capsys input disable

[georgeharker]

I have a baseline image diff test, for which I define an interactive mode, allowing diffs to be accepted or declined one by one. This works with X11 for the diff frontend and I block on the validation assert until input accepts or not (when in interactive mode).

I want to do a terminal version of this, but capsys.disable() doesn't pass through the in flag to suspend_capture here

pytest/src/_pytest/capture.py

Line 668 in 958981c

def suspend_capturing(self, in_: bool = False) -> None:

The intervening call to suspend_global_capture on multicapture here

pytest/src/_pytest/capture.py

Line 788 in 958981c

def suspend_global_capture(self, in_: bool = False) -> None:

does deal with input.

I think it would be as easy as modifying CaptureFixture to pass the boolean for input here

pytest/src/_pytest/capture.py

Line 987 in 958981c

def disabled(self) -> Generator[None]:

and wire it through here

pytest/src/_pytest/capture.py

Line 841 in 958981c

def global_and_fixture_disabled(self) -> Generator[None]:

with sensible current-behavior defaults.

I appreciate reading from stdin isn't a great idea for automated tests - this is intended for interactive maintenance of baseline image diffs only.

Meanwhile I can do the utterly horrible

from _pytest.capture import CaptureManager

        # ...
        capmanager: CaptureManager = capsys.request.config.pluginmanager.getplugin(
            "capturemanager"
        )
        capmanager.suspend(in_=True)
        # ...
        capmanager.resume()

Any thoughts?

[GLec1er]

For a dedicated interactive maintenance run, a public-API workaround is to add -s to that invocation and keep using the explicit capsys fixture:

pytest -s path/to/test_images.py

Keep your existing opt-in interactive flag too, if you have one. The distinction is that -s disables global capture (including the stdin blocker), but it does not disable an explicitly requested capture fixture. You can still use capsys.disabled() just around the prompt:

def test_prompt(capsys):
    print("captured before")
    assert capsys.readouterr().out == "captured before\n"

    with capsys.disabled():
        answer = input("Accept diff? [y/N] ")

    assert answer.lower() == "y"
    print("captured after")
    assert capsys.readouterr().out == "captured after\n"

I checked this with pytest 8.4.1 and 9.1.1: without -s, the prompt reproduces the stdin-capture OSError; with -s, it reads the supplied answer and the fixture captures output before and after the disabled block again. The capture documentation also notes that capture fixtures take precedence over -s.

This avoids reaching into _pytest.capture for that workflow. It does change global capture for the whole interactive invocation; it does not add a per-block stdin flag to capsys.disabled(). If keeping global capture enabled throughout is a requirement, that API question remains separate.

Both the failing and working invocations were checked locally.