Unifying Owner Telescope

For the simple data types, the owner telescopes of their constructors are trivial: the data telescope. But the indexed data types are not, the owner telescopes of their constructors depend on the bindings of each pattern. For example:

open data Vec Type Nat
  | A, zero => vnil
  | A, suc n => vcons A (Vec A n)

The owner telescope of vnil is (A : Type) (size = 1); But for vcons, it is (A : Type) (n : Nat) (size = 2).

However, we implemented this incorrectly (see CallTerm.Data#conHead, PatToTerm.java )

This should be a potential bug.

[ice1000]

I spent 3 hrs debugging this and there is finally a clue. See branch xyy.

[HoshinoTented]

Why do you think so? Did you mean I should not use conTele2?

I mean it is buggy that conHead constructs a ConHead with just DataCall#args. For example:

open data Vic (n : Nat)
| O => vnil
| S n' => vcons (Vic n')

If you want to construct a ConHead of vcons for some DataCall: Vic (S O), it should be vcons {O} instead of vcons {S O}.

PatTycker didn’t use the wrong ConHead#args, but PatClassifier (which I am not familiar with) might use.