Merge lp://qastaging/~allenap/maas/log-power-on-a-bit-later into lp://qastaging/~maas-committers/maas/trunk

The change itself makes sense, but I'm lost with the test.

Before I go into my main question with the test: when injecting unhandled exceptions, I'd recommend creating an ad-hoc exception class over just picking an existing exception class. Otherwise the reader is left wondering if your choice of exception has any significance. For a while I was wondering why you seemed to be saying that division was difficult, and I still don't see what you mean by “This doesn't get any easier”!

But now the main thing. How does the test establish that power_change_starting is called “first”? I don't see anything to prove that anything else doesn't happen before the exception. That must have been the purpose of the cheating that your comment mentions, so did you forget a check?

> The change itself makes sense, but I'm lost with the test.
>
> Before I go into my main question with the test: when injecting
> unhandled exceptions, I'd recommend creating an ad-hoc exception class
> over just picking an existing exception class. Otherwise the reader is
> left wondering if your choice of exception has any significance. For a
> while I was wondering why you seemed to be saying that division was
> difficult, and I still don't see what you mean by “This doesn't get
> any easier”!

Okay, good point, done.

>
> But now the main thing. How does the test establish that
> power_change_starting is called “first”? I don't see anything to prove
> that anything else doesn't happen before the exception. That must have
> been the purpose of the cheating that your comment mentions, so did
> you forget a check?

That it's called first isn't /that/ interesting, just that it's called
early on. I've added some comments to explain what the intent is, and
I've changed the name of the test. The sentinel argument values lend
some weight to "happens early on" because they would probably cause
failures later on in the change_power_state(). The comments also explain
that the exception "cheat" also avoids doing set-up for later parts of
change_power_state(), again lending weight to "happens early on".

Thanks!

Wonderful, thanks.