Support layout annotations

[goldfirere]

This refreshes the work in ccasin/ocaml-jst#1 and makes changes to layout annotations in types as agreed to in the type-systems meeting.

[antalsz]

I left my comments, but this looks broadly good. The biggest thing I was confused about is the situation with the test cases; a lot of them seemed to be incomplete. I didn't end up reviewing most of them.

EDIT: Oops, I hit "submit" while I meant to be drafting. Some more comments forthcoming…

[antalsz]

Okay, good, I hadn't missed too much. As discussed, I didn't review typing/; as mentioned above, I didn't review testsuite/ yet, I'm awaiting clarification.

[antalsz]

Reviewed up through 860b01b "More work on extension constructors"

[goldfirere]

Here are review instructions:

General notes:

• The commits are meant to be reviewable individually. And if GitHub had good support for commit-by-commit review (e.g. allowing you to say you had read a commit and then to get you to focus your efforts on other parts of a file when viewing it later), I would advocate for commit-by-commit. But that's not the world we live in, so I've come up with another plan.
• The other plan (details below) could be implemented more smoothly by doing some big restructures of this PR. But I estimate that the effort of doing so is not worth it. Indeed, I've already probably spent the better part of a day (cumulative) trying to keep the commits clean, and I'm not sure that effort has yielded any benefit, so I'm not inclined to invest more in this direction.
• Exception to the general rule that individual commits make sense: the testsuite. Because the testsuite cares about extension levels, and extension levels are tied in with the jane-syntax approach, it didn't make sense for me to keep the testsuite up-to-date at every commit. So if you build an arbitrary commit, you should expect the annots.ml and annots_beta.ml tests to fail, along with two test files that are labeled "failing". (But that's it -- just those four tests should fail. I kept the testsuite up-to-date otherwise.)

I've broken the review responsibilities down across the two requested reviewers, @antalsz and @ccasin.

@antalsz

1. Review "Pprintast now depends on Printast" (bdac10c). This could be a separate PR, but see comments above about investment of time making clean PRs. It's just a little refactor of dependencies. Upstream has offered support for this direction of travel, and I expect to upstream this particular commit after this PR has landed; I've made an internal ticket to track this task.
2. Recall that you've previously done review here. Commits starting from "Comments from Antal" are new to you.
3. Review "Comments from Antal" through to the end. It's unclear whether this will be easier going commit-by-commit or just as a chunk. (GitHub supports displaying changes from a range of commits.) These commits also fix up the testsuite, hopefully clearing up the (quite reasonable -- sorry!) confusion about the testsuite from earlier.
4. Confirm that the final changeset includes only a conservative change to ast_helper.mli and no changes at all to parsetree.mli.
5. Take a pass through open conversations from previous reviews -- hopefully these issues are resolved now.

@ccasin

This PR does not remove the mechanism for writing e.g. type t [@@void], and writing type t : void is interpreted as the former, which does not use the new jane-syntax mechanism. The goal in this PR is simply to remove changes to the parsetree that I introduced previously. I will return to fixing up the remaining non-jane-syntax annotations in a future PR. Of course, there will be a small challenge in that we need to continue to support [@@immediate] in that spot, but it's a story for another day.

1. Review the final changes to the testsuite, to see if you like the final destination of this PR.
2. Remember that you gave an in-depth review of Add support for annotating types with layouts. ccasin/ocaml-jst#1. All conversations there were Resolved. That PR got squashed and has become the commit "Initial implementation of layout annotations" (2e2cac3). You may want to glance through this commit to refresh your memory, but no need for a re-review. The rebase was uninteresting.
3. In commit Rename ~reason to ~context for annotations (only) (1d35034), confirm that the commit really is just a renaming. Nothing else happens here.
4. Review type-checker changes in "Move layout annotations into aliases" (edd32a5).
5. Review type-checker changes in "Support for annotations on _" (91b3a78).
6. Review type-checker changes in "Handle type parameters uniformly" (9e2b5e2). This adds a new check in typeclass.ml and does non-trivial restructuring in typedecl.ml / typetexp.ml.
7. Review commit "Incorporate test case from old PR, improving msg" (3166795), which comes from your comment on the old PR.

---

I hope this is manageable! If we ever need to do a temporary parsetree change again, we should follow the workflow pioneered by @ccasin and @rtjoa and review commits as they get merged into a long-running branch, which would avoid this particular review challenge.

[antalsz]

I believe my review is complete, modulo my outstanding comments – I didn't review the typing/ directory thoroughly, which IIUC was the plan. EDIT: To clarify, I reviewed that the changes to typing/ just moved code around, but did not review that the code so moved was correct.

[antalsz]

Approving for non-typing changes

[ccasin]

I have some questions and suggestions, but this looks great.

[goldfirere]

OK. Finished responding.

[ccasin]

Looks good now, thanks!

[Ekdohibs]

So, the build failure seems to be due to the minimizer (make minimizer and make minimizer-upstream) not building anymore; you'll need to at least update the code of compat.jst.ml to make it compile again both with ocaml-jst and upstream ocaml.

[Ekdohibs]

For files in chamelon/.

[mshinwell]

dune and driver/ changes