Ganden Schaffner

> Can you elaborate on how this is different from what anyio does right now? `get_async_backend` wants to use a cache for performance, so that it has a fast-path that...

> Yes, I can see that this is indeed a preferable solution. Why the `WeakValueDictionary` though? > It can be a normal dict though; the weak-value dictionary seems unnecessary. yeah,...

Closes #282.

Resolves the test regression on Python < 3.3 caused by #268; see #276.

@skorokithakis Maybe I misunderstand, but I don't think that a small fix/feature soon would necessitate a major version bump if this was merged. Under SemVer, dropping support for old versions...

Sorry for the delayed response! Following @BvB93's comment above, I think the underlying problem here is how #268 passes around kwargs. Schema is trying to pass `validate`'s kwargs to the...

On second thought: a major version bump is *technically* not necessary under SemVer even if this is merged, since Schema is still in major version 0. Schema still being v0.x.x...

did you mean to include that `assert` in your reproducer? I would expect `assert receive.receive_nowait() == "item"`, if reachable, to raise `WouldBlock`: * `receive_nowait()` can only receive an item from...

if I remove that assertion from your reproducer, then I believe your reproducer demonstrates at least one bug, but possibly two: 1. the bug of AnyIO dropping the item in...

> Ah, my intention was to create a memory object stream with a buffer. Then the `assert` makes sense, yes? yes, agreed, assuming that your intention is to test for...