bpo-33608: Factor out a private, per-interpreter _Py_AddPendingCall(). #12360
Conversation
|
@vstinner, actually let me fix the deadlock in test_capi first. I'll let you know when I'm ready. :) |
|
Note for myself: the first time that this change has been merged, it caused issues at Python finalization, causing buildbot failures. https://bugs.python.org/issue33608#msg337116 |
ericsnowcurrently
force-pushed
the
pending-calls-unrevert
branch
from
376d9da to
b5fdc67
Compare
ericsnowcurrently
force-pushed
the
pending-calls-unrevert
branch
from
b5fdc67 to
48a23ef
Compare
|
FYI, I've fixed the problem. :) @vstinner, could you try this branch against that VM you set up? |
|
I plan on merging this on Friday (in 2 days). If we still see the same buildbot failures (or @vstinner still sees the problem in his VM), we can go from there. If a revert (again) is necessary, it won't be nearly as disruptive since this PR is much more focused. However, I don't expect any problems at this point. Since the revert I've re-merged most of the original changes (as noted above, in the PR summary), with no apparent negative impact. I've also landed several fixes (e.g. to finalization, to locks after fork), found while working on the un-revert, which likely prevent the buildbot failures for which the revert happened. Furthermore, this PR includes several fixes (as noted above) that likewise probably help with those revert-motivating buildbot failures. Consequently I'm confident that the underlying issue has been resolved. |
|
I have a bad news for you Eric: I'm able again to reproduce the crash at commit f13c5c8: https://bugs.python.org/issue33608#msg340069 |
|
That's too bad. I'll revert then. |
|
Thanks for checking that. |
|
@vstinner, it isn't failing without this branch, right? |
…ngCall(). (pythongh-12360)" This reverts commit f13c5c8.
I ran my test using 2 terminals for 8 minutes: I failed to reproduce the crash, whereas with your change I got a crash way faster. |
This is effectively an un-revert of #11617 and #12024 (reverted in #12159). Portions of those were merged in other PRs (with lower risk) and this represents the remainder. Note that I found 3 different bugs in the original PRs and have fixed them here.
@vstinner, would you mind running this against that VM you set up, to see if it causes the original problem? I expect it won't.
https://bugs.python.org/issue33608