feedparser and timeoutsocket

[Sam Ruby]

Jeff Waugh wrote:
> Hey,
> 
> While running trunk on Planet GNOME and my personal planet, I've discovered
> that the timeoutsocket code in feedparser has been removed [1] (at least it
> was after a long history of changes of it being disabled, enabled, removed,
> added back, and then finally, apparently removed for good).
> 
> This has resulted in a few things:
> 
>  * I'm getting way more 500 errors than before, which you'd expect
> 
>  * Planet takes even longer to run
> 
>  * Every now and then, the process appears to get stuck, though I've yet to
>    figure out why (the Python-happy gdb backtrace wasn't entirely helpful to
>    me, perhaps others might find it informative [2])
> 
> So - thoughts? Should we badger the feedparser people to put it back in (as
> an option, maybe)? Should we set it up ourselves in Planet? Are there good
> reasons to not use timeoutsocket at all?

Looking at the code, my thoughts are that this involves a Python
release-dependent policy decision that globally affects all sockets and
there is no logging done.  As such, I don't believe that it belongs in a
library function.

Since people were happy with the way it was, I've implemented it in my
repository, with the defaults exactly the way they were.  With logging.

And at this point, I might as well fess up.  When I said I thought I was
done now, I lied.  I've been having too much fun (most of the other
changes in my repository are Atom specific)

http://www.intertwingly.net/blog/2006/05/08/Affordances-and-Accomodations

http://planet.intertwingly.net/bzr/

http://planet.intertwingly.net/planet.log

- Sam Ruby