Context of a function does not form a node

[ocharles]

If I have the following:

foo :: Eq a => a -> a

I may later want to add more constraints to the context of this function. I would like to move the point over Eq a and use ) to select the whole Eq a node and then surround it with parenthesis (M-(). Unfortunately, the node selected is Eq a =>, which is too large a selection,

[chrisdone]

Agreed, I noticed this recently myself. This is how HSE treats constraints in the AST. Similarly, foo { } record syntax behaves like this too (you can select foo { }, but not the { }—that's annoying).

What we can do in genHSE is add some special-cases to reconstruct the sub-parts and output more source spans that the Emacs-side can take advantage of transparently.

There's a field srcInfoPoints which sometimes contains additional source spans (which is true for the RecordUpdate case above), so for those we can easily construct a span from those, all that would need to be done is to make sure they have an appropriate (but nonce) name and constructor (e.g. RecordUpdateFields). Because the output to Emacs is always:

[<type> <constructor> <start-line> <start-col> <end-line> <end-col>]

E.g.

["Decl SrcSpanInfo" TypeSig 1 1 1 36]

In the case of class constraints, they don't have this handy srcInfoPoints decomposition, but given that we simply need to omit the " =>" part, this should be trivial to do, too.

[chrisdone]

I'll come back to this another time. I fixed record fields, but class constraints are a mess. The source spans in the AST are completely useless.

[chrisdone]

Another item to fix:

case x of
  Foo bar mu -> X

One cannot select the Foo bar mu separately. This should be possible, it's a common use-case.

[chrisdone]

I don't think I'll get round to this issue.