On Tue, Feb 7, 2017 at 11:29 AM, Glyph Lefkowitz <glyph@twistedmatrix.com> wrote:

On Feb 7, 2017, at 6:59 AM, Jean-Paul Calderone <exarkun@twistedmatrix.com> wrote:

On Fri, Jan 13, 2017 at 1:13 AM, Glyph Lefkowitz <glyph@twistedmatrix.com> wrote:
Thanks for highlighting those.  I've put the link in the other direction as well.

Craig seems eager to go ahead with reverting this change in behavior.

As far as I can tell, no one has weighed in on the other side.  So I'm inclined to go along with the reversion.

My 2¢ for the other side is: if trial does this, but twist and twistd don't, then it will be possible to get a passing test run for a plugin that doesn't get loaded.  I think it would be simpler and easier to debug to leave these consistent.

This is an interesting corner case, but I think the twistd and twist issues should be pursued in
separate discussions and tickets.

For trial, I would like to proceed with https://github.com/twisted/twisted/pull/672/ .  With my conversion
of trial to a console script, the new behavior was unintentional on my part.  Since all the unit tests
passed, I did not notice.  In this ticket: https://twistedmatrix.com/trac/ticket/8978
Job-Evers-Meltzer provided a use-case where the new behavior of trial
broke existing usage.

trial can be used as a general-purpose tool for running unittest-style tests,
much like pytest or nose.  The new behavior is disruptive, for minimal benefit,
so I would like to restore the old behavior.

I added a unit test for this, so in future, if this breaks again, we will catch it.