[Q] How to use the capture from a first pattern as a predicate for a second one?

[gagbo]

It is somewhat linked to #981

For context, I put everything in that Gist with a screenshot, a repro, and everything: https://gist.github.com/gagbo/f9e9ff04633391408064068356bca4c6

Since I can't match elements at any nested level, I'm trying to make a query with 2 patterns, where the first pattern detects an import name ("autometrics"), and the second pattern only matches nodes based on that first pattern detection.

Currently when I'm trying the pattern, I get a false positive (on line 7 of the sample code), where the name of the function in the call_expression doesn't match the name that got saved in the earlier pattern.

Is there something I'm missing to make this work, or am I obligated to run this as 2 separate query steps ? (with templating to fill the second query with the result of the first)

[sogaiu]

I don't have a whole lot of helpful bits to contribute, but FWIW:

• Tried things out and noticed that the version of tree-sitter-javascript at the "official" playground doesn't appear to understand string_fragment
• I ran a local version of the playground with the latest for tree-sitter-javascript and replicated the reported results
• IIUC, #eq? and friend predicates are all "re"-implemented per language binding so the behavior that's being described here may be specific to the web / wasm bindings (lib/binding_web) -- at least when observing via the playground. May be that makes this code relevant?

The docs here suggest that using more than one capture name with a predicate is supposed to work, but it's not clear to me whether a capture name from one pattern is supposed to work inside another pattern. (But note the point above about the predicates being "per-binding".)

[gagbo]

I think you need to use the tree-sitter-typescript bindings. Even with a somewhat old version (0.12.something) on my local setup I had the string_fragment in the nodes, and I also had it in my 0.20.something version on my other local setup.

But all the other points are still relevant, thanks :)

I know the predicates are implemented per language binding, and I thought that all bindings at least implemented #eq?; but maybe they have different implementations for the case where we want to compare 2 captures

[sogaiu]

I think you need to use the tree-sitter-typescript bindings.

Oh, oops :)

Thanks for pointing that out.

[sogaiu]

FWIW, when I tried with a local version of the typescript grammar, and (#eq? @wrapperAtCall "autometrics") substituted in for (#eq? @wrapperAtCall @wrapperAtName), req and res are not captured, while the other desired captures are.

When I leave out (#eq? @wrapperAtCall @wrapperAtName) completely, I get the results described in the gist (so req and res are captured as well as the 3 other things on other lines).

Perhaps (#eq? @wrapperAtCall @wrapperAtName) is just being ignored?

[gagbo]

Maybe the whole predicate is being ignored, but then I suppose it would be a bug in the TS bindings, since it seems to have special code to deal with #eq? when the 2nd operand is a capture.

If the reason my predicate is not effective is that @wrapperAtName is not "defined" in the 2nd pattern, then I'd like to know what the workaround @ahlinc talked about in 981 is, because I can't figure it out

[sogaiu]

it seems to have special code to deal with #eq? when the 2nd operand is a capture.

Yes, it does look like there is some code to deal with the 2nd operand being a capture -- but may be that only applies if both captures are part of the same pattern?

To quote ahlinc:

you can workaround this by writing a query to find all class_declaration nodes and after that on every class declaration node run a query that would match all identifiers in invocation expressions.

That seems like two queries to me. Or am I misunderstanding?