Fix various memleaks discovered by ASAN #3568
Conversation
There was a problem hiding this comment.
Great, thanks!
One suggestion: since the "fix memleak in * test" commits don't have any further description, it may be a bit less noise to combine them into one commit and then maybe say something about the the specific ASAN environment that caught them? The one libjob fix can be its own commit and might do well to use the Problem: format.
Or if that just feels like busywork, I'm actually OK with just setting MWP. Your choice.
chu11
force-pushed
the
issue3566_toss4_asan
branch
from
40ea3d4
to
f1e3ea3
Compare
|
re-pushed per @garlick comments above. |
|
Looks good! Feel freel to set MWP when you're ready. |
Using the configure option --enable-sanitizer=address, many memleaks were found in unit tests. This commit fixes many of them.
Problem: In the event of a invalid input error, there is a chance memory allocated during a json_loads() may leak. Solution: Call json_decref() when returning from an input error.
SteVwonder
force-pushed
the
issue3566_toss4_asan
branch
from
f1e3ea3
to
9ef6008
Compare
|
the bionic builder hung (also did in #3569 .. hopefully not a start of a pattern). restarting it |
|
Hung again. Looks like in this case, everything ran, but In #3569 it looked like it stopped during configure. Hmm, I'll go ahead and restart again. |
Codecov Report
@@ Coverage Diff @@
## master #3568 +/- ##
=======================================
Coverage 82.47% 82.47%
=======================================
Files 323 323
Lines 48770 48771 +1
=======================================
+ Hits 40221 40225 +4
+ Misses 8549 8546 -3
|
So due to a ASAN failure in #3548, I tried to run ASAN on TOSS4 and hit all of these other ASAN discovered memleaks when trying to debug my PR's issue (i.e.
make checkwas constantly failing all over the place, never getting to the unit test failure I hit in CI).I thought these were new ASAN discoveries in TOSS4 over TOSS3, but @grondo reminded me that we turn off leak detection in the CI b/c there are too many and many can't be fixed. But by this point I had already gone down the path of fixing a bunch of these leaks.
I saw no reason to waste these fixes, so here they are and some extras as I continued the exercise. I did find one actual memleak in
libjob, so the exercise wasn't a complete waste of time.