MetaOCaml: reserved >. token

[Octachron]

Submitted on behalf of Oleg, cc @yallop .

This PR is part of a series of patches sent to me by Oleg that proposes to reduce the difference between OCaml and MetaOCaml at the syntactic level.

This first PR modifies OCaml lexer to make sure that all OCaml program are syntactically valid in MetaOCaml.
In particular, it removes ">." from the set of infix operators:

let (>.) _ _ = ()
let () = 1 >. 2

emits a syntax error after this PR.
Similarly, the tokenisation of >.>. is modified to >. >..

[Octachron]

After some more discussion with Oleg and @yallop , removing >. from the set of infix operators create some incompatibility with the existing ocaml libraries (like Core). A more compatible option is probably to have two lexing modes: mainline and MetaOCaml and make the lexer switches between the two modes. The question is then how to trigger thus switch. After some back and forth, it seems that there is three possible options:

• implicit: the first .< token triggers the switch
• explicit: a boolean -metaocaml (metaocaml_lexing?) Clflags controls the lexing mode.
• hybrid: the -metaocamlflag can be either implicit (aka switch on first .<, true, or false (same as explicit above).
  Mainline OCaml would use false as default, whereas MetaOCaml would set the default to implicit.

[yallop]

I think any of these three solutions could be made to work. Do any of them actually require changes to OCaml, or is it sufficient in each case to change MetaOCaml?

[johnyob]

One issue with the proposed solution is we don't provide a method of escaping this "MetaOCaml mode".
For example, one could imagine writing something like:

open Core

module Float : Robustly_comparable with type t = float = struct ... end
let true_ = .< Float.(2. >. 1.) >.

[yallop]

That's true, and I'm not sure there's a reasonable way to treat >. both as an operator and as a closing delimiter. There's an easy workaround (aliasing >.), but it's not really ideal. It might be better to use extension points, like this:

let true_ = [%code Float.(x >. y)]

(as is under consideration) rather than building in the slightly fragile .< >. syntax.

[Octachron]

With the syntax proposed in ocaml/RFCs#27, MetaOCaml, users that would have activated >. as a keyword could still used the escaped syntax Float.(\#>.) 2. 1.. Nowadays, I tend to think that registering >. as an optional keyword along the lines proposed by ocaml/RFCs#27 is the most principled alternative.

[gasche]

I believe that this PR is superseded by #12470, which includes MetaOCaml's new "adaptative lexing" mode to avoid incompatibilities pointed out by @yallop.