Parmatch: introduce a type for simplified pattern heads

[trefis]

This is a mostly mechanical change, that has been a long time coming.

I think @lpw25 is a good candidate to review this, given that he too had wanted to implement this (or a similar) idea for a long time.

[stedolan]

Looks good! Glad to see a bunch of assert false go away.

[Octachron]

I like the code. Nevertheless, I have some small minor nitpicking remarks. In particular, the code often uses names of the form d... for pattern head variables. Could we use more mnemonic names? Like h.. as a shorthand for head (which is also used) for instance?

[Octachron]

Minor nitpicking: why not use make at this point?

[lpw25]

Note from a brief discussion with @trefis. In a future PR it might be worth adding a version of Pattern_head.destruct that returned a list of "destructed" patterns, allowing it to work on or-patterns. It would allow for simpler code in places like compat and simplify_head_pat.

[gasche]

I have thought about extending the notion of head to or-patterns (which may have several heads). One property that we lose as soon as we leave the fragment of simple patterns is the equivalence between pattern_desc and head * pattern list. Indeed, this equivalence does not take into account the binding structure "above" the head of the pattern, as present in the cases Tpat_var and Tpat_alias.

So I think that rather than extending destruct, we should have a different API for this case that emphasize that we are talking about the shape of the matched values, as a lossy projection from the pattern: heads : pattern -> head list.

[trefis]

@Octachron I updated as you suggested. I'll clean up the history and merge later today.

@gasche:

So I think that rather than extending destruct, we should have a different API for this case

I believe that is what @lpw25 was suggesting.

[trefis]

Ah, I realize I forgot to answer:

In particular, the code often uses names of the form d... for pattern head variables. Could we use more mnemonic names? Like h.. as a shorthand for head (which is also used) for instance?

d... is often used to indicate we're working with a "discriminating" pattern (which I guess we should call "discriminating head") as produced by discr_pat, not just any head.
I will have another look at the code to see if it's actually used properly.

[trefis]

I pushed some minor renaming along these lines, but things were mostly OK IMO.

[Octachron]

The renamings look nice (in particular in simple_match_args), and make it much more clearer that d is a shorthand for discriminating, thanks!

[trefis]

I rebased and cleaned up the history.
I'll merge once the CI is green.