Fix: When using Nuitka to package a project, an error occurred when reading the file: "KeyError: 'self'" #12923

[CrazySJX]

Fixes #12923

The reason the current code cannot correctly retrieve the calling parameters is that the inspect module does not store complete frame information for each normal call by default; it only populates the call stack information when an exception occurs. In other words, the frame information that this code relies on may not be complete, which could lead to the inability to retrieve the parameters.

[welcome]

Hello! 👋 Thanks for opening your first pull request here! ❤️ We will try to get back to you soon. 🚴

[larsoner]

Rather than adding this to every module before calling super().__init__(...), is there any way this magic could live inside super.__init__(...) itself instead? You mention:

[inspect] only populates the call stack information when an exception occurs

So in theory it should be possible to use inspect.currentframe().f_back to get back to this level and do something there, maybe?

Also not sure why you do __import__("inspect") rather than just import inspect at the top of the file...?

[drammock]

the inspect module does not store complete frame information for each normal call by default

is there a reference / explanation for this? I saw it mentioned in the nuitka issue thread but it's not immediately obvious to me from reading the docs for the inspect module. I do see this note:

CPython implementation detail: This function relies on Python stack frame support in the interpreter, which isn’t guaranteed to exist in all implementations of Python. If running in an implementation without Python stack frame support this function returns None.

which I think is referring to the problem you encountered, but it would be nice to have that confirmed / have a fuller explanation of what is going wrong.

[larsoner]

Also a side note -- IIRC the raw._init_kwargs stack frame stuff is a feature used (99% sure!) only by mne-bids, and it's private so we can change it without warning (to anyone other than mne-bids, since I think we have an implicit contract with them to support it). So we could make MNE-Python more robust to incorrect or incomplete information being available based on the Python implementation being used, i.e., make the _get_argvalues() return None when the init kwargs can't be resolved.

In other words, maybe we don't want to try super hard to make this work in all cases? Like if we just set raw._init_kwargs = None when it couldn't be resolved, to hit a problem in practice, a user would have to be using nuitka and MNE-BIDS to write data in a specific way that makes use of _init_kwargs. I think this is a very deep corner case. And to make it work we're jumping through more hoops that we'll need to support. Might not be worth the effort and maintenance burden of whatever solution we come up with...