Call remove_pos only where a type comes into the system

[zuiderkwast]

Remove_pos normalizes the syntactical representation of a type, by recursively removing the location annotation (line and column). This is to be able to compare types using equality and pattern matching in the type checker.

This should only be done once on each type and only when the type comes into the type checker, i.e. when types are extracted from a parse tree.

This PR sorts out the mess, makes sure it's done where it should and removes it where it shouldn't be done.

Note that gradualizer_db doesn't use this, so we call remove_pos on the types returned by gradualizer_db.

[erszcz]

This looks good, but I think there are a few places where remove_pos is now redundant:

• Gradualizer/src/typechecker.erl

  Line 772 in 0657d76

  normalize_rec(typelib:remove_pos(T), Env, UnfoldedNew);

• possibly

  Gradualizer/src/typechecker.erl

  Line 3060 in 0657d76

  lists:map(fun typelib:remove_pos/1, Assocs)),

• possibly

  Gradualizer/src/typechecker.erl

  Line 3074 in 0657d76

  case take_assoc(typelib:remove_pos(Key), Assocs, []) of

Apart from the above, there are a few cases where it's called on a value returned from various gradualizer_db functions - for example:

Gradualizer/src/typechecker.erl

Line 785 in 0657d76

normalize_rec(typelib:remove_pos(T), Env, Unfolded);

Is there a reason to keep types with their position in the DB or could we also remove it there, to remove the redundancy at call sites?

[zuiderkwast]

I think there are a few places where remove_pos is now redundant

Thanks, the tests still pass when I remove these.

Is there a reason to keep types with their position in the DB or could we also remove it there, to remove the redundancy at call sites?

The reason I didn't do it already is that this is no mess so far. Remove_pos was simply not done there, without exceptions. The mess was only in typechecker. But since gradualizer_db does other stuff such as absform:normalize_function_type_list/1 and absform:normalize_record_field/1, I think it can do remove_pos too. It will make everything simpler. 👍 I'll do it tomorrow.

[erszcz]

That's neat!