A fast heap memory profiler for Linux.
Details
Sep 26 2021
PRIORITY: should this be unsigned instead of int?
Sep 25 2021
May 17 2018
fixed by using an updated libbacktrace
Mar 8 2018
code review is taking place on github for this change request: https://github.com/KDE/heaptrack/pull/12
Feb 28 2018
Jan 21 2018
I see from your recent commits that you've figured most things out here.
Jan 17 2018
I've adapted and pushed this now. Many thanks Ivan! Much appreciated.
Aha, -DTAKE_ADDR fails when I use ld.bfd instead of ld.gold, I assume you do that too? I still don't see a difference with using -Wl,-z,now though...
I fail to reproduce the issue with the provided test case on ArchLinux. When I compile the three variants, I always properly detect the free calls.
Thanks, I'll take over then. What email address should I use to attribute (parts of) the patch to?
Jan 15 2018
I was curious enough to figure out what compiler behavior causes the different type of relocation entries, so here's a test program with extensive comments. I was originally seeing the behavior with libcairo, but this should make testing easy.
I'm not much of a C++ guy, so take it from here, Milian, if this patch still needs work. :)
Dec 20 2017
hey, thanks for the contribution! do you have a code snippet that showed the behavior we could use for testing purposes?
Dec 8 2017
Aug 31 2017
abandoning since https://phabricator.kde.org/D7621 got landed
taking over to close
thanks, I'll fix this up and merge this
Added link to bug 383889.
Aug 30 2017
I opened https://phabricator.kde.org/D7621 which includes the requested changes.
Aug 27 2017
Please also add a link to https://bugs.kde.org/show_bug.cgi?id=383889
Jun 27 2017
May 2 2017
Apr 5 2017
A WIP for this can be found in wip/buildid, it will include the buildid in the heaptrack file. But for interpreting, it isn't enough to just load the debug file only. It has to be used in addition apparently, which isn't possible without patching libbacktrace like is done in vogl.
Jan 22 2017
some ideas on how to do it:
need to add configuration for the editor preferences, similar to what gammaray does
Jun 29 2016
additionally, we could try to use seccomp(-bpf) to get notified about mmap syscalls via SIGSYS, but I'm not sure whether we get access to the syscall args from the interrupt handler
there's now a public heaptrack_api.h for pool allocators.
Apr 6 2016
Apr 5 2016
Mar 24 2016
implemented now
Mar 23 2016
implemented now
Mar 15 2016
implemented now with
Mar 11 2016
Mar 9 2016
On a 32bit machine:
Mar 5 2016
We should look at rr, and how they monkey-patch the vdso to trace all syscalls. That could be used then to track I/O and futex stuff. It would also be useful for heaptrack itself to trace mmap/munmap/sbrk to check heapspace vs. the "real" requested memory by the allocator.
Mar 1 2016
Also an option: http://diamon.org/ctf/