[issue34795] loop.sock_recv failure because of delayed callback handling

Change by Andrew Svetlov : -- assignee: -> asvetlov ___ Python tracker ___ ___ Python-bugs-list mailing list Unsubscribe:

[issue34795] loop.sock_recv failure because of delayed callback handling

Andrew Svetlov added the comment: Looks like a hack but we have no choice -- ___ Python tracker ___ ___ Python-bugs-list mailing

[issue34795] loop.sock_recv failure because of delayed callback handling

Yury Selivanov added the comment: yes -- ___ Python tracker ___ ___ Python-bugs-list mailing list Unsubscribe:

[issue34795] loop.sock_recv failure because of delayed callback handling

Andrew Svetlov added the comment: Yuri, do you mean `socket._io_refs` manipulations? -- ___ Python tracker ___ ___

[issue34795] loop.sock_recv failure because of delayed callback handling

Neil Booth added the comment: My library user reports: I can't reproduce the issue with uvloop on linux. (looks like uvloop does not work on windows atm) -- ___ Python tracker

[issue34795] loop.sock_recv failure because of delayed callback handling

Neil Booth added the comment: This seems related: https://bugs.python.org/issue30064 -- ___ Python tracker ___ ___

[issue34795] loop.sock_recv failure because of delayed callback handling

Yury Selivanov added the comment: Can you reproduce this with uvloop? Andrew: I think we need to "lock" sockets from closing in sock_recv and friends the same way we do it in uvloop. -- ___ Python tracker

[issue34795] loop.sock_recv failure because of delayed callback handling

New submission from Neil Booth : In certain circumstances the coroutine loop.sock_recv() registers a callback internally, which is called on e.g. task cancellation. The callback assumes a file descriptor that was open and valid at the time the callback was registered is still open and valid