Rewrite the parser as hand-written recursive descent, removing goyacc

BUILD.bazel

@@ -8,15 +8,44 @@ exports_files([
 go_library(
     name = "parser",
     srcs = [
+        "api.go",
         "digester.go",
-        "generate.go",
-        "hintparser.go",
+        "hint_token_kinds.go",
         "hintparserimpl.go",
+        "keyword_classes.go",
         "keywords.go",
         "lexer.go",
         "misc.go",
-        "parser.go",
-        "yy_parser.go",
+        "parse_admin.go",
+        "parse_alter.go",
+        "parse_analyze.go",
+        "parse_binding.go",
+        "parse_brie.go",
+        "parse_column.go",
+        "parse_create_index.go",
+        "parse_create_misc.go",
+        "parse_create_table.go",
+        "parse_create_view.go",
+        "parse_dml.go",
+        "parse_drop.go",
+        "parse_expr.go",
+        "parse_flashback.go",
+        "parse_func.go",
+        "parse_grant.go",
+        "parse_hint.go",
+        "parse_misc.go",
+        "parse_procedure.go",
+        "parse_rename.go",
+        "parse_select.go",
+        "parse_set.go",
+        "parse_show.go",
+        "parse_tidb.go",
+        "parse_txn.go",
+        "parse_types.go",
+        "rd_errors.go",
+        "rd_parser.go",
+        "redact.go",
+        "token_kinds.go",
     ],
     importpath = "github.com/sqlc-dev/marino/parser",
     visibility = ["//visibility:public"],
@@ -48,6 +77,7 @@ go_test(
         "lexer_test.go",
         "main_test.go",
         "parser_test.go",
+        "rd_errors_test.go",
     ],
     data = glob(["**"]),
     embed = [":parser"],

CLAUDE.md

@@ -0,0 +1,44 @@
+# marino development guide
+
+marino is a hand-written recursive-descent parser for the MySQL dialect
+(TiDB-flavored). It was ported from a goyacc grammar; PLAN.md records the
+architecture of that rewrite and the decisions that survive it.
+
+## Layout
+
+- `parser/rd_parser.go` — parser core: token window over the streaming
+  lexer, dispatch (`rdRegister`, one owner per leading token), statement
+  loop with the historical statement-text bookkeeping.
+- `parser/parse_*.go` — one file per statement family. Every nontrivial
+  parse function carries a comment naming the grammar production(s) it
+  implements (the grammar reference is the parser.y of the last goyacc
+  commit; see PLAN.md).
+- `parser/parse_expr.go` / `parse_func.go` — the expression precedence
+  ladder and atoms.
+- `parser/token_kinds.go`, `hint_token_kinds.go` — token constants and
+  lexer value types (snapshotted from the generated parser at removal;
+  hand-maintained now). NOTE: `any`, `recover`, `dump`, `format` shadow
+  Go builtins/imports inside package parser.
+- `parser/keyword_classes.go` + `keywords.go` + `misc.go` tokenMap — the
+  keyword tables, kept in exact agreement by `TestKeywordConsistent`.
+- `parser/parse_hint.go` — the optimizer-hint sub-parser.
+- `parser/rd_errors.go` — error construction; messages reproduce the
+  goyacc format byte-for-byte, validated by `TestRDErrorFidelity`
+  against `parser/testdata/errors.json` (goldens recorded from the
+  goyacc parser before its removal).
+
+## Rules
+
+- The `ast` package is frozen, and the public `parser` API is stable.
+- Error messages are part of the contract: `line N column M near "..."`
+  built from the offending token's recorded position; action errors use
+  `r.actionErrorf` (reduce-time lookahead position). Failures during
+  speculation report the farthest position reached (`farthestFail`).
+- One deliberate deviation from goyacc, documented in PLAN.md:
+  `OriginTextPosition` is the deterministic production-start offset.
+- Always run parser tests with a timeout: they are fast, and a hang
+  means an infinite loop in the parser:
+
+```sh
+go test ./... -count=1 -timeout 120s
+```

Makefile

@@ -1,40 +1,14 @@
-.PHONY: all parser clean
+.PHONY: all test fmt clean
 
-all: fmt parser generate
+all: fmt
 
-test: fmt parser
+test: fmt
 	sh test.sh
 
-parser: parser/parser.go parser/hintparser.go
-
-genkeyword: generate_keyword/genkeyword.go
-	go build -C generate_keyword -o ../parser/genkeyword
-
-generate: genkeyword parser/parser.y
-	go generate ./parser/...
-
-parser/parser.go: parser/parser.y bin/goyacc
-	@echo "bin/goyacc -o $@ -p yy -t Parser $<"
-	@bin/goyacc -o $@ -p yy -t Parser $< || ( rm -f $@ && echo 'Please check y.output for more information' && exit 1 )
-	@rm -f y.output
-
-parser/hintparser.go: parser/hintparser.y bin/goyacc
-	@echo "bin/goyacc -o $@ -p yyhint -t hintParser $<"
-	@bin/goyacc -o $@ -p yyhint -t hintParser $< || ( rm -f $@ && echo 'Please check y.output for more information' && exit 1 )
-	@rm -f y.output
-
-parser/%arser_golden.y: parser/%arser.y
-	@bin/goyacc -fmt -fmtout $@ $<
-	@(git diff --no-index --exit-code $< $@ && rm $@) || (mv $@ $< && >&2 echo "formatted $<" && exit 1)
-
-bin/goyacc: goyacc/main.go goyacc/format_yacc.go
-	GO111MODULE=on go build -C goyacc -o ../bin/goyacc .
-
-fmt: bin/goyacc parser/parser_golden.y parser/hintparser_golden.y
+fmt:
 	@echo "gofmt (simplify)"
 	@gofmt -s -l -w . 2>&1 | awk '{print} END{if(NR>0) {exit 1}}'
 
 clean:
 	go clean -i ./...
 	rm -rf *.out
-	rm -f parser/parser.go parser/hintparser.go parser/genkeyword