get_class_from_modulename mishandles ImportErrors unrelated to its scope

[whatisgalen]

this method:

def get_class_from_modulename(modulename, classname, extension_type: ExtensionType):
    mod_path = modulename.replace(".py", "")
    module = None
    import_success = False
    import_error = None
    for directory in get_directories(extension_type):
        try:
            module = importlib.import_module(directory + ".%s" % mod_path)
        except ImportError as e:
            import_error = e
            continue

captures an ImportError that would come up if the module wasn't actually present in the directory. However, it inadvertently also handles for, say, an ImportError that would arise from the module's own logic.

For example, in my_module.py:

from arches.app.models.nonexistent import NonexistentClass

class thing(NonexistentClass):
    ...

this logic on its own would raise an ImportError but the trace gets effectively overridden when the module is imported somewhere else by what the logic assumes is a totally unrelated error of the module not being there in the first place.

[whatisgalen]

@jacobtylerwalls idk why but this might be of interest to you

[jacobtylerwalls]

Good catch @whatisgalen and good spidy sense -- in other projects I've used from importlib.util import find_spec for exactly this kind of thing, just to sniff whether the module exists without importing. Here we could do that to finish the search loop, then with the right module in hand do the import.