Update tree-sitter-kotlin and adapt the queries

[interkelstar]

Why

The pinned grammar revision (4e909d6) mis-parses a class carrying two or more annotations when another declaration follows it. class and the class name come out as simple_identifiers and the body as a lambda_literal, so the tree contains no class_declaration at all — and no ERROR node, so the file looks healthy.

@Component
@ConfigurationProperties(prefix = "app")
class Config {
    var url: String? = null
}

class Other

Visible effects: the class name is highlighted as a variable rather than a type, and the class is missing from the outline and from breadcrumbs while its own properties still show. Spring-shaped code hits this constantly.

This is already fixed upstream, 185 commits past the pinned revision.

Query changes

Four node changes since then need the queries to follow. The first one is not cosmetic — it stops outline.scm compiling at all against the newer grammar, so the bump cannot land without it:

Change	Query
property_declaration holds binding_pattern_kind instead of bare "val"/"var" tokens	outline.scm — the ["val" "var"] alternation became an impossible pattern
"null" is now the named null_literal	highlights.scm — was invalid node type
"!is" / "!in" are no longer single tokens (optional "!" + "is"/"in")	highlights.scm — dropped; "!", "is" and "in" are all already in the operator list
interpolation exposes interpolation_identifier_start, interpolation_expression_start, interpolation_expression_end (upstream #270)	highlights.scm — replaces the "$", "${", "}" tokens

Verification

Every query was run against both grammar revisions over a real 21-file Kotlin project (Spring, Gradle) and the capture counts compared, to catch patterns that still compile but silently stop matching:

query	before	after
outline	948	952	+4 — the annotated classes that were missing
highlights	11480	10855	−625, see below
brackets	2236	2236	unchanged
indents	2136	2136	unchanged
injections	72	72	unchanged
overrides	334	334	unchanged
runnables	200	200	unchanged

Breaking the highlight delta down by capture kind, it is punctuation.delimiter (−619) and punctuation.bracket (−13); everything else moves by at most ±4, in the positive direction.

The punctuation.delimiter drop is the dots inside import statements: the newer grammar builds the import path as one token, so its dots are no longer separately addressable and no query can colour them independently. Dots in package headers and in navigation expressions are unaffected. This is a deliberate upstream change, not something the queries can restore.

Known remaining case

The same family of mis-parse still affects declarations led by two or more plain modifiers (internal open class First {} followed by another declaration) — upstream issue fwcd/tree-sitter-kotlin#277, with a fix proposed in fwcd/tree-sitter-kotlin#280. That one is independent of this bump; the annotation case above is fixed by the revision this PR moves to.

[interkelstar]

Moved the pin from c8ac3d2 to 1852ea1, the current tip of upstream main.

c8ac3d2 predates two fixes for exactly the case this PR is about. fwcd/tree-sitter-kotlin#278 stopped a modifier-led class before an adjacent declaration from parsing as an infix expression, and #280 did the same for object, which #278 did not cover. Both landed on 2026-08-01.

Re-verified against the new grammar with the CLI upstream uses, cache cleared:

@Configuration
@EnableConfigurationProperties
class DataSourceConfig(private val url: String) { ... }

internal open class Other

object Registry { ... }

sealed interface Shape

parses as three class_declarations and one object_declaration, with no ERROR and no infix_expression. All seven query files still compile against it, and the annotated class is back in the outline, which is what was missing from breadcrumbs.

[MrSubidubi]

Thank you once again!