Thanks for merging #740.
Could you publish a release, including PyPI wheels, containing the socket cancellation / FD double-close fix in commit dc680eb20c526d8babacaf4751a12f51431ea246?
As of September 22, 2026, the latest release on both GitHub and PyPI is still 0.22.1, which does not contain this fix. A normal package upgrade therefore cannot pick it up yet.
The existing upstream report and patch describe a correctness issue: when socket adoption through create_connection(sock=...) or create_unix_connection(sock=...) is cancelled or raises, the Python socket can retain a descriptor already closed by libuv. If that descriptor is reused, subsequent cleanup can close an unrelated connection. The merged patch detaches the socket on the error path and includes TCP and Unix socket regression tests.
A published release would let downstream applications adopt the fix without maintaining custom builds or switching event-loop implementations.
Would a patch release containing #740 be feasible? If a release is already planned, an approximate timeline would be helpful.
Thank you for maintaining uvloop.
Thanks for merging #740.
Could you publish a release, including PyPI wheels, containing the socket cancellation / FD double-close fix in commit dc680eb20c526d8babacaf4751a12f51431ea246?
As of September 22, 2026, the latest release on both GitHub and PyPI is still
0.22.1, which does not contain this fix. A normal package upgrade therefore cannot pick it up yet.The existing upstream report and patch describe a correctness issue: when socket adoption through
create_connection(sock=...)orcreate_unix_connection(sock=...)is cancelled or raises, the Python socket can retain a descriptor already closed by libuv. If that descriptor is reused, subsequent cleanup can close an unrelated connection. The merged patch detaches the socket on the error path and includes TCP and Unix socket regression tests.A published release would let downstream applications adopt the fix without maintaining custom builds or switching event-loop implementations.
Would a patch release containing #740 be feasible? If a release is already planned, an approximate timeline would be helpful.
Thank you for maintaining uvloop.