On 21 April 2011 10:23, Andrew Bennetts <email address hidden> wrote:
> My only concern is that by registering a signal handler we may be
> inviting the return of some EINTR bugs? Hopefully if we're just going
> to cleanup and exit the process anyway that's not a big deal, but we
> should at least keep an eye out for that sort of bug reports after this
> lands.
That is a concern. More generally I think it's a reason not to sink
too much effort into sophisticated cleanup from signals: first of all
the ssh connection may be dead already, and secondly handling them
safely and reliably in Python is difficult. Not only eintr things,
but also cases like the toomanyconcurrentrequests from trying to clean
up objects that are already in use.
On 21 April 2011 10:23, Andrew Bennetts <email address hidden> wrote:
> My only concern is that by registering a signal handler we may be
> inviting the return of some EINTR bugs? Hopefully if we're just going
> to cleanup and exit the process anyway that's not a big deal, but we
> should at least keep an eye out for that sort of bug reports after this
> lands.
That is a concern. More generally I think it's a reason not to sink ntrequests from trying to clean
too much effort into sophisticated cleanup from signals: first of all
the ssh connection may be dead already, and secondly handling them
safely and reliably in Python is difficult. Not only eintr things,
but also cases like the toomanyconcurre
up objects that are already in use.