On 30/09/2010 15:53, Phil Mayers wrote:
So, I appear to be back to the case where I can either gracefully handle the exception *or* gracefully handle the errback, but not both?
It should be possible if using Twisted APIs correctly to reliably capture errors.
If it is not, that is a Twisted bug (or possibly API limitation)
However, your "test_looping.py" is not using the API correctly; the reactor.callLater throws away the call stack, so the error has nowhere to go.
Right, but that's what appears to be happening with the "real code", and I get the mysterious GeneratorExit...
Since you're not using reactor.callLater in your real code, the example is not valid. Hopefully I'm being more clear here!
Yes, but I do feel a bit like I'm banging my head against a wall too. The effect is the same as the real live observed problem, only I can't replicate the real problem without an ftp server that only happens to accept active connections. I don't have one of those around, and *even* if I did, I want to make the scheduler bulletproof, not *just* go and fix the one error that has cropped up so far... cheers, Chris