xref-find-references for keywords

[vemv]

One often can wonder "where is this keyword used?". Especially (but not limited to) ns-qualified keywords.

So our existing xref-find-references functionality could be extended to support keywords, based on simple source code analysis.

It would be our first xref-like feature that would support clojurescript.

And also, I believe, one of our very first features using a more comprehensive static source code analysis (e.g. traverse many files in search of matches), with all it takes (performance, caching etc).

I wouldn't want to create our own clj-kondo/clojure-lsp style local databases - those projects are much more refined than what one could handroll of a one-off feature.

I'm thinking of something in-memory, parallel, maybe backed by a LRU cache.

Normally, keyword search is limited to a project, so we can take advantage of that to quite greatly limit the search corpus (maybe to the extent of not needing a cache. IDK).

So, in a way, this would be like a glorified grep, except that it understands keywords better:

• for the input parsing, it understands that our query of ::foo means :some.ns/foo
• while searching a given ns for matches, it knows to search for any of ::foo, ::bar/foo, :a.b.c/foo, depending on the context
• matches are exact (grep can struggle when a given keyword is a subset of a longer keyword)

Leaving it here for the hammock.

[vemv]

One quick idea is that the cache could be rather small by design - instead of caching large data like slurped file contents, or an AST, etc, we keep this simple mapping:

• ns -> <keywords used by that namespace>

e.g. {my.ns [:bar/baz :quux/quuz]}

So the whole cache for a given project would be a quite tiny hashmap.