http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59061
--- Comment #7 from Kostya Serebryany <kcc at gcc dot gnu.org> --- > > Clang supports -fsanitize=leak which simply links a standalone lsan library > > (no instrumentation at compile time required). > > Perhaps gcc can add such option too. > > I agree that such an option would be useful (see also below). I am not an expert in the gcc build system so this will have to be done by someone else. Also, I am heavily frightened by the amount of differences between the clang and gcc builds of libsanitizer. On my... The only good solution I see will hardly be accepted by the gcc community (which is: export compiler-rt as is, with all of its build and test system, anything else is going to increase the maintenance cost). > For now, I put '-fsanitize=address' for the link step only, which has the > same effect. It has a bit different effect because asan's allocator has greater memory footprint compared to a plain lsan's allocator. But -fsanitize=leak as a separate entity is not on top of our priorities. It works, but we don't test it on anything large. > > Additionally, it seems important to have -g -fno-omit-frame-pointer in the > options. Maybe gcc should add these options for best 'user experience', for > example as part of -fsanitize=address/leak. Power users can always override > with -fsanitize=leak -fomit-frame-pointer. -g: please no. -g incurs very large compile-time and binary size overhead. we need something like -gmlt (in google's gcc branch, not sure if it's available in trunk) or clang's -gline-tables-only. -fno-omit-frame-pointer: that may or may not be a good idea, I don't know. If we do any of these flags as default for -fsanitize=x, it would be nice to have the same defaults in clang.
