test(offline): real-dictionary move-generator conformance in CI
CI / changes (pull_request) Successful in 1s
CI / unit (pull_request) Successful in 10s
CI / integration (pull_request) Successful in 15s
CI / ui (pull_request) Successful in 1m8s
CI / conformance (pull_request) Successful in 9s
CI / gate (pull_request) Successful in 0s
CI / deploy (pull_request) Successful in 1m38s

Phase A (A4): prove the ported move generator (#188) against the FULL shipped
dictionaries, not just the tiny samples — the deep graphs and complete 26/33-letter
alphabets the samples cannot reach.

- backend/cmd/movegen: add a -dawg-dir mode that emits per-variant golden move-gen
  vectors from the real dawgs (a bounded first-move + a blank case + a deep 7-tile
  mid-game position). Regenerated in CI to /tmp, never committed (like the
  dictgen/validategen vectors), so no dictionary version is pinned into the repo.
- ui/src/lib/dict/generate.realparity.test.ts: env-gated (DICT_DAWG_DIR +
  DICT_MOVEGEN_DIR) parity against that golden — 9 positions across scrabble_en /
  scrabble_ru / erudit_ru match the Go solver exactly. Skips cleanly when unset.
- .gitea/workflows/ci.yaml: the conformance job now generates the movegen golden
  and points the gated vitest at it (DICT_MOVEGEN_DIR).
This commit is contained in:
Ilia Denisov
2026-07-06 02:09:42 +02:00
parent d694ead7b6
commit a9c8f1ecfe
3 changed files with 194 additions and 0 deletions
+79
View File
@@ -19,6 +19,7 @@
package main
import (
"bytes"
"encoding/json"
"flag"
"log"
@@ -108,12 +109,18 @@ type genCase struct {
func main() {
out := flag.String("out", "ui/src/lib/dict/testdata", "output directory for fixtures")
dawgDir := flag.String("dawg-dir", "", "when set, emit real-dictionary move-gen golden from the .dawg files in this dir (conformance mode) instead of the committed samples")
flag.Parse()
if err := os.MkdirAll(*out, 0o755); err != nil {
log.Fatalf("movegen: mkdir %s: %v", *out, err)
}
if *dawgDir != "" {
buildReal(*dawgDir, *out)
return
}
buildSample(*out, "en", rules.English(), sampleWordsEN, []genCase{
emptyCase("empty-cared", englishRack("caredts", 0), scrabble.Both, false),
emptyCase("empty-dogs", englishRack("dogsent", 0), scrabble.Both, false),
@@ -172,6 +179,78 @@ func buildSample(out, tag string, rs *rules.Ruleset, words []string, cases []gen
log.Printf("movegen[%s]: %d words, %d cases", tag, finder.NumAdded(), len(cases))
}
// realVariant maps a shipped dictionary file to the ruleset that scores it and the racks
// the conformance positions use. smallRack/blankRack keep the first-move (empty board)
// lists bounded on a dense dictionary; fullRack drives a deep 7-tile mid-game position,
// kept small by the anchors around the already-placed word.
type realVariant struct {
file, variant, smallRack, blankRack, fullRack string
rs *rules.Ruleset
}
// buildReal emits move-generation golden from the real shipped dictionaries in dawgDir —
// the full alphabets and deep graphs the tiny samples cannot reach — one
// <variant>.movegen.json per variant. Like the dictgen/validategen vectors it is
// regenerated in CI and never committed, so it pins no dictionary version into the repo.
func buildReal(dawgDir, out string) {
reals := []realVariant{
{"en_sowpods", "scrabble_en", "aine", "ain", "aeinrst", rules.English()},
{"ru_scrabble", "scrabble_ru", "аеин", "аен", "аеиноср", rules.RussianScrabble()},
{"ru_erudit", "erudit_ru", "аеин", "аен", "аеиноср", rules.Erudit()},
}
for _, v := range reals {
data, err := os.ReadFile(filepath.Join(dawgDir, v.file+".dawg"))
if err != nil {
log.Fatalf("movegen[%s]: read dawg: %v", v.variant, err)
}
finder, err := dawg.Read(bytes.NewReader(data), 0)
if err != nil {
log.Fatalf("movegen[%s]: parse dawg: %v", v.variant, err)
}
s := scrabble.NewSolver(v.rs, finder)
cases := []genCase{
emptyCase("first-move", encRack(v.rs, v.smallRack, 0), scrabble.Both, false),
emptyCase("first-move-blank", encRack(v.rs, v.blankRack, 1), scrabble.Both, false),
}
for i := range cases {
runCase(s, v.rs, &cases[i], nil)
}
// A deep 7-tile mid-game: place the top first move, then generate again. The
// anchors around the placed word bound the list while still exercising a full rack,
// deep left/right extension and wide cross-sets over the real graph.
full := encRack(v.rs, v.fullRack, 0)
b := board.New(v.rs.Rows, v.rs.Cols)
if m1 := s.GenerateMovesOpts(b, toRack(v.rs.Size(), full), scrabble.Both, scrabble.PlayOptions{}); len(m1) > 0 {
mid := genCase{Name: "mid-game", Rack: full, Mode: int(scrabble.Both)}
runCase(s, v.rs, &mid, tilesOf(m1[0].Tiles))
cases = append(cases, mid)
}
writeJSON(filepath.Join(out, v.variant+".movegen.json"), genFixture{Ruleset: rulesetOf(v.rs), Cases: cases})
total := 0
for _, c := range cases {
total += len(c.Moves)
}
_ = finder.Close()
log.Printf("movegen[%s]: %d cases, %d golden moves", v.variant, len(cases), total)
}
}
// encRack encodes a rack given as the variant's letters (plus a blank count) into the
// index-based genRack the fixtures carry.
func encRack(rs *rules.Ruleset, letters string, blanks int) genRack {
enc, err := rs.Alphabet.Encode(letters)
if err != nil {
log.Fatalf("movegen: encode rack %q: %v", letters, err)
}
idx := make([]int, len(enc))
for i, b := range enc {
idx[i] = int(b)
}
return genRack{Letters: idx, Blanks: blanks}
}
// runCase fills a case's Moves by generating on a board holding the given placed
// tiles (nil = empty board).
func runCase(s *scrabble.Solver, rs *rules.Ruleset, c *genCase, placed []genTile) {