PR published as draft for review of general approach and review of usage/ergonomics of CLI interface.

TODO:

• CLI code to be polished - description is missing etc.
• Unittests to be implemented.

In general I haven't personally been keen to push cython_freeze type functionality too far. Mainly because I think it has limited use without being able to bundle external non-Cython modules. And supporting that is a huge task that I don't think Cython should be taking on. (I actually think the whole --embed feature was probably a mistake because it gives the expectation that building a "single application" works when mostly it really doesn't)

So I don't know. This extra addition is fairly minor, but I'm not sure if it's a direction Cython really wants to go.

Ignoring that: what commands should I be using here? Is it cython lcmath.pyx first to build lcmath then your command with --embed-modules?

Commands (files are from Demos: https://github.com/cython/cython/tree/master/Demos/freeze):

cython --embed --embed-modules=lcmath combinatorics.pyx # creates combinatorics.c
cython lcmath.pyx # creates lcmath.c

gcc -c combinatorics.c ... -o combinatorics.o # compile combinatorics
gcc -c lcmath.c ... -o lcmath.o # compile lcmath
gcc lcmath.o combinatorics.o -o combinatorics #link them together statically

In general I haven't personally been keen to push cython_freeze type functionality too far. Mainly because I think it has limited use without being able to bundle external non-Cython modules. And supporting that is a huge task that I don't think Cython should be taking on.

Here is my humble point of view:

• definitely bundling all dependencies is totally out of scope of cython
• embedding can be interesting for someone who does not want distribute source code (EDIT: but this can be done by compiling all modules and having stupit main python script)
• there is pretty popular question in stackoverflow regarding embed: https://stackoverflow.com/questions/22507592/making-an-executable-in-cython
• Currently in cython there is half-baked support of embedding - there is no easy way to compile it (e.g. cythonize or setuptools), poorly documented etc... I think it should be decided to make it usable or remove it as out of scope for cython.

(I actually think the whole --embed feature was probably a mistake because it gives the expectation that building a "single application" works when mostly it really doesn't)

I personally think that this is the result of current state of documentation of this feature. This can be easily fixed by documenting --embed properly and adding one warning bar there.

I've seen people use --embed on the ML from time to time, so I'd not consider it unhelpful as it is.
I concur that it shouldn't cover more complex use cases, though.
I think we can leave the feature in. It doesn't hurt and seems to have a purpose. I also don't think we keep others from building something better by having the simple feature included.

To clarify, I definitely wasn't suggesting removing any existing features.

This can be easily fixed by documenting --embed properly and adding one warning bar there.

I did add a bit more details on the limitations to the docs a few months ago (76b22ac). So hopefully that's a little better

there is pretty popular question in stackoverflow regarding embed [...]

And the fact it's popular is exactly why I don't like it. Lots of people want to use Cython to create applications, but we can't actually do what they want (often).

So the question here is: is this PR a small enough extension to the existing feature that we can justify adding it. My current view is that I don't think it is (but I don't feel strongly).

I think this is a tiny enough feature to just include it. Some users might find it helpful.

As mentioned, documentation and test are missing.

I don't think we need to remove cython_freeze.py right now. Just leave it where it is until we have a reason to remove it.