Unable to find declaration of type when editor position is in another declaration

[Tim-ats-d]

See ocaml/ocaml-lsp#1586

[voodoos]

This one is interesting: locate uses a context around the cursor to guess the kind of the item being located: is it a value, a type, a module, etc ?
If the user did not provide a specific type, it thus reasonable, when on a type declaration, to answer "already at origin, doing nothing".

merlin/src/analysis/locate.ml

Lines 835 to 859 in 8ba0ef8

	let infer_namespace ?namespaces ~pos lid browse is_label =
	match namespaces with
	| Some nss ->
	if not is_label then `Ok (nss :> Env_lookup.Namespace.inferred list)
	else if List.mem `Labels ~set:nss then (
	log ~title:"from_string" "restricting namespaces to labels";
	`Ok [ `Labels ])
	else (
	log ~title:"from_string"
	"input is clearly a label, but the given namespaces don't cover that";
	`Error `Missing_labels_namespace)
	| None -> (
	match
	(Context.inspect_browse_tree ~cursor:pos lid [ browse ], is_label)
	with
	| None, _ ->
	log ~title:"from_string" "already at origin, doing nothing";
	`Error `At_origin
	| Some (Label _ as ctxt), true | Some ctxt, false ->
	log ~title:"from_string" "inferred context: %s" (Context.to_string ctxt);
	`Ok (Env_lookup.Namespace.from_context ctxt)
	| _, true ->
	log ~title:"from_string"
	"dropping inferred context, it is not precise enough";
	`Ok [ `Labels ])

But it's not a reasonable shortcut if the user asked to locate a specific type different from the one whose declaration is under the cursor. I guess we could try to check the name of the type declared under the cursor and see if it matches with the one specifier by the user.

The issue then, is too guess what is the kind of the given name, and we don't really have a way to do that. Using the same heuristic based on the cursor position might be good enough.