Details
Diff Detail
- Repository
- rL LLVM
Event Timeline
Kostya, do you think we'd better relax the test a bit and use 15K on all platforms? I'd vote for that.
Kostya, do you think we'd better relax the test a bit and use 15K on all platforms? I'd vote for that.
I agree.
In addition, if we find defaults in FreeBSD that are at odds with all other contemporary platforms the FreeBSD team should give consideration to changing those defaults.
test/asan/TestCases/deep_call_stack.cc | ||
---|---|---|
34 ↗ | (On Diff #11579) | Should we have a "middle" value for FreeBSD too? |
Further investigation indicates that the problem is the main thread stack size and not the system stack size.
According to the discussion at:
http://freebsd.1045724.n5.nabble.com/libthr-and-main-thread-stack-size-td5935621.html
the main thread stack size is set to 4M on x86-64 and cannot be changed. For i386 it is half of that value.
What if we reduce the depth of the recursion for all platforms and add 'ulimit -s 4096' to the run lines?
I'm slightly concerned about portability, but it looks like we can use "ulimit -s" on Mac OS. I'll defer to Kostya to decide if we need even less deep recursion, or fine with making this test Posix-only.
test/asan/TestCases/deep_call_stack.cc | ||
---|---|---|
4 ↗ | (On Diff #12301) | please don't modify limits for the other lines. |