Drop timing-sensitive sha1 count assertion#63
Conversation
With Python 3.14, Reader and HasherPool threads start before collect() is called. For 8-byte pieces all 1000 hashes land in hash_queue before the main thread calls hash_queue.get(), so sha1_mock.call_count always equals t.pieces regardless of whether cancellation works
|
Thank you for the fixes. Do you need another release? And if so, should I hold off on making a new release? Or are you fairly |
No worries, happy to help! Is there any reason why this PR was closed and not merged? Without this change, the tests on Debian CI infra fail
It would be wonderful to see a new release! That way, I could easily include the patches directly in Debian and have everything packaged more smoothly |
|
Is there any reason why this PR was closed and not merged?
It's listed here for me: https://github.com/rndusr/torf/commits/master/
Can you confirm?
I don't know what happened there. Usually when I `git pull ...` and `git push ...`, GitHub marks the
PR automatically as merged, but not for this one. I'm sure it's my fault, but I have no idea what I
did wrong.
It would be wonderful to see a new release! That way, I could easily include the patches directly in Debian and have everything packaged more smoothly
I can make a bugfix release to make your life easier. I just don't want to make another release in
two weeks with another minor patch. So I'd like to know if you have any other fixes lined up.
The safest bet for me would be to wait and see if you find anything else, but I don't know how
annoying that is for you.
|
|
@rndusr oh, I see that the commit is there. Thanks for accepting the change! No need to release it now, I’ll also try to package torf-cli this or next week and will see if there are any new bugs with this program. But thank you for your collaboration, I really appreciate that! |
With Python 3.14, Reader and HasherPool threads start before collect() is called. For 8-byte pieces all 1000 hashes land in hash_queue before the main thread calls hash_queue.get(), so sha1_mock.call_count always equals t.pieces regardless of whether cancellation works
The error initially appeared during the CI autopkgtest run for the torf package in Debian on amd64 and arm64