Support -m flag

[jacobsvante]

So one can run the equivalent of e.g. python -m alembic.config revision --autogenerate

[gotcha]

I do not understand what you mean. Can you gibe an example with ipdb ?

[jacobsvante]

I just mean that it would be nice if ipdb could support the -m flag just like Python. Nothing more nothing less.

$ python --help
[...]
-m mod : run library module as a script (terminates option list)
[...]

The part I mentioned in the previous message is just an example where I call a module named alembic.config (which itself takes the argument revision and the flag --autogenerate). You can run the command yourself if you just install alembic with pip. Doing this won't work though:

ipdb -m alembic.config revision --autogenerate

It gives me Error: -m does not exist.

[gotcha]

It is supposed to work without -m as explained when running ipdb -h

usage: ipdb.py scriptfile [arg] ...

Have you tried ?

[jacobsvante]

Yeah and that won't work with packages that use relative imports (from . import db). The error thrown is ValueError: Attempted relative import in non-package.

Also it's really nice to not have to know the entire path if it's a 3rd party script buried somewhere on my file system in site-packages. I.e. instead of doing:

ipdb /my/path/to/venv/lib/python2.7/site-packages/mypackage/mymodule.py

It would be nice to be able to do:

ipdb -m mypackage.mymodule

[wodow]

+1 to this.

If I am already running a module by python -m mypackage.mymodule then being able to immediately replace python with ipdb on the command line to rerun with the debugger enabled would be very useful indeed.

[kprussing]

The issue is that the __package__ variable is not getting properly set to None per PEP 366. It appears like this is specifically a Python 3 bug because the problem also appears in pdb but only in Python 3. I've opened an issue here.

As a work around, you can either use Python 2 or simply inject __package__ = None at the top of your script while debugging.

[gotcha]

@kprussing If I understand well what you say, this is not an ipdb bug but a Python 3 bug. Correct ?

[kprussing]

Partially. After reviewing the discussion, I realize that my comment is a bit off topic. I ran into a similar issue trying to debug a script that uses relative imports. According to PEP 366, the loader is supposed to set the __package__ variable to None when a script is directly executed. Thus, when run as python script.py, __package__ is supposed to be None; whereas, it is properly set when run as python -m script. The Python 3 version of pdb does not do this. Rather, it sets __package__ to the empty string. I haven't fully reviewed your source, but I suspect this is affecting the relative import in your case too. The minimum working example I attached to the linked issue above shows this behavior in both pdb and ipdb.

I like the idea of supporting the -m flag, but if the erroneous loader behavior is present in Python, you would have to manually work around it. I opened the issue on the Python project because it seems to me it should be handled at the underlying source.