[Twisted-Python] If the errbacks of a canceled Deferred are called with error other than CancelledError, is this acceptable?

[exarkun at twistedmatrix.com]

On 10:26 am, terry at jon.es wrote:
>Hi Tom
>>Since most apis in twisted don't pass a canceller, the behavior is 
>>a2+a3.
>
>Right. That was the intention (glyph's). It was expected that over time
>more Twisted apis that create deferreds would add cancellers. As I
>understand it, Kai Zhang is now adding cancellers to some apis 
>returning
>deferreds, starting with the POP3 client.
>
>Based on JP's comment, I still think I'm missing something here.
>
>The only suggestion I've been making is that, if we wanted to, we could
>allow a value to be passed to the cancel call. That would allow errback
>functions to receive more information when a Deferred is canceled. 
>You'd
>still get a CancelledError in any case, but the value would be in the
>exception instance as args[0], as in

An idea very much like this was discussed when Deferred cancellation was 
being implemented.  My recollection is that it was ultimately explicitly 
rejected, though I don't remember why nor do I find discussion of it on 
the ticket (#990) (lack of this information somewhere seems like a 
failure of the development process).  I admit the idea of being able to 
get additional information passed through the system seems appealing to 
me, but *not* having any additional information does simplify the 
interface.  Considering how much trouble people already have with 
Deferreds without even bringing cancellation into play, maybe the 
simpler interface is a good thing.

My earlier point was that any Deferred for which cancellation is not 
already explicitly implemented already has a cancellation behavior: it 
will fire its errback chain with `CancelledError`.  It is an 
incompatible change to replace this with a different exception type - 
such as `ConnectionDone`.

This is slightly different than your suggestion, Terry, I think - in 
that you propose sticking more information onto the `CancelledError`, 
not changing the type of the `Failure` that is sent down the errback 
chain.  However, the thread was started specifically with the question 
of whether changing `CancelledError` to `ConnectionDone` is acceptable 
and that's the point I was mainly focused on.

Jean-Paul