Dot composition detection

[hmgibson23]

As per #589

Seems to work pretty well but I'm welcome to alternatives!

I removed the forall check since it seems redundant now.

[hmgibson23]

Ah okay. Sure will try those.

[hmgibson23]

@gracjan not sure I understand the issue with the lonely . since it will return nil - isn't that expected?

[gracjan]

It should not crash.

[hmgibson23]

It doesn't crash for me. What version of Emacs are you using?

[gracjan]

I haven't checked it, I just asked after seeing code.

[hmgibson23]

Okay well it seems to work fine. I've put back the forall code and all seems to work well.

[gracjan]

I'll check it out soon.

[deviant-logic]

Tried it. It's better than the current state of things, so if I have a vote, I vote merge it (or to be more consistent merge a change that looks like:

(defun haskell-font-lock-dot-is-not-composition (start)
  "Return non-nil if the \".\" at START is not a composition operator.
This is the case if the \".\" is part of a \"forall <tvar> . <type>\"."
  (save-excursion
    (goto-char start)
    (or (re-search-backward "\\<forall\\>[^.\"]*\\="
                            (line-beginning-position) t)
        (not (or (eq ?\  (char-before start))
                 (eq ?\  (char-after  start)))))))

Which uses eq and ?\ instead of turning (char-before start) into a string to compare it to " ". I don't know if that actually matters, but it's how the rest of the file seems to go about things.

This change still misses the case where (.) is used as composition without spaces around it, which is common in lens-style code like _1._2._3. For that matter, it also misses (.). I conjecture that the rule we're looking for is closer to: "if the character before start is is a word character, and the identifier before start starts with a capital letter, and we're not dealing with the forall case, then dot isn't composition"?

[gracjan]

What you really want is:

(?:MODID\.)*([allsymbolchars]+)

and then use prettyfication only when (match-string 1) is exactly equal to a single dot.

Current prettyfication engine is not ready for such logic.

[gracjan]

@deviant-logic: You solutions is slightly better but it is still a hack. I'll merge the version by @hmgibson23 because it is good enough for the time being.

Meanwhile we wait for a skilled volunteer to actually fix the whole mess.