The smacker fork has been stale since 2024 and pinned us to an old ABI.
Move every adapter to the upstream bindings (v0.25.0) so we can track
new grammar releases.
* internal/parser/tsitter/ — thin compat shim. Re-exports the upstream
API under smacker-style names (Type/Content/StartPoint/NamedChild(int))
so the ~28 language adapters only needed an import-path swap. Adds
Node.Equal()/Node.Id() for identity, since the new bindings hand back
a fresh *Node on every traversal — pointer comparisons no longer mean
anything.
* internal/parser/treesitter.go — RunQuery / RunPrepared / EachMatch
rewritten on top of the new iterator-based QueryCursor.Matches; pool
*ts.QueryCursor directly (the new API has no per-call state leak).
* Per-grammar shim sub-packages under internal/parser/tsitter/<lang>/
wrap each upstream binding's Language() into a smacker-shaped
GetLanguage() *tsitter.Language. Most pull from
github.com/tree-sitter/tree-sitter-<lang>; a handful use third-party
forks (camdencheek dockerfile, elixir-lang, fwcd kotlin, DerekStride
sql, alex-pinkus swift) where the tree-sitter org has no grammar.
* Locally vendor grammars whose upstream Go bindings are unavailable or
broken: markdown (no Go binding), protobuf (no Go binding), dockerfile
(parent module path conflicts with submodule path), swift + sql
(gitignore the generated parser.c). Each ships parser.c + scanner.c +
tree_sitter/{parser,alloc,array}.h compiled directly through CGO.
* Lua extractor rewritten for tree-sitter-grammars/tree-sitter-lua's
current node vocabulary: function_statement → function_declaration
with field-based name access, variables now under
variable_declaration → assignment_statement → variable_list, method
calls via method_index_expression.
* Kotlin pinned to fwcd/tree-sitter-kotlin: the
tree-sitter-grammars/tree-sitter-kotlin v1.x grammar drops the
type_identifier alias and would force a rewrite of the extractor.
* PHP Laravel middleware extractor: replace pointer-keyed processed map
and pointer comparison on parent.ChildByFieldName("object") with
Node.Id() / Node.Equal() — both broke once *Node became
per-traversal.
go.mod loses smacker entirely and gains the upstream + grammar modules.
Andrey Kumanyaev
8ee3ca43aa