On 28 August 2017 at 01:23, Jeff King <peff@xxxxxxxx> wrote: > On Sun, Aug 27, 2017 at 10:04:55PM +0200, Lars Schneider wrote: > >> I did run all tests under valgrind using "make valgrind" and I found >> the following files with potential issues: >> >> cat valgrind.out | perl -nE 'say /^==.+\((.+)\:.+\)$/' | sort | uniq -c >> 7102 >> 2 clone.c >> 33 common-main.c >> 6 connect.c >> 64 git.c >> 4 ls-remote.c >> 126 run-command.c >> 12 transport.c >> 7 worktree.c > > I'm not sure where valgrind.out comes from. The individual > test-results/*.out files may have valgrind output, but I don't think > they usually contain leak output. > > Doing "valgrind ./git-upload-pack . </dev/null >/dev/null" mentions > leaked memory but not the locations. Adding --leak-check=full shows that > most of it comes from format_packet(). > > And applying Martin's patch drops the "definitely lost" category down to > 0 bytes (there's still 550k in "still reachable", but those are in the > "exit will free them for us" category). > >> No mention of "pkt-line.c". Did you run Git with valgrind on one of >> your repositories to find it? > > I'm curious, too. I don't think the valgrind setup in our test suite is > great for finding leaks right now. Sorry for being brief. I've patched t/valgrind/valgrind.sh to say "--leak-check=yes". Then I run "./t0000 --valgrind", simply because running the complete suite gives more reports than I could possibly process. Then I check the first few leaks, verify that they're "ok" and add them to a suppressions-list. Lather, rinse, repeat. A couple of very targeted and well-motivated suppressions in git.git could perhaps be motivated, but there are many many reported leaks. My suppressions-list is getting gross. I started with t0000 and t0001 because I figure, once I have those basic suppressions in place (or leaks fixed), I can run some other more interesting tests. Of course, the concept of "this leak is ok" is a bit subjective. For example, we might do "return !!create_object(...);" (function name invented on the fly), which is a leak, and unreachable. But if this is only done once in builtin/foo.c and the object created is small, then this could be deemed "ok", since in practice this leak will never bring anyone over the cliff. If clean-ups in such code would not just be code churn, then I can of course adjust my definition of "ok" accordingly. This is not an attempt to find and fix a huge number of leaks, it's more to have a good reason to go through call-stacks, convince myself I know what the code wants to do and how it does it. Looking at only "unreachable" leaks seems like it should be an improvement for finding the interesting cases. I'll have less time for Git this week, but can try it out as time permits. Thanks for your feedback, both of you. Martin
