What a few responders have said is along the lines that my proposal

1. Can already be done.
2. Has gaps/is too simple.

I will try to address those points here, but before I start, I'm getting the impression I don't have a great grasp of the communication style of this list. I assumed that a simple proposal with _some value_ would meet less resistance than a complex one with more value, so I left out features (like handling arguments). I will try to avoid MVP-ing my proposals in the future, as I think that led to a more "meh" initial response than my idea deserves.


Can calling a function from the command line already be done? Yes, I led with that. But my idea would lead to the following improvements:

1. Being able to call two different functions from the same module. (AND)
2. Being able to call some functions that exist today without modification or rearrangement. (AND)
3. Being able to invoke these things from python without installing a module from pip.

There are solutions that do some of these, but nothing that does them all. I think it's surprising that the python makers have found worthwhile three different solutions for invoking modules, but this list so far has seen little value in invoking a function.


Is my proposal oversimplified? Maybe? I actually think it's pretty nice just to be able to call a couple of arbitrary functions by name that know how to access args via argparse or whatever.

It would definitely be more empowering to be able to describe the command line args to such an invocation using the args to the function.

But, I guess the real weakness of my proposal was that it required anything special about the function at all. It'd be easy to handle some functions with arguments, certainly if they're all strings. Other primitive scalar types may be doable if we map type hints to constructors. But I can't think of a reasonable way to handle a function that expects a complex object for an argument.

So, how about if the proposal expands to include functions with any arity of string-only arguments? I think that makes it possible to use a wide variety of functions without any special work.

We could also say that positional arguments are treated positionally, and kwargs need to be expressed on the command line as --kwarg=value

What else is missing?

On Wed, Oct 21, 2020 at 13:15 Christopher Barker <pythonchb@gmail.com> wrote:
FWIW, one of my intro students just yesterday asked about this:

Is there a way to call a function in a python file directly from the command line?

I have no idea what their programming background is, or why they expected that to be possible, but there you go -- at least one newbie expected it to be possible :-)

However: This was literally class number one -- I had not yet showed them the "if __name__ == '__main__':" idiom yet.

Also, I'm pretty sure this student's expectation was that there would not have to be anything special about the function (except maybe taking only string arguments?)  -- so it's probably not a good idea to try to solve that use case anyway.

But that was the first time I was ever asked that question -- and here it is on python-ideas -- quite a coincidence!


On Tue, Oct 20, 2020 at 11:16 PM Paul Sokolovsky <pmiscml@gmail.com> wrote:

On Wed, 21 Oct 2020 00:24:47 -0400
Michael Smith <michael@smith-li.com> wrote:

> On Tue, Oct 20, 2020 at 23:43 Guido van Rossum <guido@python.org>
> wrote:
> > The ‘if __name__…’ idiom makes one function special. Hm... So if the
> > module has that, will that main() still be called with your
> > proposal? I.e., is __name__ still ‘__main__’?
> > 
> Perhaps not. If __name__ were still '__main__' and calling the
> function implied importing the module, then anything in `if __name__
> == '__main__'` would still get run. I think that would be
> inconvenient for folks.

Alternative valuation is that your proposal is inconvenient to folks,
as it leaves too many gaps. Which, as it seems, you're eager to
bravely patch, but that does/will lead to only more confusion.

> I think __name__ should be the module name in the context of the
> module, and the function name in the context of the function, in this
> case.

__name__ is the name of the module. Or, as a special exception, it's
"__main__" for a module directly invoked as a script or via "-m"

There's no need to complicate it any further (it's already complicated
enough; my bet that half of Python users won't be able to formulate it
like above by heart).

Back to the original proposal, there's no general need to invoke
an arbitrary function in an arbitrary module. That just makes no sense,
as most functions depend on context.

If *you* need that, then, besides other options you mentioned in
your original mail, it's trivial to write a module which will do it for
you, like:

python -m my_super_invocator module:thunk

Actually, if you look on PyPI, I bet you'll find a few like that.

> On Tue, Oct 20, 2020 at 20:32 Michael Smith <michael@smith-li.com>
> wrote:
> > 
> >>
> >>
> >> On Tue, Oct 20, 2020 at 23:12 Guido van Rossum <guido@python.org>
> >> wrote:
> >>> On Tue, Oct 20, 2020 at 8:06 PM Michael Smith
> >>> <michael@smith-li.com> wrote:
> >>> 
> >>>> On Tue, Oct 20, 2020 at 10:19 PM Guido van Rossum
> >>>> <guido@python.org> wrote: 
> >>>> >
> >>>> > I think it would be a tad more convincing if there was a way
> >>>> > to pass 
> >>>> arguments too (even if just a list of strings). At the very
> >>>> least extra arguments should end up in sys.argv[1:].
> >>>>
> >>>> Could python -m 'module:thunk' have exactly the same behavior
> >>>> with respect to arguments as `python3.8 -m module` does today?
> >>>>
> >>>> ```
> >>>> $ cat bar.py
> >>>> import pprint, sys
> >>>>
> >>>> def thunk():
> >>>>     pprint.pprint(sys.argv)
> >>>>
> >>>> if __name__ == "__main__":
> >>>>     thunk()
> >>>>
> >>>> $ python3.8 -m bar -- -1 --two --three=3
> >>>> ['/Users/michael/bar.py', '--', '-1', '--two', '--three=3']
> >>>> ```
> >>>>
> >>>> So then with the same bar.py, `python -m bar:thunk -- -2 --three
> >>>> --four=4` would print `['/Users/michael/bar.py', '--', '-1',
> >>>> '--two', '--three=3']`. I like this better than my previous
> >>>> suggestion to shorthand python -c.
> >>>> 
> >>>
> >>> Actually it should print the same except for sys.argv[0]. (We
> >>> could argue about what sys.argv[0] should be.)
> >>>
> >>> 
> >>>> > Then again, presumably the function must be specially crafted
> >>>> > for 
> >>>> this usage. Why can't you just specially craft a module's main()?
> >>>>
> >>>> I'm not sure I know what you mean by "specially crafted", other
> >>>> than the function only needs not require any formal parameters.
> >>>> It doesn't need to be special-er than that. It can handle args
> >>>> via sys.argv, as you suggested. Most of the `main` functions I
> >>>> write today are just like that.
> >>>> 
> >>>
> >>> Okay, so this is just a way to choose an alternative main()
> >>> function.
> >>
> >> Am I missing something important here? What's special about naming
> >> a function "main"? Don't you still have to do something else to
> >> invoke it from the cli, such as calling it expressly in the module
> >> or pointing console entry points to it?
> >>
> >> ```
> >> $ echo 'def main(): print("hi")' > xyz.py
> >> $ python -m xyz
> >> <no output>
> >> $
> >> ```
> >>
> >> I'm kind of meh at this point, I'll leave it to the usual
> >> crowd. :-) 
> >>> 
> >>
> >> I appreciate the engagement you've given this already. :)
> >>
> >> -- 
> >>> --Guido van Rossum (python.org/~guido)
> >>> *Pronouns: he/him **(why is my pronoun here?)*
> >>> <http://feministing.com/2015/02/03/how-using-they-as-a-singular-pronoun-can-change-the-world/>
> >>> 
> >> -- 
> > --Guido (mobile)
> > 

Best regards,
 Paul                          mailto:pmiscml@gmail.com

Christopher Barker, PhD

Python Language Consulting
  - Teaching
  - Scientific Software Development
  - Desktop GUI and Web Development
  - wxPython, numpy, scipy, Cython