[python-committers] Can we choose between mailing list and discuss.python.org?

[Paul Moore]

On Tue, 12 Feb 2019 at 22:00, Antoine Pitrou <antoine at python.org> wrote:
> Here is a 161-message Discourse thread (at the time of this writing):
> https://discuss.python.org/t/pep-517-backend-bootstrapping/789

As someone directly involved in that discussion, with a strong need to
understand all of the points being made, that's a great example of
both the benefits and the flaws of the discourse model.

> I know I can browse easily through a 161-message mailing-list or
> newsgroup thread using a traditional threaded view, read what I want,
> come back later to read the rest, etc.  But Discourse's linear
> presentation pretty much kills that ability.  It doesn't even allow
> *seeing* the structure of the discussion.

I don't use a threaded mail client (I use gmail's web interface) so I
don't get any of the benefits of threading from a mailing list. So to
that extent, Discourse's lack of threading is no different for me, and
shouldn't affect my ability to follow the discussion. But it *does*,
and in practice, it's substantially worse than a traditional mailing
list. (Note: this is only a comment about long, complex discussions
like this one, for shorter threads the Discourse view is fine).

The problem isn't, IMO, so much the lack of threading as the lack of
*context*. We're all used to (and frustrated by) mailing list threads
that are 90% quoted text. But Discourse goes to the other extreme, of
having very *little* context - no thread structure, a tendency towards
minimal quoting, and an *extremely* non-obvious "reply" UI (you can
"reply" to any message, or to the thread as a whole, but the
distinction is almost invisible, and doesn't support "replying to"
*part* of a long comment.

Also, the lack of any "mark unread" functionality makes it easy to
lose track of where you're up to - I popped into that discussion to
check some facts for this post, and found myself needing to read a
number of quite detailed messages, as otherwise they would no longer
show as "unread" for me, and I'd risk losing my place in the
discussion. I know there are bookmarks, but they don't match my mental
model which is "I saw these posts, but haven't *read* them".

Anyway, I remain generally happy with Discourse for lower-traffic
lists that have relatively short threads. Medium sized ones (like
packaging replacing distutils-sig) I'm not certain about yet, but I
think "probably no worse" is as far as I'd go right now. For groups
like python-dev or (worse still) python-ideas I feel like they would
be a terrible fit. There's also the interaction effect - high traffic
in one category pushes out information about what's new in *other*
categories, and there's no "list of categories with a count of unread
messages" view to mitigate it.

tl;dr; I don't think discourse scales particularly well to long,
complex discussions, but I think it's less about threading than about
other aspects of the UI. At the end of the day, managing long, complex
discussions is *hard* and I think Discourse is optimised for different
parts of the spectrum than mailing lists. But while the day to day
volume of traffic might be shorter threads, the massive, complex,
rambling threads are the lifeblood of Python development (much as we
might all hate them ;-)) and we need to be cautious about making
decisions for those cases based on evidence from other, simpler,
situations.

Paul