[WIP] Suspend capturing when stdin is being read (optionally)

[blueyed]

No description provided.

[asottile]

I'd rather raise a helpful error message telling the user to use -s instead of magically suspending capture.

The capture code is already some of the more complex parts of pytest and adding more edges seems the wrong direction there

[blueyed]

I'd rather raise a helpful error message telling the user to use -s instead of magically suspending capture.

Good point about a more helpful error message (I've created #5086 for it).

(You should see the code that I have to make this work with runpytest_subprocess to make it a real TTY (using threads to display and capture/forward streams).. ;))

But seriously, passing -s does not wotk with testdir.runpytest (both methods (inline and subprocess)).
What do you suggest here?
I usually end up re-running the tests from the created tempdir directly, but that is rather tedious.

As for runpytest_inprocess, it could maybe just not capture anything if -s is in args (or better have a kwarg for this). This way you could use capture=False to not use capturing for this test, but it would make the following assertions against stdout/stderr fail then - for this to work they could be duplicated maybe (or handled via threads, to be cross platform).

By now it works better to pass through stdin, which could serve as an indicator, but would still need to not capture (or duplicate) stdout/stderr then.

The capture code is already some of the more complex parts of pytest and adding more edges seems the wrong direction there

Agreed in general, but not being able to easily debug pytest's own tests does not help to improve this situation.
And most of this patch is to keep the current behavior - it would be easier if only the new behavior would have been supported in the first place, with maybe an option then to close it explicitly.

(just some quick thoughts - happy to iterate from here)

[asottile]

But seriously, passing -s does not wotk with testdir.runpytest (both methods (inline and subprocess)).
What do you suggest here?
I usually end up re-running the tests from the created tempdir directly, but that is rather tedious.

Maybe borrow from the subprocess py3 interface and have an input=b'...'?

pytest is currently lacking a good "capsys but for stdin", which (at least in my code) has led to some wacky hacks to patch out input / sys.stdin.read / etc. -- maybe the answer is to expose something like that, maybe even added into the current capturing system

(spitballing)

def test_stdin(capsys):
    capsys.feed_input(b'foo\n')
    assert input() == 'foo'


def test_stdin_not_fed(capsys):
    with pytest.raises(...) as excinfo:
        assert input() == 'foo'
    msg, = excinfo.value.args
    assert msg == 'idk some useful message about input but no provided input'

[nicoddemus]

Hi @blueyed,

This has been on WIP for a long time now, and has some conflicts. I'm closing this for now to clear up our PR queue a bit, but please do re-open this when you have the time.

Cheers!

[blueyed]

@nicoddemus
btw: if you do another round of closing stale PRs it would be helpful to attach some label - so e.g. it is easier to find them again.

[nicoddemus]

Good idea, done. 👍