This is interesting. We could definitely match this reference, and I think it would be useful. We should experiment with doing this. Tentatively putting in the v0.17.0 milestone, as v0.16 is filling up.

CC @adonovan

FWIW the inlay hints show the field names so the info is already wired in somewhere :)

By design gopls' references algorithm reports the locations of identifiers that reference a given symbol, whereas in your example there is no field identifier. So the operation is working as intended.

The algorithm does not report all of the many kinds of implicit references that arise in other places, which, in addition to your example, include references:

• to anonymous fields, in selections where x.f is shorthand for x.a.b.c.f;
• to parameters, when passing argument values in a function call;
• to named result values, in a blank return statement;
• to types, in assignments x = y that involve an implicit conversion x = T(y);

We have had other discussions about and requests for different kinds of 'references' queries, that are not invoked through the usual LSP 'references' operation. For example: show all the places where this struct field or slice element is accessed, even implicitly. It is easy to define the queries as code actions ("Source actions..." in VS Code); the main challenge is the lack of a way to communicate to the client that the result of this operation is a list of annotated source locations that should be displayed using essentially the same UI elements as a true references query. I think we should lobby for the inclusion of such a feature into LSP.

Thank you for the thorough reply.

I understand your point and agree with most examples, but don't agree using "identifier is spelled-out" should be a hard rule, and think each case is worth examining from the point of "does this bring value to the user" and "is that value worth the work/complexity".

Maybe the intended use case can motivate some rule bending? It is finding all assignments to a field, which ATM cannot be easily done as you need to check all field references + all struct references. And doing that second search is not intuitive at all IMO.

I think we should lobby for the inclusion of such a feature into LSP.

I agree, having more than one type of symbol search would definitely be useful.

I understand your point and agree with most examples, but don't agree using "identifier is spelled-out" should be a hard rule, and think each case is worth examining from the point of "does this bring value to the user" and "is that value worth the work/complexity".

Many users would be surprised if they asked to see references to "foo" and the result included a line of text that doesn't contain foo. A better UX for this feature would be a query report that annotates the location with a comment such as "implicit selection of field foo". But again that requires LSP changes.

At least in that case you're presented with information so you ask yourself a question. The current situation doesn't confront you to anything and you're very likely to miss some instances of what you're looking for.
EDIT: also the inlay hint makes it pretty obvious.

Anyways feel free to close or rescope to something else if you don't plan on implementing what I opened the issue for originally.

I've filed microsoft/language-server-protocol#1911 upstream for LSP support for generalized reference queries.