Walker now dispatches through a `languageKindMap` keyed on language
name before falling back to the conventional `*_definition` /
`*_declaration` suffix matcher. Covers grammar rule names that
don't match the standard convention: Erlang `fun_decl`, Haskell
`function` / `signature` / `data_type` / `type_synomym` (upstream
typo preserved), Crystal `class_def` / `method_def`, Nim
`proc_declaration` / `func_declaration`, Ada `function_specification`
/ `procedure_specification`, Odin `procedure_declaration`, Apex
`class_declaration` / `method_declaration` / `trigger_declaration`,
Solidity `contract_declaration` / `interface_declaration` /
`modifier_definition`, F# `function_or_value_defn` / `named_module`
/ `record_type_defn`, ReScript `let_declaration` / `module_declaration`,
ObjC `class_interface` / `class_implementation`, AL
`codeunit_declaration` / `table_declaration` / `page_declaration`,
GDScript `class_name_statement`, Jinja / Twig `macro_statement`.
`nodeName` recurses three levels deep so wrappers like `fun_decl ▶
function_clause ▶ atom` resolve to the inner identifier;
`isIdentifierKind` recognises `constant` / `atom` / `variable` in
addition to the conventional `*identifier*` patterns.
17 regex extractors migrated to forest+regex hybrid — forest covers
structural definitions via the walker, regex stays for
idiom-specific edges (imports, framework-specific metadata): R,
Gleam, Erlang, Crystal, Haskell, Nim, Hare, Vala, Zig, Groovy, Ada,
D, Matlab, Odin, F#, Jinja, ReScript. Net change: −469 lines across
22 files; each migration deletes ~50–200 lines of bespoke regex and
delegates definition extraction to the forest grammar.
Adds `internal/parser/forest/dump_kinds_test.go` — a parameterised
research helper. Future contributors map a new grammar's
node-kind landscape in ~5 minutes by adding a fixture to its table
and running `go test -run TestDumpGrammarKinds/<lang>`. Lifts
`lineAt` / `findBlockEnd` out of `zig.go` into a shared
`helpers_lines.go` so other hybrid extractors don't import them
from a language-specific file.
Lisp-family and uniform-grammar languages (Clojure, Tcl, Pascal,
Cobol, Liquid, Blade, Pug, Nix) stay regex-only — their grammars
don't encode definitions as distinct AST node kinds, so the walker
can't distinguish `(defn foo …)` from `(println …)` without
symbol-text matching (which is what tags.scm `(#match? …)`
predicates do; forest's packages for these languages don't ship
one). Languages attempted and reverted: Vim (special function-name
chars), Julia (short-form `name() = body`), CMake (deeply-nested
name), Perl, PowerShell, GDScript, Twig, ObjC, AL, Apex.
Andrey Kumanyaev
1e4f1d0da7