Add safe_close helper to standardize socket cleanup in tests

[denini08]

Hi! First of all, thanks for the great work on this project.

While looking at the socket-based tests, I noticed that sockets are being closed directly with close(). In most cases this works fine, but the recommended pattern when dealing with sockets is to first call shutdown() and then close(). This allows the connection to be terminated more cleanly, ensuring that both read and write channels are properly closed before releasing the resource.

According to the Python documentation:

shutdown() should be called before close() to signal that no more data will be sent or received, and to allow the other side to handle the connection termination properly.
https://docs.python.org/3/library/socket.html#socket.socket.shutdown

In tests, failures caused by not doing this are rare, but when they happen they are usually hard to diagnose (for example, resource warnings, hanging sockets, or platform-dependent behavior). Using a small helper like safe_close makes the cleanup more explicit and consistent across the test code.

This PR introduces a safe_close function that:

• Tries to call shutdown() first
• Falls back gracefully if the socket is already closed or in an invalid state
• Always attempts to call close()

The main goal here is not to fix a concrete bug, but to improve the cleanup pattern and make it more robust and standardized for current and future tests. It also makes the intent clearer: sockets are always closed in a safe and predictable way. This was identified during an ongoing research project.

[njsmith]

I don't think it's true that shutdown first is better than simply closing? There are patterns where it could make a difference like doing a shutdown and then waiting for the other side to respond with a close before finishing the close, if you have unpredictable duplex traffic + a protocol with no GOAWAY message type. I can't think of when that scenario would come up but it probably has for somebody at some point. But afaict a shutdown immediately followed by a close doesn't solve anything or even behave differently in any way than a simple close.

…

On Sun, Jan 25, 2026, 06:56 Denini Gabriel ***@***.***> wrote: Hi! First of all, thanks for the great work on this project. While looking at the socket-based tests, I noticed that sockets are being closed directly with close(). In most cases this works fine, but the recommended pattern when dealing with sockets is to first call shutdown() and then close(). This allows the connection to be terminated more cleanly, ensuring that both read and write channels are properly closed before releasing the resource. According to the Python documentation: shutdown() should be called before close() to signal that no more data will be sent or received, and to allow the other side to handle the connection termination properly. https://docs.python.org/3/library/socket.html#socket.socket.shutdown In tests, failures caused by not doing this are rare, but when they happen they are usually hard to diagnose (for example, resource warnings, hanging sockets, or platform-dependent behavior). Using a small helper like safe_close makes the cleanup more explicit and consistent across the test code. This PR introduces a safe_close function that: - Tries to call shutdown(SHUT_RDWR) first - Falls back gracefully if the socket is already closed or in an invalid state - Always attempts to call close() The main goal here is not to fix a concrete bug, but to improve the cleanup pattern and make it more robust and standardized for current and future tests. It also makes the intent clearer: sockets are always closed in a safe and predictable way. This was identified during an ongoing research project. ------------------------------ You can view, comment on, or merge this pull request online at: #698 Commit Summary - 36da767 <36da767> Add safe_close function for socket cleanup in tests File Changes (1 file <https://github.com/python-trio/trustme/pull/698/files>) - *M* tests/test_trustme.py <https://github.com/python-trio/trustme/pull/698/files#diff-7f31a89310e2fd6438037acd7f0e5e1eadd05c8155193f69fd1d4aad05b1b375> (35) Patch Links: - https://github.com/python-trio/trustme/pull/698.patch - https://github.com/python-trio/trustme/pull/698.diff — Reply to this email directly, view it on GitHub <#698>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEU42CXTP2EHDSN3VAOXE34ITKQFAVCNFSM6AAAAACS2MLZE6VHI2DSMVQWIX3LMV43ASLTON2WKOZTHA2TGMZWGYZDCOI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[njsmith]

Oic the Python docs are talking about the case where you've dup'ed the fd so there are multiple references to the underlying socket. In that case close only decrements the refcount and doesn't affect any of the other references, and shutdown can be used to immediately force the network state transition that normally doesn't happen until the last reference is closed. This is a pretty exotic situation or thing to want (arguably if you ever get into this state then you should rethink your design) and I can't see how it would every be relevant to our tests?

…

On Sun, Jan 25, 2026, 06:56 Denini Gabriel ***@***.***> wrote: Hi! First of all, thanks for the great work on this project. While looking at the socket-based tests, I noticed that sockets are being closed directly with close(). In most cases this works fine, but the recommended pattern when dealing with sockets is to first call shutdown() and then close(). This allows the connection to be terminated more cleanly, ensuring that both read and write channels are properly closed before releasing the resource. According to the Python documentation: shutdown() should be called before close() to signal that no more data will be sent or received, and to allow the other side to handle the connection termination properly. https://docs.python.org/3/library/socket.html#socket.socket.shutdown In tests, failures caused by not doing this are rare, but when they happen they are usually hard to diagnose (for example, resource warnings, hanging sockets, or platform-dependent behavior). Using a small helper like safe_close makes the cleanup more explicit and consistent across the test code. This PR introduces a safe_close function that: - Tries to call shutdown(SHUT_RDWR) first - Falls back gracefully if the socket is already closed or in an invalid state - Always attempts to call close() The main goal here is not to fix a concrete bug, but to improve the cleanup pattern and make it more robust and standardized for current and future tests. It also makes the intent clearer: sockets are always closed in a safe and predictable way. This was identified during an ongoing research project. ------------------------------ You can view, comment on, or merge this pull request online at: #698 Commit Summary - 36da767 <36da767> Add safe_close function for socket cleanup in tests File Changes (1 file <https://github.com/python-trio/trustme/pull/698/files>) - *M* tests/test_trustme.py <https://github.com/python-trio/trustme/pull/698/files#diff-7f31a89310e2fd6438037acd7f0e5e1eadd05c8155193f69fd1d4aad05b1b375> (35) Patch Links: - https://github.com/python-trio/trustme/pull/698.patch - https://github.com/python-trio/trustme/pull/698.diff — Reply to this email directly, view it on GitHub <#698>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEU42CXTP2EHDSN3VAOXE34ITKQFAVCNFSM6AAAAACS2MLZE6VHI2DSMVQWIX3LMV43ASLTON2WKOZTHA2TGMZWGYZDCOI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>