Integrate with conda-forge

[bernhard-42]

Add jupyter-cadquery to conda-forge

[bernhard-42]

see some details can be found in #34 with @bollwyvl

[bollwyvl]

Some things while digging around:

• the hard (e.g. ==) pins in install_requires are pretty hard for downstream maintainers over time
  • a bottom pin is always great, but at least having a top pin, which is reviewed prior to release, helps give some wiggle room, as you can at least lobby the upstream for a point release with a backport-in-hand of e.g. security fixes
  • using tox to test against a range of python/runtime versions is the gold standard e.g.
    • the oldest viable environment
    • the highest viable environment
    • an entirely unpinned version (potentially with --pre enabled) to help detect upcoming issues
• having the jupyterlab pin in both places makes things a little more fragile
  • potentially just have the -widgets package carry that dependency
• looks like the nbextensions are now handled by the jupyter-cadquery-widget package, but are still listed in data_files
• not having a cadquery version pin seems like it will be problematic over time, even if it's not possible to install from pypi
  • since it's not installable from pypi, one can at least put it behind an extras_require so that the version required/tested is at least documented somewhere.
    • this can be used by downstreams to test, though not through "normal means" like pip check (which ignores extras anyway)

Again, still no closer to having you your upstreams on conda-forge... the test cycle is quite slow, and I'm somewhat shooting in the dark on some of these things.

[bernhard-42]

the hard (e.g. ==) pins in install_requires are pretty hard for downstream maintainers over time

Happy to change
"webcolors==1.11.1"
"sidecar==0.5.0"
to
"webcolors~=1.11"
"sidecar~=0.5"
if that helps

having the jupyterlab pin in both places makes things a little more fragile

Agreed, I can remove it from juypter-cadquery

looks like the nbextensions are now handled by the jupyter-cadquery-widget package, but are still listed in data_files

That's a bug, will remove it

not having a cadquery version pin seems like it will be problematic over time

Happy to add cadquery 2.1 to extra requires. I didn't need it up to now since it is not on pypi as you said

If I do these changes, would I then need to deploy both packages with newer versions (2.0.3 for jupyter-cadquery-widgets and 2.0.1 for jupyter-cadquery) or would it be sufficient to have the changes in master?

[bollwyvl]

since i can't package your stuff anyway yet, no need to publish!

[shimwell]

I notice that the jupyter-cadquery v3 has a different dependency stack, is it worth taking another look conda-forge deployment

[bernhard-42]

@shimwell What would be your requirements?
jupyter-cadquery plays well with conda as long as it is provided in the pip section of the environment yaml file.

One of the biggest issues I have is that CadQuery doesn't have versions or tags. In order to participate on features like Sketch I have to recommend to install cadquery=master. How to you solve this iswsue?

[shimwell]

I am able to install into a conda environment nicely

However I am quite keen to build a conda package that makes use of jupyter-cadquery and I think that when making a conda package all dependencies must also be available on conda. Therefore I can't include jupyter-cadquery in the package I am currently making. The meta.yml for my package is here and you can see I have jupyter-cadquery commented out.

So currently I have a two part install process which is conda install my package (this brings in cadquery and others) then pip install jupyter_cadquery.

I guess on the cadquery front that lack of tag will be solved once cadquery 2.2 is released. I wonder if in theory jupyter-cadquery could be published to either conda-forge channel, cadquery channel or another channel?

I have a package on conda that makes use of the master branch of cadquery which appears to bring in the version tagged as master

[bernhard-42]

I have never built a conda package, so I first need to understand how to do so.
And I need to build conda-forge packages for all three sub projects I guess. I need to check the process ...

I guess on the cadquery front that lack of tag will be solved once cadquery 2.2 is released. I wonder if in theory jupyter-cadquery could be published to either conda-forge channel, cadquery channel or another channel?

btw. my problem around cadquery=master is for the docker container. I artificially need to change something else in my Dockerfile in order to pick up the latest CadQuery master. In general I would prefer to pin the version to a version I have tested it on. But, no tags exist ...

[bernhard-42]

I will also take inspiration from your conda packages and meta.yaml files ...

[shimwell]

I recall there was a reason that CadQuery2 is in a dedicated channel and noticed this issue that might be of interest

I think conda-forge packages need all their dependencies to be on conda forge so perhaps I should ask this question over on cadquery repo first

[bernhard-42]

Good point.
Could you please add me to your question there? I would like to follow and understand whether conda-forge would work. Thanks!

[bernhard-42]

Since cadquery is available in pypi now, this is not necessary any more from my point of view