bug#43682: 28.0.50; Clean up nnimap server buffers?

[Eric Abrahamsen]

On 10/11/21 05:53 AM, Stefan Kangas wrote:
> Eric Abrahamsen <eric@ericabrahamsen.net> writes:
>
>> On 10/01/20 18:01 PM, Lars Ingebrigtsen wrote:
>>> Eric Abrahamsen <eric@ericabrahamsen.net> writes:
>>>
>>>> It's possible (I'm not claiming to understand all the code) that all we
>>>> would need to do is fix `gnus-async-wait-for-article' to replace its
>>>> calls to `nntp-find-connection' and `nntp-accept-process-output' with
>>>> something generalized. Those two functions deal with directly with
>>>> `nntp-connection-alist', so we'd need something that would do the
>>>> equivalent with `nnimap-connection-alist'.
>>>
>>> Yup.
>>
>> This is something I wouldn't want to tackle until we have generic
>> functions.
>>
>>>> Anyway, in the interest of completing this far less ambitious patch: if
>>>> the nnimap connection has timed out, we should remove this connection
>>>> from `nnimap-connection-alist', so this version of the patch does that.
>>>> If async has opened a second connection, I guess we should leave that
>>>> alone, though I don't have too much confidence that the whole process
>>>> will recover gracefully from the main connection dying...
>>>
>>> Well, the connections are separate, and there's all kinds of reasons for
>>> the server to close a connection, so...
>>>
>>>> +      (unless (memq (process-status (get-buffer-process buffer))
>>>> +                    '(open run))
>>>
>>> Aka `process-live-p'.
>>
>> I forgot we have that!
>>
>>> Otherwise looks fine to me (but I haven't tested the code).
>>
>> Okay, I'll run this for a bit, as well.
>
> Any news here?  Should the fix be installed?

Okay, I've pushed a commit that fixes nearly all of the issue. There are
still some mysterious circumstances under which a single dead imap
connection buffer remains in `nnimap-process-buffers', and I don't know
how it gets there. But it's better than 50+ dead connection buffers, so
I'm going to close this for now.