Hey! Yes, re-iterating my comment on Reddit: What you get from the analysis data is just data, no AST. But you can match this data on location on whatever AST structure you produce using rewrite-clj.

E.g. getting the fully qualified var name or whether something is a local, etc, you should be able to look up in the analysis data

Oh, I see, I thought there's something new I've missed from clj-kondo. This approach doesn't seem suitable for me. I find it more straightforward to have a full AST map at hand right away, as someone who's not familiar with clj-kondo internals.

Yes, I get that. But to produce such a structure, you can use the clj-kondo analysis. Else you would have to write your static analysis yourself

So while walking the raw rewrite-clj structure, you can look into the clj-kondo analysis if some symbol is a local or var, then add that data and produce your AST.

Hmmm, that makes sense. I can see var defs in analysis output, but not local vars. Is there a config switch for that or something?

yes: https://github.com/clj-kondo/clj-kondo/blob/master/analysis/README.md#extra-analysis

nice! I thought :analysis true would include everything

locals and keywords is a lot of data, this is why it wasn't included by default

ok, I'm seeing the data now. Still it means depending on clj-kondo would mean splitting the analyzer in two pieces and parsing inputs twice: to get vars data from clj-kondo and produce AST for everything else myself, which doesn't feel solid to me.

This approach is what several tools are doing, clojure-lsp being one of them. Works quite well. Yes, if you use a tool do to analysis, it becomes a dependency. Parsing the code twice isn't really a problem, it's only a small fraction of the total time taken.

You also benefit from clj-kondo hooks that people write for their custom macros

When clojure-lsp adopted clj-kondo we considering sharing the AST but it wasn't worth the trouble

Good point about macros!