Test isopen on response_stream

[omus]

Follow up to #467 and #466. The original test for issue #466 used eof which could cause the tests to hang if the code was configured incorrectly. The tests have been revised to use isopen which doesn't have this issue.

I further improved upon the tests by checking the behaviour of using "response_stream" and "return_stream". As it turns out it has revealed a bug with the DownloadsBackend which I won't be fixing at this time (#471)

Finally, I also duplicated this test suite for MinIO as this allows this behaviour to be tested locally and not just on the CI. I can remove this if desired.

[omus]

Note I've tested these changes locally using HTTP 0.9.14 and 0.9.15

[omus]

cc: @nickrobinson251

[omus]

bors try

[bors]

try

Build succeeded:

• Julia 1 - ubuntu-latest - x64
• Julia 1.5 - ubuntu-latest - x64
• Julia 1.3 - ubuntu-latest - x64
• Julia 1 - macOS-latest - x64

[nickrobinson251]

Changing to isopen seems a good idea.

I don't understand the behaviour of different backends -- i know you said there's a bug in Downloads.jl-backend, but from the code itself i don't follow what the expected behaviour is and which bits are broken/buggy. It looks like we expect different behaviour from the different backends, but that is surprising to me, so i think warrants a comment with a link to an issue explaining

[nickrobinson251]

Why do different backends have different behaviour?

[omus]

They really shouldn't. I think this all comes the fact that HTTPBackend returns a Base.BufferStream by default while DownloadsBackend returns an IOBuffer by default. For this PR I'm just recording the current behaviour and avoiding changing it because as we know closing/not closing a file can be a breaking behaviour

[omus]

I'm going to address these inconsistencies with #384

[nickrobinson251]

thanks, i think the changes here would benefit from a comment explaining that they are testing the current behaviour and also linking to the issue you intend to open about possibly changing this behaviour (at first i thought we were testing we had "buggy" behaviour, which seemed an odd thing to test, but now i understand)

The changes themselves all LGTM, i was just struggling to follow what it is that these tests are documenting

[nickrobinson251]

it's expected to be test_broken above, but to work here -- why is that?

[omus]

This is the reason for the difference in behaviour. Calling read_body may or may not close the IO:

AWS.jl/src/utilities/downloads_backend.jl

Lines 35 to 39 in 1722aa5

	read_body(x::IOBuffer) = take!(x)
	function read_body(x::IO)
	close(x)
	return read(x)
	end

Whether this function is called is dependent on a couple of factors:

AWS.jl/src/utilities/downloads_backend.jl

Lines 70 to 74 in 1722aa5

	body_arg = if request.request_method == "HEAD" || request.return_stream
	() -> NamedTuple()
	else
	() -> (; body=read_body(request.response_stream))
	end

[omus]

I'll squash and merge this PR as this doesn't need to be more than a single commit