[283375]

• Rework play result warnings into enum-driven model
• Fix warning correctness
• Add validator unit tests
• Add a hint and a detailed message per warning
• Add docs/play-result-mechanisms.md for future reference
• Moved to :shared (Prepare multiplatform support #24):
  • ArcaeaFormatters
  • Warning UI components and their string resources
  • Basic text/UI helpers and DISABLED_ALPHA

[283375]

/opencode Review this PR. Avoid external directory access (/home, /tmp, etc.) as it will block unattended session with permission requests.

[opencode-agent]

Review: Play result validator improvements

Overall this is a solid refactor. Moving the validator into :shared, replacing the ad-hoc warning classes with a PlayResultWarning enum + PlayResultWarningDetail, and centralizing the null-handling policy is a clear improvement, and the new PlayResultValidatorTest (974 lines) covers the tricky cases well. I found no blocking defects.

Findings

1. Historical Room schemas/migrations were edited (shared/schemas/.../{7..13}.json, Migration_6_7.kt:19, Migration_7_8.kt:84)
The 100000000.0 → 10000000.0 and score >= 100000000 → score >= 10000000 changes are a real bug fix (the old factor was 10× too large; the Kotlin PlayResultCalculated had the same bug and is fixed too). But editing already-shipped migration sources/schema snapshots is non-standard. It doesn't repair DBs that already ran those migrations, and schemas 4–6 still carry the old buggy SQL, so the historical record is now inconsistent. Confirm this is only for migration-test consistency, since views are dropped at v14/15 anyway.

2. Integer overflow on corrupt inputs

• PlayResultWarning.kt:17: 10_000_000 + notes overflows for notes > Int.MAX_VALUE - 10_000_000, making SCORE_IMPOSSIBLE's maximum negative and SCORE_OUT_OF_RANGE's guard always return early.
• Common.kt:20-21: base.toInt() wraps for very large pure/far (e.g. notes = 1, pure = 2_000_000_000).

The validator is specifically for corrupted data, so using Long/clamping here would be more robust. Low severity (real charts are small).

3. Behavior change: score 0 with positive judgments and no chart info is no longer flagged (PlayResultWarning.kt:49-62)
EMPTY_PLAY_RESULT only fires when score and all counts are zero/null. The old SCORE_IS_ZERO warned on any score == 0. With a missing ChartInfo, a record like score=0, pure=5, far=0, lost=0 now produces no warning (SCORE_OUT_OF_RANGE needs notes). If this is intentional, ignore; otherwise a cheap score == 0 && (pure > 0 || far > 0) check would preserve the old coverage.

4. Nit: missing final newline in app/src/main/res/values/strings.xml and app/src/main/res/values-zh-rCN/strings.xml, while .editorconfig sets insert_final_newline = true.

5. Nit: ArcaeaPlayResultClearType.toDisplayString() (ArcaeaPlayResultClearType.kt:16) now hardcodes English Title-Case labels in core/constants. It's still unlocalized (it was uppercase English before), so not a regression, but presentation strings in core are a layering smell and the app is bilingual. Consider resolving these via Compose resources where displayed.

6. Scope note: 2,083 additions mix validator refactor, score/precision fixes, DB migration/schema edits, resource moves, and label unification. The individual commits look coherent, but splitting the score/migration bug fixes from the validator refactor would make this easier to review and revert.

Verified

• All old ArcaeaPlayResultValidator* / helpers.ArcaeaFormatters / helpers.*ItemAlpha references were updated; no stale imports.
• Warning/action/message resource keys are complete and consistent between values and values-zh-rCN; no leftover app play_result_validator_* keys.
• New score formula matches the intended floor((pure·10M + far·5M)/notes) and the added sweep over note counts 1..3000 is a good regression guard.
• Null-value policies in the enum KDoc match the implementations, and CLEAR_TYPE_MISSING remains B50-only as before.

github run