un/registerJsModule fails various expectations about pyodide HOT 2 OPEN

Thanks for opening the issue @WebReflection!

There was a similar discussion in pyodide/micropip#51, so the conclusion in this issue might also be applied to that one.

It's not very common to install modules at runtime, and even less common to uninstall them, so it's hard to say which way is better. It is also complicated by the fact that in Python, if an imported module is already in use, deleting it from sys.modules does not necessarily affect the module in use...

I am +1 with removing the module object from sys.modules when a module is re-registered / un-registered. This won't solve all of the problems, but it's somewhat close to "expected" behavior IMO.

from pyodide.

I am +1 with removing the module object from sys.modules when a module is re-registered / un-registered. This won't solve all of the problems, but it's somewhat close to "expected" behavior IMO.

yeah, expectations is all I am after ... I understand if stuff was previously referenced and never let go the FinalizationRegistry won't get ever a chance to free those resources but I really believe once an explicit unregisterJsModule happens something must be different in further executions and, when a module is re-registered, something gotta happen ... either silently update the sys.modules cache or throw an error because already used in the wild ... I am all for replacing it with the new updated references that any further script can consume, thanks for considering the issue and proposing your suggested changes!

from pyodide.