the test objective can be reasonably verified just by counting the nr of blocks. The nr of bytes leaked is not critical to verify the heuristics are working. If the filter transforms all 'byte values' in xxx bytes (similarly for the increase), and we filter the suppressed: line, I think we can end up with a single .exp file for all our current cases ciombinations (32 bits, 64 bits, solaris; old or new gcc/glib) -- You are receiving this mail because: You are watching all bug changes.
