Re: Gnus fetch freezes emacs

[Eric Abrahamsen <eric@ericabrahamsen.net>]

On 07/05/23 10:04 AM, Stephen Berman wrote:
> On Tue, 04 Jul 2023 20:50:05 -0700 Eric Abrahamsen <eric@ericabrahamsen.net> wrote:
>
>> Stephen Berman <stephen.berman@gmx.net> writes:
>>
>>> On Tue, 04 Jul 2023 10:02:34 -0700 Eric Abrahamsen <eric@ericabrahamsen.net> wrote:
>>>
>>>> Stephen Berman <stephen.berman@gmx.net> writes:
>>
>> [...]
>>
>>> This:
>>>
>>>  (defun srb-gnus-group-get-new-news (&optional arg one-level)
>>>    (interactive "P")
>>>    (with-timeout (1 (kill-buffer (nntp-find-connection-buffer nntp-server-buffer))
>>>  		   (gnus-group-get-new-news))
>>>      (gnus-group-get-new-news arg one-level)))
>>>
>>>  (define-key gnus-group-mode-map "g" 'srb-gnus-group-get-new-news)
>>>
>>>>                                      Eric F is just describing the
>>>> unfortunate behavior of nntp-connection-timeout, which interrupts the
>>>> entire fetching process when it hits the timeout.
>>>
>>> Is that different than what the above function does with the kill-buffer
>>> sexp?  (Not a rhetorical question, I know next to nothing about news
>>> servers and their connectivity issues.)
>>
>> The `nntp-connection-timeout' variable has different behavior in that
>> NNTP servers are allowed one "retry" if the connection fails. The code
>> around that is very confusing to me (which is why my earlier fix was
>> buggy).
>
> I don't follow you, but no need to elaborate further here.
>
>>>>>> Yeah, I'd put in a dumb fix for this that turned out to be buggy, so we
>>>>>> just recently reverted it. I have a more thorough fix in progress
>>>>>> somewhere here, that would report a server connection failure without
>>>>>> interrupting the rest of the servers, but it's not done yet. I've had
>>>>>> very little time for coding recently, but will get to it At Some Point.
>>>>>>
>>>>>> Glad it's at least better than it was. I wonder if we should have some
>>>>>> generous timeout set by default...
>>>>>
>>>>> It might make sense to continue this discussion in bug#52735.
>>>>
>>>> This doesn't seem like the same issue -- this problem is pretty well
>>>> understood.
>>>
>>> Hm, I had understood from both Prashant Tak and Eric Fraga that the
>>> problem they have is essentially the same as I do and what I reported in
>>> that bug.  But that problem doesn't seem to be understood.  If by the
>>> understood problem you mean the effect of nntp-connection-timeout,
>>> doesn't that just mean using it isn't a real fix for the hang the three
>>> of us (at least) are experiencing?  That's why I thought other
>>> approaches need to be considered and bug#52735 seems like the
>>> appropriate venue for that.  But I'm fine with continuing the discussion
>>> here instead.
>>
>> Oh I see what you mean. In your bug report I'd gotten the idea that
>> something was going wrong with accepting process output, and had a
>> missed-the-forest-for-the-trees moment around it simply being a dead
>> process.
>>
>> Using `nntp-connection-timeout' is the proper fix for this problem, it's
>> just got a bit of unfortunate behavior that needs to be remedied. I'd be
>> inclined to start a whole new bug report for a fix for that, because
>> it's really a new issue, with its own larger-reaching design decisions.
>> I suppose we could merge #52735 with that, though.
>
> Feel free to open a new bug for fixing nntp-connection-timeout.  I don't
> know if I can help, other than trying out suggestions and providing
> feedback.  In the meantime I'll keep using my workaround replacement
> function.
>
> But I wonder, could this issue have been triggered by some change in
> news.gmane.io around early to mid December 2021?  Because that's when
> the problem start for me, and prior to that I don't recall ever having
> this problem (perhaps sporadically but not with such persistance).

The nntp-connection-timeout variable has been present and nil since
1999. I put my original buggy fix in at the end of October 2021, so that
seems suspicious, but that should only have had an effect if you had set
nntp-connection-timeout to something other than the default of nil.

In general it's pretty hard to say, though...