Make proper executable with entry_point

[bollwyvl]

I'd recommend the entry_point approach, as it is generally preferable to using python -m if possible, as -m can have... unwanted side-effects. Further, a conda downstream will do a good job of making it robust on different platforms.

[martinRenou]

I did use the entry_point approach: https://github.com/martinRenou/json-lsp/blob/master/setup.py#L20-L24

Should I do it in another way?

[bollwyvl]

sorry, yes, the lsp entry_point is 👍

I mean for the thing-what-runs, e.g. https://python-packaging.readthedocs.io/en/latest/command-line-scripts.html

setup(
    ...
    entry_points = {
        'console_scripts': ['json-lsp-or-whatever=json_lsp.main:main'],
    }
    ...
)

[martinRenou]

I see. So I would need two entry points, one for the console script (that actually runs the json-lsp binary) and one for lsp (that provides the spec)?

[bollwyvl]

Or go to flit
https://flit.readthedocs.io/en/latest/pyproject_toml.html#scripts-section

...and get reproducible wheels, too 🎡

[bollwyvl]

right. then you can either use the shell helper instead of the python one... or just return the dictionary, and not have a dependency on on jupyter_lsp at all... that was for us to reduce boilerplate. we did the thing for pyls for... reasons. probably windows. don't remember now.

[martinRenou]

I didn't know about flit! Thanks for the pointer