Design specification for what a post based repodata fetching might look like

[mariusvniekerk]

One feature that we could theoretically look at would be to modify the behavior of conda so that instead of it just doing a GET to <CHANNEL_URL>/repodata.json.bz2 it instead does a post to the same url with an payload of the packages already present in the target environment

{
   'packages': [
      'python 3.7.1', ... 
   ]  
}

This would allow us to dynamically be able to generate the constained repodata without needing to resort to fairly gross url path based approaches.

[dhirschfeld]

without needing to resort to fairly gross url path based approaches

Wouldn't url parameters be slightly less gross (assuming conda just passed them through)

<HOSTNAME>?channels=conda-forge,defaults&packages=pandas,ipython,scikit-learn

...though I think POST is probably a cleaner solution still.

[scopatz]

That's the problem, conda doesn't just pass them through because it appends things like 'repodata.json' to to URL you provide

[mariusvniekerk]

The cleanest short term api is a modification to conda that allows us to pass through this information as request headers