Documentation

Compiled extensions that "want to be interruptible by user requests" need to call PyErr_CheckSignals periodically; when it returns -1, they should stop work, clean up after themselves, and return to the interpreter.

This function is not mentioned at all in the extending and embedding guide, and its reference documentation makes it sound unimportant. As a result, many extension authors are (it seems to me) not aware that it exists or that they ought to be calling it, and so many extensions that should be interruptible by user requests aren't. For example, if you execute this code

>>> import numpy as np
>>> rng = np.random.default_rng()
>>> x = rng.random(1000000000)

and then immediately hit control-C, you'll be waiting at least a few seconds before you get the prompt back.

PyErr_CheckSignals also has the unusual property of potentially executing arbitrary Python code before returning to its caller, which makes it easy to use unsafely. Making an extension interruptible throughout may necessitate calling PyErr_CheckSignals from innermost loops where one would ideally release the GIL, which makes that worse (see #133465).

Thus, I suggest that a guide to correct use of PyErr_CheckSignals should be added to the extending and embedding guide.

Linked PRs

• gh-134075: Document how to make C extensions interruptible #159571