I couldn't find answer after having read all the following:
- PEP 338 Executing modules as scripts
- documentation of
runpy
standard module
- description of Python interpreter's
-m
option
Rationale:
When a test script which uses relative imports is being run without -m
option I could print a warning message instead of leaving user with standard traceback leading to ValueError: Attempted relative import in non-package
exception. Without knowing this I can catch this exception and only suggest lack of -m
option could be the reason of error.
Another observation is that __package__
is set to None
when executing the script directly and to the package name when using -m
(using the empty string when the module isn't included in any package, so it's still different from None
).
Disclaimer: this is just an observation, I have not seen it in the docs so it is probably implementation dependent and might not be consistent across different Python versions.
I have noticed that when calling a script using a -m
option a variable called __loader__
is added to the namespace, so at the top of your script you could check for existence of that variable:
if '__loader__' in globals():
# called with -m
For some extra safety you could check to see if __loader__
is an instance of pkgutil.ImpLoader
:
if '__loader__' in globals() and __loader__.__class__.__name__ == 'ImpLoader':