Repository navigation
Implement Android app features and enhance release readiness - #10
Merged
Merged
Conversation
…ures - Added support for fullscreen detection and handling in the Android app. - Implemented app update prompts for outdated versions in the Android app. - Configured production API URL for the Android app in a new .env.android file. - Created documentation for building and testing the Android app. - Developed build script for the Android app to ensure proper configuration and assets. - Added tests for Android-specific functionality and configurations. - Integrated native back button handling and lifecycle management for the Android app. - Ensured system bars are hidden and safe areas are respected in the Android app.
The audit of the ranked backend, replay security and the failing touch/menu browser suites, moved from a scratch file into the repository. Probe output is kept as evidence; references to the scratch folder are described instead of linked. File and line references are to a8acf66. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Chain Lightning added a seventh entry to POWERUP_INFO, but the guide still laid rows out at a fixed 74 px pitch from y=130, so the last row (Revive) and its two-line description ran under the BACK button at every phone and desktop size. The rows now start higher and their pitch shrinks to fit the space between the title and BACK (capped at the old 74 px). Found by the menu.mjs layout review once the suite could run past its stale PLAY tap. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Point-to-Fly "four diagonals" check failed at random (2 of 5 runs in the audit). recentre() emptied the asteroids but never the UFOs: a UFO was alive in every direction case, and when it hit the ship (lives 99 -> 98) the ship respawned facing 0 and was still re-turning when the window ended. recentre() and quiet() now remove UFOs and their shots and stop new ones; the failure detail includes the ship's lives so a hit shows up. Movement section: 6/6 clean runs afterwards, against 1 hit in 4 before. The "Assisted Flight gets going up sooner" check measured between two wall-clock snapshots, so its window varied by a tick or more (~5 px) with browser latency, once landing inside the 10 px margin. It now traces sim.step and measures exactly 0.35 s of simulation time from the step that first received the flick. Same assertion and margins; the result is now identical on every run (Assisted 91 px up / 26 px sideways, Point-to-Fly 75 / 55). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
5033ac8 made PLAY on a one-player phone the ranked entry (casual play is PRACTICE) and a740fe1 / ccce3ce changed some texts, but touch.mjs, menu.mjs and ranked-browser.mjs still expected the old behaviour. Each stale PLAY tap timed out and aborted the rest of its suite. - touch.mjs / menu.mjs: tap PRACTICE where a casual game is wanted; a phone in keyboard mode keeps the PLAY / PRACTICE entry, so check that, and that a tap on PRACTICE asks for a keyboard first and a finger cannot press START WITH KEYBOARD; ?touch=0 is checked by the touch-on button rather than the old "PLAY WITH KEYBOARD" label; the rotate-prompt check taps where PRACTICE is (a tap on PLAY never starts a game on a phone). - menu.mjs: the legend lists six one-player power-ups (LIGHTNING / Chain hits, and MEGA POWER says "4 powers!"). - ranked-browser.mjs: nicknames are shown as typed, and the outage page says "YOU CAN STILL PLAY PRACTICE.". touch 360/360, menu 127/127, ranked-browser 61/61. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The router counted every POST /runs/:id/submit against submit:global before submitRun had loaded the run or checked its token, so junk requests to random run ids from many addresses could use up the hourly verification budget (default 1200) and make every honest submission fail with 429. The per-address bucket still runs first, before any work. The global bucket is now charged by submitRun once the run exists, the caller has shown its token and the run is still open (not rejected or expired), just before the body is read. Authentication, the per-address limit and the global ceiling on verification are unchanged for requests that can actually reach verification. Regression tests: unknown runs, malformed ids, missing and wrong tokens leave the global counter at zero and a real run still verifies; token holders are still capped (third of three gets 429 and stays issued); expired and rejected runs answer 410 / 409, not 429. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Until now the Worker verified exactly one gameplay rules version, compiled in from src/replay, and ignored what the client played. Since PLAY on a phone is ranked, the first rules bump after the Android release would have broken the main button of every installed app, and rejected any run in progress at deploy time (see docs/android-release-audit.md, #1). - server/rules/<version>/engine.mjs: the simulation + verifier for one rules version, built from src/replay exactly like dist/node (one self-contained module, no imports), plus engine.d.mts and gzipped reference replays with the results that engine computes. Written by `npm run rules:freeze`; hashes recorded in server/rules/versions.json; stored byte-for-byte (-text) so Windows and CI agree. 1.2.0 frozen from a8acf66 (identical to today's dist/node build). - server/src/engines.ts: the explicit allowlist. Untrusted version strings only match the registry's own keys (no prototype keys). - Issue: POST /api/v1/runs?rules=&schema= gets that version when it is registered with that schema; anything else, or no query (existing web clients), gets the current version as before, which the client already refuses with "update". No new error path, no client change needed for old builds. - Verify: the engine is chosen by runs.rules_version, stored at issue, never by the submission; a replay naming other rules is rejected before any engine runs; an unregistered version is rules_mismatch. Tokens, idempotency, claim races, limits, elapsed-time and expiry rules are untouched. No migration: each engine reads exactly one schema, so the stored rules version fixes both. - Health adds supportedRules; rulesVersion / schemaVersion keep their meaning (the current version). - Client: requestChallenge names GAMEPLAY_RULES_VERSION and REPLAY_SCHEMA_VERSION in the query string (keeps the request CORS-simple). - The Worker bundle now contains no live simulation code: 121 KiB (32.6 KiB gzipped) in a wrangler --dry-run build. Tests: issuing by version (fallbacks, hostile values, body ignored), a 1.2.0 run verified by the frozen engine on a server whose current version is newer, a rules update mid-run not invalidating the run, a submission unable to pick its verifier, retirement. Browser suites match the challenge request by path now that it has a query string, and ranked-browser / worker-e2e check the query and supportedRules for real (worker-e2e also stops expecting the old 1.1.0). The chain-lightning verifier test spies on the frozen engine's GameSim, which is now the authoritative one. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- server/rules/released.json: every Android build handed out, the rules it plays, and when a successor replaced it; policy.supportDays = 90. The release candidate (1.0, versionCode 1, rules 1.2.0) is recorded with distributed: null, so 1.2.0 is protected from now on. - server/tests/rulesCompat.test.ts (runs with npm test, before the Worker deploys): engines.ts registers exactly the versions in versions.json; the current GAMEPLAY_RULES_VERSION is frozen and registered; every version a released build still needs is supported; each frozen file matches its pinned hash and nothing is added or missing; engines import nothing; the reference replays give the recorded results through their frozen engine, and the current version's also through the live sources, so a gameplay change without a rules bump fails here (checked: a 1-degree turn-speed change fails 4 reference checks and the lifecycle tests). - scripts/rules-guard.mjs (CI step, against the merge base; checkout now fetches full history): frozen files may be added, never modified, and deleted only once retired; versions.json entries never change; released.json records are permanent and only gain their dates, without backdating supersededOn. node:test runs it against scratch git repos. - scripts/production-health.mjs: after deploy, the Worker's supportedRules must include the current rules and every version distributed builds still need. - scripts/build-android.mjs: refuses to package unless the app's rules are frozen and its versionCode is recorded in released.json with those rules. Its tests now run in CI too. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…licy server/rules/README.md describes the frozen engines, how a version is chosen at issue and verify, what each check protects, the rules-bump procedure (backend serves both versions -> web and the new app ship -> the old version retires after its support period) and the initial retirement policy: Android rules stay supported at least 90 days after their last build is superseded (CI-enforced) and until they are under about 2% of issued runs; web-only versions at least 7 days. leaderboard-backend.md, deployment.md and android.md drop the one-version-at-a-time description (android.md keeps it as history), and the audit gets a status note listing what has been done since. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.