Not sure why the line order changed here -- does the expensive comment line still apply? Also qvarid appears to be "qualified varid", so you should probably use that instead of the hand-rolled [[:alnum:]] etc.

Well, font-lock syntax for haskell is in a need for refactoring. To answer your question:

qvarid is only module prefixed version and does not apply to both. What we need is a qualified-or-plain-varid. This is not defined so that is why I went with just inline regex.

Order of rules needed to be changed because otherwise a line later that matches conid and qconid would match conid and qconid. So my line had to be before.

Leftovers like ;; Expensive or Emacs 21 support needs to be cleared up.

Note that the line

(,reservedsym 1 haskell-operator-face)

appears three times for no reason. That is the best testimony to the quality of this fragment of the code.

Does anybody have pointers to an example how to write unit tests for font-lock or have some more interactive way of working with syntax definitions?

That is the best testimony to the quality of this fragment of the code.

:-)

Does anybody have pointers to an example how to write unit tests for font-lock or have some more interactive way of working with syntax definitions?

There's https://github.com/Lindydancer/font-lock-studio, which might help.

Is this merged?

Is this merged?

Not yet, no. Isn't there a way to combine qvarid and varid instead of repeating the [[:alnum:]] incantation? Something like (concat "\\(?:" varid "\\|" qvarid "\\)") would perhaps do the trick.

Sadly modid is defined using normal group not the shy group, so it does not compose:

(modid (concat "\\b" conid "\\(\\." conid "\\)*\\b"))

So using qvarid and qconid might bring trouble in places. So I went with the obvious four-liner that you see in the commit (the new one).

@purcell: what about this?