There's two things I'm concerned about with removing types altogether:

1. In a syntactically invalid file, Merlin might incorrectly infer an expression context when a user wants to write a type. (To be clear, I'm saying "I suspect Merlin will sometimes do this", not "I've observed Merlin do this.")
2. Users might actually sometimes want to write a type in an expression context. One example that comes to mind is in [%sexp_of: t] (t is actually an expression, not a type). Or maybe the user will make the following series of edits: let foo = f in -> let foo = f t in -> let foo = f (x : t) in.

Is there a test that witnesses the behavior change?

The motivation to tweak the behavior is to improve the quality of fuzzy search behavior.

For example, you have a type string and a module String available at the cursor, and typing an expression should only show the module String rather than the type string. In many languages, typing str yields auto-completion to the class/module name, and we want to replicate that behavior for ocaml lsp. If the type shows up in the list before the module name, it can be a bit annoying.

I'm not sure tweaking the priority is the best approach. I still think we should remove types from the list of completions on normal expressions.

@liam923 pointed out that this might behave worse in ppxes (good catch!). I think we should provide more items only for ppxes given its flexibility.

@JimmyZJX I don't think there's an easy way for Merlin to be aware that it's in a location that a ppx treats like a type, so that would be difficult to do.

@liam923 Ack. I'm fine with just changing the order and hope that clients would sort as expected.

@goldfirere I've added tests - see the "Completions in an expression context" one. Without the change in completion.ml, the type foobar comes before the module Foobar.