Restarting a reactor should raise an exception

[TimothyFitz]

About once a month someone asks twisted-python why they have really strange behaviour, and happen to be restarting the reactor. Restarting the reactor is unsupported (currently, by all implementors, nothing prevents it from being implemented) so these all fail, but always fail poorly. A call to reactor.run() after reactor.stop() should raise an exception that clearly says "This reactor does not support restarting" or "Reactors do not support restarting".

[TimothyFitz]

Glyph Lefkowitz

On Thu, 7 Sep 2006 15:01:08 -0400, James Y Knight <foom@…> wrote:

However, even though it'd be possible to fix, perhaps it is better to simply throw an exception on the second reactor.run(), since most people who do that shouldn't actually be running the reactor twice, they just didn't know the better thing to do yet.

I think that's a good idea. Based on the nature of the questions that I've seen, I >think that this would probably be a good idea even if we *did* have a restartable >reactor. Add an interface, IRestartableReactor, and require people who want to use >it to call 'reactor.restart()'. While this might not actually *do* anything, the >conspicuous absence of it from any example code and a potentially useful error >message from reactor.run would prevent the question from being asked repeatedly...

This bug will cover making reactor.run() raise on a second call. Someone else can pick up IRestartableReactor and the like.

[mishok13]

First attempt at the actual fix

[mishok13]

The test is basically a rewrite of test_multipleRun and the patch introduces new instance variable (which sucks).

[lvh]

(In [31008]) Branching to 'reactor-restart-raise-2066'

[lvh]

(In [31009]) Apply patch, refs #2066

[lvh]

Looks great! Merging.

[lvh]

(Okay, that's actually a lie: it's missing a topfile, but I'm going to add that for you)

[lvh]

(In [31011]) Raise when you try to restart an unrestartable reactor

Author: mishok13 Reviewer: lvh Fixes: #2066

Unrestartable reactors not complaining was confusing. When you try to do it, you get an exception now, which is a much nicer failure mode.