Highlight source locations when reporting errors

[Armael]

This PR adds support for reporting source location when displaying errors, similar to the error reports in OCaml >= 4.08. This relies on the pp_loc library.
The easy part was plugging in the library.
The harder part was to fix the locations produced by gospel. The solution was twofold:

• modify the preprocessor to carefully preserve column alignment of the payload within (*@ *) blocs (by emitting newlines within the [@@@gospel ...] where appropriate), using line directives to make up for the extra lines introduced
• carefully handle and "repair" the locations at the time of displaying the error, to only rely on line and column numbers, but not byte offsets from the start of the file (which happen to refer to the file after preprocessing rather than before preprocessing as one might hope).

Most of the tricky location handling logic is factored out as helpers in pp_loc, so here most of the code is about modifying the preprocessor.

The result of the new error messages can be seen in the last commit, which updates the .expected files of the testsuite.

[pascutto]

This is really great, thank you!
I'm surprised (in a good way) that the locations were not too broken besides the preprocessing part.
Could you also add a changelog entry under "Improved"?
You're also missing a dune build @fmt --auto-promote.
And Lexing.set_filename is not supported before 4.11 unfortunately.

[pascutto]

I just pushed a change in the test format to allow this. Expected results are now displayed in

(* {gospel_expected|
   ...
   |gospel_expected} *)

Let me know what you think @Armael :)

[pascutto]

It looks like pp_loc outputs a different message depending on the OCaml version (with a change in 4.11). Is that a known behaviour?
The locations seem wrong on < 4.11

[pascutto]

I believe I found the issue.
ppxlib actually does not stabilize the OCaml parser, so it parses the code with the current OCaml's parser, then converts the AST to the stabilized version (which we use).
As a consequence, if an AST node did not previously contain a location, the translation will use a dummy (or approximate) translation in the stabilized translation.
In our case, the locations added in ocaml/ocaml@abc53d1, which you now use, were not present prior to 4.11, and therefore cause the issues in this PR.

To me that would be a sufficient argument to upgrade to 4.11 (which is already two years old) and avoid this.

[Armael]

ah, I see. Upgrading to 4.11 sounds good to me.