Hello,

after a recent system update, I noticed that Memory-annotated functions defined within a Jupyter notebook
would be called after kernel restarts even though the arguments passed were the same across runs. This has worked without problems in the months before.

After some investigation, I think I pinpointed the underlying issue to be a possible naming scheme change in IPython that is not caught by the check in line 130 of func_inspect.py. If I am not mistaken, this function expects to see the name of the source file containing the function to contain <ipython-input. Instead, on my machine the structure of the source file path is /tmp/ipykernel_XXXXXXX/YYYYYYYYYY.py where the X are digits that change slightly on kernel restart and the Y seem to be a stable identifier of the source (at least I have not seen a variance there so far). Consequently, the check in parts[-1] does not work any more to condense a run-agnostic source identifier.

I have fixed the issue on my machine by inserting another case in get_func_name (see my fork). If wanted, I can create a PR, although I have limited resources at the moment to check the fix on other machines (or in non-Jupyter use cases).

Best
Malte

PS: I work on an Arch linux system and the Jupyter notebook gives the following version information:

Python 3.9.6
IPython 7.26.0