TST Occasional failures of test_keyboard_interrupt

[rth]

I have seen this test fail at least once in the CI and once locally. Rerunning the test fixes it, but it does suggest that it's maybe not very stable.

___________________________________________________________________________________ test_keyboard_interrupt[firefox] ____________________________________________________________________________________

selenium = <conftest.FirefoxWrapper object at 0x7f6963c4f7f0>

    def test_keyboard_interrupt(selenium):
>       assert selenium.run_js(
            """
            x = new Int8Array(1)
            pyodide._module.setInterruptBuffer(x)
            window.triggerKeyboardInterrupt = function(){
                x[0] = 2;
            }
            try {
                pyodide.runPython(`
                    from js import triggerKeyboardInterrupt
                    x = 0
                    while True:
                        x += 1
                        if x == 2000:
                            triggerKeyboardInterrupt()
                `)
            } catch(e){}
            return pyodide.runPython(`
                2000 < x < 2500
            `)
            """
        )
E       AssertionError: assert None
E        +  where None = <bound method SeleniumWrapper.run_js of <conftest.FirefoxWrapper object at 0x7f6963c4f7f0>>('\n        x = new Int8Array(1)\n        pyodide._module.setInterruptBuffer(x)\n        window.triggerKeyboardInterrup...       `)\n        } catch(e){}\n        return pyodide.runPython(`\n            2000 < x < 2500\n        `)\n        ')
E        +    where <bound method SeleniumWrapper.run_js of <conftest.FirefoxWrapper object at 0x7f6963c4f7f0>> = <conftest.FirefoxWrapper object at 0x7f6963c4f7f0>.run_js

src/tests/test_pyodide.py:198: AssertionError

On chrome it produces,

E       AssertionError: assert {}
E        +  where {} = <bound method SeleniumWrapper.run_js of <conftest.ChromeWrapper object at 0x7fa3564e7e80>>('\n        x = new Int8Array(1)\n        pyodide._module.setInterruptBuffer(x)\n        window.triggerKeyboardInterrup...       `)\n        } catch(e){}\n        return pyodide.runPython(`\n            2000 < x < 2500\n        `)\n        ')
E        +    where <bound method SeleniumWrapper.run_js of <conftest.ChromeWrapper object at 0x7fa3564e7e80>> = <conftest.ChromeWrapper object at 0x7fa3564e7e80>.run_js

src/tests/test_pyodide.py:198: AssertionError
--------------------------------------------------------------------------------------- Captured stdout teardown ----------------------------------------------------------------------------------------
Python exception:
Traceback (most recent call last):
  File "<exec>", line 6, in eval_code
  File "/lib/python3.8/site-packages/pyodide/_base.py", line 302, in eval_code
    return CodeRunner(
  File "/lib/python3.8/site-packages/pyodide/_base.py", line 225, in run
    exec(mod, self.globals, self.locals)
  File "<exec>", line 5, in <module>
KeyboardInterrupt

cc @hoodmane

[rth]

Actually I can consistently reproduce it with,

pytest src/tests/test_pyodide.py -k firefox

using both chrome or firefox. However when it's run separately,

pytest src/tests/test_pyodide.py -k "firefox and test_keyboard_interrupt"

it doesn't fail. So there is likely some bad interaction with other tests in test_pyodide.py. Which also explains why it occasionally fails in CI, as the run order is not fully deterministic with pytest-xdist.

[hoodmane]

Thanks for pointing this out @rth!

likely some bad interaction with other tests

Weird, well that's great that it reproduces. Hopefully I'll get the same effect, I'll check it out and try to see if I can fix it.

[hoodmane]

This seems to have been caused by #1167 somehow.

[hoodmane]

In particular, it doesn't seem to be a problem with the test or the test suite, the type conversions logic is actually wrong now. Confusing...

[hoodmane]

Oh wait! It is a problem with the tests: test_monkeypatch_eval_code doesn't undo the monkey patch. It's actually weird that this hasn't shown up until now.