Implement versioned .js.flow

[goodmind]

Proof of concept for #4917
Has bad performance on all resolutions

Performance checks

Current measurements: https://gist.github.com/goodmind/65695c3b7a3aa0766812b15779a984dc

1. Build flow binary
2. Run flow check --profile on codebase
3. Record results
4. Repeat with regular flow binary and this PR, properly stopping all servers before running new binary

[nmote]

I skimmed this and overall it looks reasonable. I'm taking tomorrow off, but I'll import this and test it out internally on Monday. If performance looks good I'll give it a more thorough review. In the meantime, could you make the following changes?

First, could you update the summary to include a bit more information, including the results and methodology for the performance tests we've discussed on Discord?

Could you add a test that the Flow server restarts correctly if it sees a change in the flow field of a package.json file? See the package_incompatible function in module_js.ml. Based on the implementation it should restart, but let's just cover our bases.

Could you also add a test for the case where the flow property exists but contains no typesVersion property, as well as a test for the case where the typesVersion property is empty?

Also see my inline comments.

[goodmind]

@nmote how can I test server restart?

[nmote]

I took a quick look and didn't see any existing examples of testing for server restart. With that in mind, I would be happy if you manually test that the server restarts and update the PR description with the steps you took.

[facebook-github-bot]

@nmote has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[nmote]

It appears that we don't have Core's String module available for our internal build, so the build here is failing. I would have mentioned this sooner but this was masked by another build issue that was easy to fix.

@mroch, do you know if it would be reasonable to pull in Core's String? @goodmind, if not, could you rewrite this without using that library?

Sorry, I know it's not a good developer experience to have a separate build system that you can't see or use. Unfortunately there's not a whole lot we can do about that currently.

[goodmind]

Yeah, it just needs substr_index, couldn't find this in jane street core

[goodmind]

@nmote I removed Base

[facebook-github-bot]

@nmote has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[nmote]

Unfortunately the test you've added in this commit fails on Linux. Do you have access to a Linux development environment to look into this failure?

[facebook-github-bot]

@nmote has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[goodmind]

@nmote any news?

[nmote]

Yes, I've been trying to get to this. Sorry again for the delay 😞

I'm planning to rebase it past the ocamlformat change for you, since I'm not sure whether we have ocamlformat properly set up on GitHub yet. I was hoping to get to that yesterday, but ran out of time. I should be able to do that and give you a review today.

[nmote]

I've rebased this past the ocamlformat update and I put up #8135 with the updated code. Feel free to take what's there and push to your branch here.

I'll do a round of review over there, so that I don't have to wait for you to update this PR.

[nmote]

I've done a round of reviewing over there. Let me know when you get a chance to respond.

[nmote]

I'm still concerned about the code in the types_versions_candidates function and node_paths_part.ml -- see my comment in #8135

[nmote]

Closing since there hasn't been any movement on this for some time