-
-
Notifications
You must be signed in to change notification settings - Fork 5.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Win32 segfault in numbers test with precompiled sys.dll #8895
Comments
Tests pass on Linux 32 bit, so I believe this is a Win32 problem. |
Thanks Jake. Deleting |
Tests still pass with |
This might not be in the numbers test at all - if I add a bunch of This is probably another "hope it goes away when we upgrade LLVM" issue. |
Doesn't happen on my Linux nightlies either, FWIW. |
We're seeing segfaults on Win32 with JuMP also: https://ci.appveyor.com/project/mlubin/jump-jl/build/1.0.592/job/g3m4lk4d48lfxxr5 |
@mlubin interesting, but probably unrelated as the Windows binaries shouldn't be including sys.dll right now |
…ues are solved (#8895, among others)
…ues are solved (#8895, among others)
this is strange, since the sys.dll file should be fine to include for win32 |
I'll note that I haven't had this happen as repeatably for the last week or so. Not sure what changed, or if that's conclusive enough to close the issue as the underlying cause might still be present? |
its always possible that it got fixed along the way, but seems somewhat unlikely. it'll be hard to fix if it isn't reproducible however – sometimes it's easier to just close the issue and wait to see if it comes up again |
It should be reproducible at any of the commits that are listed above as |
…ues are solved (JuliaLang#8895, among others)
This is back on master at 84612d4, but now with a backtrace thanks to @vtjnash's awesome recent work.
I see signs of gc and utf8proc in there? |
Looks like a missing gc root. The sys.dll thing is probably just a red herring, caused by altering the allocation/gc order |
although, it also looks like a corrupted stack. i wonder if backtraces would be even better if we disabled FPO ( |
I think the fixes for #9366 may have solved this. I'll reopen if I see it come back. |
Looks like it was caused by the call overloading merge (#8712):
This may be a Win32-specific problem, would need someone with 32 bit Linux to test and confirm.
Here's the bisect log, the skips were from failures to compile or otherwise different-looking failures, all from within the same call-overload PR I think.
The text was updated successfully, but these errors were encountered: