Add bucklescript support

[pitag-ha]

The two important commits are:

• Support binary AST's of any supported version

  Before, the OCaml version of any binary ast input had to coincide with the version the ppxlib driver got compiled with. That was a problem for bucklescript/rescript.

  With this commit, all versions from the Versions module are accepted as input. The input version is preserved by ppxlib when outputting a binary AST.

• Implement run_as_ppx_rewriter internally

  Besides taking away a burden for Astlib, implementing run_as_ppx_rewriter internally has two advantages:

  • it allows making changes with more flexibility
  • run_as_ppx_rewriter gets syncronized with standalone_run_as_ppx_rewriter

  With this change, the three flags -loc-filename, apply, dont-apply are taken into account when given. Before, they were allowed to be passed in but not taken into account.

[pitag-ha]

The ocaml-ci arm32 failure isn't related to this PR.

There is a minor change in behavior that I haven't added to the changelog: if someone hands in a binary AST with an unknown OCaml version to the standalone, we now raise instead of packing that error into an extension point and returning the corresponding AST; and same if the input file doesn't exist. Should I mention that in the changelog?

[ghost]

Looks good overall. The code definitely looks more straightforward now!

[ghost]

There is a minor change in behavior that I haven't added to the changelog: if someone hands in a binary AST with an unknown OCaml version to the standalone, we now raise instead of packing that error into an extension point and returning the corresponding AST; and same if the input file doesn't exist. Should I mention that in the changelog?

It should go into the changelog, so that users can quickly identify which version introduced this change in behaviour.

But thinking about this again, packing such errors using the compiler AST versions seems like a reasonable behaviour and is more conservative, so I think we should do that instead. Could you do this change?

[pitag-ha]

But thinking about this again, packing such errors using the compiler AST versions seems like a reasonable behaviour and is more conservative

Ok, I've done that now. Therefore, I've had to restructure the loading logic a bit.

[ghost]

The only downside that has is that if the initial exception was raised without loc parameter, here I can't add it anymore. But that only has an impact for input doesn't exist errors (see here), where the input file is reported inside the error message anyways.

About that, looking at the Location.error type in the compiler, it should be possible to write a function update_loc : error -> Location.t -> error that updates the location of an error.

[ghost]

Changes since last review are looking good. Just need to avoid the double file opening, and perhaps factorise a bit more the *run_as_ppx functions

[ghost]

I also pushed a slightly more idiomatic version of create_versioned_error_f

[pitag-ha]

Thanks! I think I've implemented all the changes you've suggested.

[ghost]

Looks all good!