Valery, I haven't tried efence on ia64, but if you can produce a simple test program that causes it to crash on ia64 but not on x86/amd64, I'd love to look into it for you.
Patrick On Tue, Jan 15, 2013 at 7:33 AM, Valery Zaporozhchenko < valeryz2...@gmail.com> wrote: > Hello, > > faced very strange behaviour of efence on ia64 and need an advice. > After moving from g++-4.3 to 4.4 several first builds of my project > were very unsuccessful, with heap crashes. That persisted for every O2 > build while with O1 all was OK. Compiling with -lefence yields very > strange result I've never seen before: it appears that the corruptor > is malloc() itself, in strlen() call. Very subtle changes in the > source code (definitely not related to the problem, if any) removed > the runtime crashes (with O2 option) completely but the behaviour of > efence didn't change: every run of the code compiled with -g -lefence > eventually ends with a heap corruption within the same malloc() call, > regardless of O1/2/3 settings. > > Сareful examination of the code supposedly causing the malloc crash > revealed nothing indicative of the error. On ia32/amd64 valgrind says > that all is OK. Nevertheless, I really expect a hidden error in the > algorithm coded rather than a compiler error (and even rather than bug > in efence). I have 3 questions: > > - what do people think about efence on ia64? any experiences? > - which kind of data, private to malloc() itself and residing in > memory not guarded by efence I could corrupt before this call to > malloc()? > - very unusual case of stack corruption pretending to be a heap > corruption, if this is possible? I have very limited knowledge of > these things, unfortunately. > > The code is a solver for one of the graph-theoretical problems. > > Thank you for any advice > > Regards > Valery > > > -- > To UNSUBSCRIBE, email to debian-ia64-requ...@lists.debian.org > with a subject of "unsubscribe". Trouble? Contact > listmas...@lists.debian.org > Archive: > http://lists.debian.org/cahcucgfruu40+qptcucfcgyhwgfafxibmokwjruzbmyv8d4...@mail.gmail.com > >
