Created attachment 108333 [details] memcheck output Found this crash when debugging project. I think it's not a normal case.
I need confirmation about if this corresponds to a bug. In that case I would like to solve it. Thanks.
There are invalid writes reported, so fix those and the crash will likely go away. It's almost certainly not a bug in valgrind, it's just your program corrupting memory, and valgrind is helpfully telling you exactly where.
Have you seen this? --11098-- VALGRIND INTERNAL ERROR: Valgrind received a signal 11 (SIGSEGV) - exiting --11098-- si_code=128; Faulting address: 0x0; sp: 0x802db5dc0 valgrind: the 'impossible' happened: Killed by fatal signal And the program happens to be the python interpreter. If it is the interpreter doing invalid writes, why valgrind says internal error and that impossible happened? Shouldn't it be prepared to handle signals?
(In reply to Daniel Gutson from comment #3) > Have you seen this? > > --11098-- VALGRIND INTERNAL ERROR: Valgrind received a signal 11 (SIGSEGV) - > exiting > --11098-- si_code=128; Faulting address: 0x0; sp: 0x802db5dc0 > > valgrind: the 'impossible' happened: > Killed by fatal signal > > > And the program happens to be the python interpreter. > If it is the interpreter doing invalid writes, why valgrind says internal > error and that impossible happened? Shouldn't it be prepared to handle > signals? For sure, valgrind can run an application that (properly) uses signal. The above SIGSEGV msg means that the signal happened at a time when valgrind was executing its own code (and not when running the guest code). Such SIGSEGV might be a bug in valgrind, or as indicated by Tom, might be the result of a problem in the guest application. Running python 'properly' under valgrind is typically implying to compile the python interpreter with special options (at least that was the case with python 2.7 IIUC). You might have with recent python a way to run more cleanly under valgrind. e.g. found the below on the web: PYTHONMALLOC=malloc python3 foobar.py It might be worth in any case to retry with a more recent valgrind either 3.13 last release, or the git trunk version, just in case ...
Thanks Philippe. We will retry with the conditions you suggest and will write the results here.
I don't think that this was a bug 5 years ago and still not a bug.