http: only close connector if it is available

[isidentical]

Stuff like asyncio-retry offer their own clients which proxy the original ClientSession, and they do not bind stuff like _connector directly. Since this is a private-attribute, it might be nicer if we access it optionally and only try to close if we can find it.

[martindurant]

For py 3.6, I think you need loop.run_until_complete

[isidentical]

Ah, right. I'll just skip these under 3.6 (as we do in test_asyn, https://github.com/intake/filesystem_spec/blob/e9bd63f795d84c35898bef3223fc7e055d49c774/fsspec/tests/test_async.py#L56)

[martindurant]

That's another solution :)

[martindurant]

I am running the tests again, but the windows async failure is a little worrying, even if it was a one-off.

[isidentical]

I recall seeing this error before too. Would you mind explaining what test_async_other_thread does?

[martindurant]

what test_async_other_thread does

The filesystem's loop is running in another explicit thread, and we manipulate the loop object directly to get it to do things (i.e., not using async/await). This is actually quite similar in effect to specifying no loop/asyncronous flags at all and calling the blocking functions.

[martindurant]

Perhaps try changing

loop = asyncio.get_event_loop()
to
loop = asyncio.new_event_loop()

(don't know why this should show up just now...)

[isidentical]

The only thing that I guess cause this to show up now is the order of tests, let me temporarily move the test to the end and see if it passes or not.

[martindurant]

OK, well it passed - we can revisit if it happens again.