Serhiy, regarding _getresp() calling resp.decode(): You are probably right that a failure decoding the response line is recoverable in some circumstances, such a GROUP command. But in other cases, say an ARTICLE command, the response won’t have been fully drained, so the connection is no longer usable. I guess I could shift the error handling to a higher level if you think it is necessary.

Regarding the workaround in my original post: It is not a good solution, because it is still trying to send a QUIT request and wait for a response in the error handler. If you are trying to interrupt a hanging network connection for example, you have to hit Ctrl+C twice, or wait a long time for the QUIT command to get some invalid data and raise its confusing protocol error.

David: The error handling inside _getlongresp() is the main place that concerned me. But as I made the changes there I identified the other places I thought could raise unexpected exceptions (network errors) or block (therefore a KeyboardInterrupt is likely).

Draining the response might be sensible for handling an error writing to the file in some cases. My /dev/full example was actually meant to be a simple example of more complicated exceptions, such as KeyboardInterrupt, network timeouts, or if the programmer wants to peek at the start of a large message and then abort the download. So it seems my simplified test case was misleading, because in these other cases, draining the response isn’t so appropriate.