Should we allow users to upload files for use in data and analysis scripts?

[WardBrian]

This could re-use the machinery added for #239

Pros:

• Make it possible to provide data that is not already JSON (e.g. CSV) but also does not live on the web
• Make it possible (though clunky) to do some kinds of model comparison by uploading CSV files generated by a different model/run into the analysis for this model/run

Cons:

• A naive implementation of this would have this be completely local, which would mean that a download or gist would no longer perfectly re-create the environment of the user who clicked Save -- some other set up would be necessary by the user
• Alternatively, we could include these arbitrary files in the data model, which would let them be save-able, but could dramatically complicate things.

This would definitely move the project further along the "a full IDE is what you really want" axis, which can be both good and bad

[magland]

I feel like once we start allowing uploads of arbitrary files to the project then as you say, we are headed toward the full IDE approach. I wonder if there should be two modes: (1) the basic one that we have now which is limited in scope (2) an experimental mode where you have an arbitrary collection of files. Maybe we can provide an export from (1) to (2).

I realize (2) is more ambitious, but it's what I had in mind at the start. As we've said, I think what we have serves an important purpose, but somehow we need to figure out how to appropriately limit its scope.

[yizhang-yiz]

Does it make sense to allow uploading but limit file size?

[WardBrian]

@charlesm93 brought up this idea to me again today, specifically to allow LOO in the analysis state.

I think for the analysis scripts we can dodge the sharable question, because already the user needs to intervene before running an analysis script, but I'm still not sure