Arvados

So here's the history of this ticket:

I wasn't really sure what it meant to "refresh" the R SDK. I started working on this ticket when both Peter and Tom were out, so I couldn't ask questions. I couldn't find any relevant documentation. I did find #13076, which describes how autoGenAPI.R generates code from the discovery document. I go and look and yeah, it looks relevant. It looks a lot like discovery2pydoc.py, except it generates implementation in addition to documentation. I run it. Spot-checking a couple method signatures it generates, yeah, this definitely seems like what I want.

I kept working on it. The changes from #19929 revealed that the script wasn't prepared to handle more involved descriptions in the discovery documentation, so I dealt with those. I rigged things up so that the R SDK could be kept up-to-date with the API automatically, rather than requiring a manual run and committing artifacts to Git. I fixed up some documentation test failures too—although I didn't really understand what changes I had made (either to the script or the discovery document) that would've caused them, but hey, what do I know about R.

With #19929 basically finished, I started getting ready for review, and that's when I noticed the method names changed between 2.7.4 and my branch. I didn't think I made any changes that would change method names. I dug through my Git history looking for it. Still can't find anything even when I'm specifically looking for it. ???

So I dig deeper in Git history, and that's when I find the culprit: de8cd4c6a56cac85dbcf7fe1fcb29abe98fe9f18. This commit basically rewrites the previously-generated Arvados.R without any corresponding change to autoGenAPI.R. No, there is not a relevant change in a nearby commit either (git log --stat sdk/R/R). I assume this current version of Arvados.R was made with editor macros or something like that.

So. What do we do from here?

Option #1 is I extend my branch to update autoGenAPI.R to generate code that's API-compatible with the current Arvados.R. I do not think this is a huge task but I do not think it is a small task either. It kind of depends on whether the changes to Arvados.R introduced lots of special cases or anything like that that will be hard to regenerate programmatically. I will look into that.

Option #2 is we abandon any kind of code generation for the R SDK and I prepare a new branch that deletes the obsolete infrastructure and, I guess, implements this ticket by updating Arvados.R by hand. We will continue to have to do this for the foreseeable future. I expect this will lead to lots of fire drills where we're preparing a release and then we have to slap together some quick fixes to get the R SDK tests to pass because they've fallen out of sync with the API.

I know which option I prefer but I also realize I've already sunk way more time into this ticket than anyone probably anticipated and someone should probably check me before I keep digging this hole.

You should add yourself to the author list (man/ArvadosR.Rd and DESCRIPTION)

I did a little poking at the package build process and it looks like as long as R CMD build R (to build the tar.gz package) does the same thing then the publishing should be unchanged.

This LGTM!