[m-u-xyz]

I was curious whether the commented-out NurseryManager class would actually work, so I turned it on ;-)
I had to insert one conditional because of an impedance mismatch between context managing and cancel handling. All tests pass.

On second thought (second patch), a Nursery can be its own context manager. This can be improved further (third patch) by deferring the initialization to __aenter__(), noting that a Nursery is never used outside of a context manager.

This change speeds up the full test run by a bit more than 1% on my system (tested with 3.5 and 3.6).

[njsmith]

I should note here: the nursery implementation is going to get majorly revisited in 0.3.0 (#136), so I've been putting this off to look at as part of that.

[njsmith]

First pass of simplified nurseries is in #375.

Do you remember why you had to add the if isinstance(exc, Cancelled): return True bit to your Nursery.__aexit__ method? I don't see how it can be correct (doesn't it make nurseries swallow all Cancelled exceptions?), but I assume you ran into trouble without it...

[m-u-xyz]

I'll test my code with the current master branch and report back.

[njsmith]

Thanks!

[kyrias]

Any updates on this? :)

[m-u-xyz]

I'll have a look soon(ish).

[njsmith]

We should probably stop using @asynccontextmanager in the nursery implementation (#229, #393). But this branch is way out of date, discussion has stalled, and I think it goes a bit to far (I like that the actual Nursery object is only returned from __aenter__, not from open_nursery). I'm going to close this for now, and we can always re-open it or steal bits from it later if we want to.