https://bugs.kde.org/show_bug.cgi?id=387766
--- Comment #6 from Julian Seward <jsew...@acm.org> --- Verified .. gcc version 8.0.0 20171210 runs clean on x86_64 linux when configured with --enable-valgrind-annotations, the trunk, and the https://bugs.kde.org/show_bug.cgi?id=387664 patch, which will land soon. I didn't try without the patch. I should note in passing, that I previously tried the 20171203 snapshot without --enable-valgrind-annotations, since I was not aware of it. I got a lot of complaints, all originating from sparseset_bit_p() in gcc/sparseset.h, because this deliberately does a comparison on undefined values. I fixed this using the patch below. I mention this because I am concerned that --enable-valgrind-annotations might "fix" the problem by zero-initialising various allocations that create memory used in sparseset_bit_p(). IMO that is the wrong solution, because it will "hide" any legitimate uninitialised values that would otherwise have been read from such allocations -- and so you reduce Memcheck's ability to detect errors in gcc. That said, this is only speculation -- I haven't actually checked what --enable-valgrind-annotations does. My fix follows. Clearly you'd actually want to ifdef the macro call so that this actually builds on targets that don't have <valgrind/memcheck.h>. --- gcc-8-20171203/gcc/sparseset.h~ 2017-01-01 13:07:43.000000000 +0100 +++ gcc-8-20171203/gcc/sparseset.h 2017-12-06 10:21:00.447480896 +0100 @@ -130,24 +130,27 @@ static inline SPARSESET_ELT_TYPE sparseset_size (sparseset s) { return s->size; } /* Return true if e is a member of the set S, otherwise return false. */ +#include <valgrind/memcheck.h> + static inline bool sparseset_bit_p (sparseset s, SPARSESET_ELT_TYPE e) { SPARSESET_ELT_TYPE idx; gcc_checking_assert (e < s->size); idx = s->sparse[e]; + VALGRIND_MAKE_MEM_DEFINED(&idx, sizeof(idx)); return idx < s->members && s->dense[idx] == e; } /* Low level insertion routine not meant for use outside of sparseset.[ch]. Assumes E is valid and not already a member of the set S. */ static inline void -- You are receiving this mail because: You are watching all bug changes.
