Skip to content

Project Tracker

Pummelchen edited this page Sep 17, 2026 · 24 revisions

Project Tracker

The single place that answers "what is being worked on, what is done, and what is known to be wrong". One row per item, whatever kind it is — a bug, a piece of work, a decision not to change something, or something waiting on a person.

Gate: ./tests/run.sh — the whole suite, 29 files and 1932 checks. CI runs it on every push and pull request. A row is Done only when it is committed to main and that is green, not when it "should work".

Version: roomcad/web/version.js is the only release source. 10.7 is deployed to https://roomcad.91.99.176.243.nip.io/, and roomcad/server/deploy.sh is what put it there. There are no tags and no releases yet.

Adding to this: a new discovery gets a row here in the same session it is found, with its status — including the ones that are decisions rather than defects, so nobody "fixes" them by accident.

Codes: A audit pass · B bug · T task · D deliberate, not a bug · H waiting on a human · N note. Code-level traps are not repeated here — they live in AGENTS.md, which is this repository's one instruction file.

Counts: 55 Done · 5 Open · 8 Won't fix · 1 Waiting · 4 Reference.

ID Item Status Outcome / evidence
A1 POST /api/login crashed with TypeError on a non-ASCII password — no response, dropped connection, and the failure was never counted, so it escaped the throttle entirely Done password_matches() compares UTF-8 bytes; a non-ASCII password now also works
A2 _read_json() returned a list or scalar for [], 5, "x", and callers reached for .get(...) — an unhandled AttributeError on /api/live-check/ Done every body goes through _read_json_object(), which answers 400
A3 get_conn() was an unlocked check-then-act: a leaked connection and a double init_db/migration Done double-checked under a lock
A4 migrate() unlinked each legacy file before committing its row, so a crash lost the room from both places Done the row is committed before the file is removed
A5 delete_room() left the room's LIVE_SEQ entry behind, one per name ever seen Done pruned with the draft
A6 rooms.db.sql published two real saved rooms in a public repository Done structure only; restores into an empty file
A7 The API ran as root with no sandboxing at all Done User=roomcadapp plus ten sandbox directives; systemd-analyze security 9.4 UNSAFE → 5.3 MEDIUM
A8 encode gzip sat at site level, wrapping the SSE stream Done moved onto the static handle
A9 .gitignore left __pycache__, the downloaded Caddy and rooms.db stageable Done all ignored
A10 A drag torn down by blur or pinch left dragTransactionActive stuck, so the client silently stopped accepting teammates' edits Done one abortDrag() path for every non-committing teardown
A11 Undo during a drag destroyed the user's most recent real history entry Done one owner for the drag snapshot; undo and redo cancel the drag
A12 liveSeq was never reset per room, so the first edit in every newly-opened room was refused as stale and discarded Done reset when the watched room changes
A13 A failed live push left liveUnpublished set, permanently disabling the drift check Done cleared on failure
A14 Walking in 3D also fired the 2D tool shortcuts (W, D, Delete) Done 2D-only keys gated on the mode
A15 Command-Z inside a text field undid the drawing instead of the text Done text-editing chords are left to the browser
A16 New Room and Open in 3D left the 3D view and its render loop running behind a 2D plan Done the change handler follows the mode
A17 A plain click with the measure tool committed a "0 cm" measurement Done a real drag is required
A18 #version-modal had no CSS rule at all Done shares #open-modal's overlay
A19 drawFixture built an invalid colour ("rgb(...)" + "30") and silently painted the wrong one Done a real rgba()
A20 dimensionBoxes was cleared after being filled, and m² captions compared device pixels against CSS pixels Done both fixed
A21 A load naming an unknown furniture kind threw, so the defensive loader could not load the documents it exists for Done unknown kinds are dropped, as other unrepairable objects already were
A22 A hand-edited diagonal wall survived the whole load path, so detectRooms reported the wrong room count silently and the 3D collider built a Z-aligned box for it Done such walls are dropped on load
A23 splitWallsAtJunctions let two junctions 4 cm apart produce a 4 cm piece that sanitize then dropped on reload, so a boundary became a hole Done cuts keep MIN_WALL_LENGTH apart
A24 The width/length clamp claimed a 2–20 m range that syncExtent() overwrote on the next line Done the extent is derived; the clamp that remains is the plate's real one
A25 The 3D loop never stopped — after one visit to 3D, WebGPU, Rapier and the 680-car traffic model ran at 60 fps behind the 2D editor Done pause()/resume(), wired to the mode switch
A26 The room was built twice on first 3D entry, the first build discarded unrendered and unlit Done build() refuses to run before an environment exists
A27 The sky dome was 200 m and the walkable city ≈272 m, so you could stand outside it Done sized from the city's reach, and it follows the viewer
A28 Physics read the traffic one frame behind what was drawn Done the city advances before the physics that consumes it
A29 Lights and their shadow maps were dropped on every rebuild and never disposed Done disposed in disposeScene
A30 The suite had no runner: no timeout, no totals, and a hang produced no usable exit code Done tests/run.sh, proven to fail on an injected assertion and on a timeout
A31 Nothing ran the suite automatically, and deploy.sh shipped without a gate Done CI on push and pull request
A32 city-physics was genuinely flaky — it failed on a machine that was merely busy Done the traffic's runtime randomness is swappable and the test seeds it; byte-identical across runs and green under six-way parallel load
A33 THIRD_PARTY_NOTICES.md did not exist, though RELEASE.md §1.6 requires it — Rapier (Apache-2.0) was redistributed with no licence header Done notices plus both licence texts
A34 AGENTS.md was wrong about the repository in four places Done corrected, along with the server README, the root README and roomcad/README.md
A35 The GitHub description said "Shophouse room planner for macOS"; the homepage and topics were wrong too Done corrected through the API
A36 Opening-width limits existed in four places that could drift apart, and isTyping() existed twice with different rules Done both single-sourced
A37 122 new checks across eight test files, and nothing proving they could fail Done each was verified to fail against the previous code
B1 Ceiling lights were dealt out in plan order, so the 17th was dead wherever you stood — and still drawn glowing, so it read as a broken lamp Done the pool follows the viewer, nearest-first, and over-budget plans are reported in the inspector. tests/room-lights.test.mjs drives the real pool methods and proves the assignment is nearest-first, follows a moving camera and is not redone once settled
B2 The only shadow-casting street lamp used bias = -0.004, the class of value the room's own comments call pure light leak Done zero, with a normal bias derived from its texel size as the sun does. tests/plan-seal.test.mjs resolves every shadow.bias in the file — literal or behind a constant — and refuses a negative one
B3 detectRooms() gave up above 60 000 cells and returned nothing, so every m² caption vanished, the floor area became the bounding box and every wall became draggable Done a real memory budget (1 000 000 cells) plus roomDetectionSkipped(), surfaced as an inspector warning. tests/plan-editing.test.mjs proves a plan the old cap rejected now measures, and one past the new cap reports the skip rather than answering wrongly
B4 The endpoint drag allowed 15 cm walls while drawing required 30 cm — the file-repair threshold used as the editor's rule, so 20 cm was legal to hold and impossible to make Done two named thresholds, the drag on the editor's, and a legacy short wall can be turned or lengthened but not shortened. tests/wall-lengths.test.mjs covers all four cases
B5 A bus drove 13 m off the edge of the city at full cruise. Two defects: the recorded junction decision held only a road index (and an index is not a junction — index 0 the two ways names two different junctions), and the stop-for-a-space path overrode turn/mustTurn without clearing that record, so a compulsory turn at the outermost junction could be lost Done the record is _junctionId(axis, dir, index), the override clears it, and _decideTurn refuses to act on a "carry on" record where there is no road ahead. Found by fuzzing once the gate was made deterministic; the seed is hunt-42, and with the fixes reverted the escape reproduces. tests/city-turns.test.mjs pins all of it
T1 login.js — the password gate — had zero tests. It has no exports and runs its session probe at import Done tests/login.test.mjs, 23 checks: the probe shows or hides the form and fails closed when the server is unreachable; the 401 hook app.js calls is the same routine; a correct password reloads, a wrong one does not and says so; a network error is reported separately; the field is emptied before the request goes out. Proven to bite by three injected regressions
T2 app.js coverage was a few lifted fragments, because it could not be imported at all — it imports walk3d.js, and the vendored addons behind that import the bare three themselves Done tests/harness/three-resolver.mjs resolves the page's import map for the whole graph, so the real module is loaded: app-wiring.test.mjs boots it, lets it adopt the room the server names, fires its own keydown handler, and checks the live stream opens; app-clicks.test.mjs clicks the real buttons; app-internals.test.mjs covers the escaping boundary, the digest, the export name and the footer thresholds
T3 tests/editor-fuzz.test.mjs compared CSS-pixel coordinates against canvas.width — device pixels. It passed only because installDOM() defaults dpr to 1 Done the helper uses the CSS box, and the block now runs against a deliberately high-DPI backing store and asserts that it is not vacuous. Against the old helper it fails with 939,861 on 1200x800
T4 Nothing limited how many live SSE streams a client could hold, and request_queue_size was socketserver's default of 5 Done MAX_WATCHERS_TOTAL 256 protects the process and MAX_WATCHERS_PER_SESSION 16 stops one client starving the rest; both are enforced before the SSE headers are written, so a refusal is a clean 503 that registers nothing. The backlog is 128
T5 The only CSRF defence was the SameSite=Lax cookie attribute — no token, no Origin check — and POST /api/logout is deliberately unauthenticated Done _require_same_origin() refuses a POST/DELETE whose Origin, or Referer, does not name the request's own Host. A request with no Origin is still allowed — curl, the suite and same-origin form posts send none — and /api/login is covered deliberately, because a cross-site login can force a session and burn the throttle
T6 serve.sh and install-caddy.sh downloaded Caddy and executed it with no verification at all Done both check the archive against the SHA-512 the release publishes, before extracting it. A mismatch, a missing entry, or a host with no hash tool aborts without running anything. Verified against a real release: genuine verifies, corrupted is rejected, and the logic was re-run under macOS's bash 3.2
T7 install-caddy.sh's default target list included /var/xaios_updater, another project's directory Done the default is RoomCAD's own path only, and an explicit list is still honoured
T8 Dead code: CORRIDOR_WIDTH, frontageLength, the write-only hasDoor/hasWindow sets, the vestigial score/alternatives, Editor2D.displaySize() Done all removed, each confirmed unreferenced first. detectRooms()'s hasDoor field was kept — svg.js and four test files read it
T9 Nothing checked that the page can boot: the import map, the module graph, the element ids Done tests/boot.test.mjs, a boot contract with no browser and no dependencies. It walks the real import graph from both entry modules, resolves every specifier through the import map, refuses an element id the code looks up but the page lacks, and refuses a module nothing loads. All four were injected and caught. It proves the graph and the DOM contract, not that anything renders
T10 walk3d.js's cityReach() recomputed city.js's reach formula from four of its constants — a copy that had already drifted once Done City.reachFor(bounds) is the single source, and the four constants left walk3d's import line
T11 Per-vehicle-per-frame string keys for the turn control, and a fresh object from _nextJunction for every vehicle every frame Done the turn-control keys are interned once per city, with strings identical to the hand-built ones; _nextJunction takes an optional object to fill, and the two per-frame callers reuse a per-vehicle scratch while the callers that hold a result still get a fresh one
T12 Nothing asserts that furnitureFootprint() and furnitureCenter() refuse an unknown furniture kind — a deliberate decision held by a comment, not by a test Open add a test that an unknown kind is refused, and that the three guarded call sites never reach it. Small
T13 The non-axis collider branch in walk3d.js is unreachable in the app, and no test executes it — "defensive" currently means never run Open drive a slanted wall through it directly, so the guard is verified rather than assumed
T14 Walk3D.dispose() has no caller and no test; the .dispose() calls in the suite are all City.dispose() Open build one via Object.create(Walk3D.prototype) with fake fields and assert teardown runs clean. walk3d.js is importable now, so this is finally possible
T15 The vendored Rapier build carries no version string, so which upstream release it is cannot be determined from the repository Open the version cannot be recovered from a minified wasm-bindgen bundle, but a SHA-256 of rapier.mjs can be recorded with a test that fails when it changes — so an upgrade can never again be silent
T16 sanitize() repairs and drops silently: walls under 15 cm, doors on too-short walls, labels with no centre, furniture of an unknown kind, diagonal walls Open a judgement call, not a defect: it could report a count the way roomDetectionSkipped() already does. "Quiet" is deliberate, and this changes what a user sees, so it needs a decision rather than a patch
T17 Deploy 10.7 to production — the first deploy to run the hardened unit, which creates the roomcadapp account and moves ownership of the database Done rehearsed first in a scratch directory against a copy of the live database, then backed up (sqlite3 .backup to /var/roomcad/backups/predeploy-10.7-20260916-230142.db, 35 rows before and after) and deployed. Verified: version.js serves 10.7; the unit runs as roomcadapp with ProtectSystem=strict; /var/roomcad itself and all three database files are roomcadapp-owned; no errors in the journal; GET / 200; /api/rooms 401 unauthenticated; the legacy :8443 links answer 200; a real login through the public URL returned the room list; an authenticated stream opens with 200 text/event-stream
T18 Rehearse the hardened deployment before trusting it, since it had never run: production was on the old unit, as root and unsandboxed, and roomcadapp did not exist Done the unit was stood up in /var/roomcad-rehearsal (since removed) against a copy of the live database, under the same sandbox directives, as an unprivileged user. It started, refused an unauthenticated call (401), accepted a login, read the live rooms and saved one — which is also what proves ReadWritePaths lets it write, WAL sidecars included. It also demonstrated the failure it prevents: with the directory left root-owned the app dies on PRAGMA journal_mode=WAL (see N2). Separately: rsync --delete has nothing to destroy (the live web/ and the repo's hold the same 25 files), the Caddy chain is consistent (roomcad-caddy runs as roomcadweb, its Caddyfile root:roomcadweb 640), and T5's check was run against production's own header shapes — 443 → 200, the legacy :8443 with the port forwarded verbatim → 200, :8443 with the port stripped from Host → 403, cross-site → 403, no Origin → 200
D1 furnitureFootprint() and furnitureCenter() stay strict about an unknown kind Won't fix sanitize() guarantees the invariant on load and the reachable call sites now guard, so a fallback here would hide a broken invariant rather than handle one
D2 walk3d.js's non-axis collider branch is unreachable Won't fix left in place deliberately: it is the collider's own guard against a wall it cannot represent, and it should not depend on another module's filter for correctness. See T13 for the test it still lacks
D3 walk3d.dispose() has no caller Won't fix the walkthrough lives for the page's lifetime, so the teardown path is defensive only. See T14
D4 GET /api/session/last answers null with a 200 on an empty database Won't fix odd for REST, but a pinned contract in tests/server-live.test.py
D5 POST /api/logout needs no valid session Won't fix an already-invalid session must still be able to clear its cookie
D6 sanitize() repairs and drops silently Won't fix it is the loader's defensive pass and is meant to be quiet. See T16
D7 The traffic is deliberately not reproducible in production Won't fix the layout is seeded but what the cars do is real Math.random, so no two reloads drive identically. setTransportRandom() exists so a gate can pin one drive, and nothing in the app calls it
D8 The vendored Rapier build carries no version string Won't fix recorded in THIRD_PARTY_NOTICES.md. See T15 for the hash that would at least make a change visible
H1 Look at the 3D in a WebGPU browser, at night Waiting the light-pool and shadow-bias fixes were verified by lifting the real methods and by resolving every shadow.bias, but nothing in this repository renders the scene — Walk3D needs WebGPU and cannot be instantiated headlessly, and tests/boot.test.mjs deliberately stops at the module graph. If the normal bias derived for the street lamp's 512² map is wrong for its reach, the symptom would be shadow acne on lit street surfaces
N1 Which of the green checks are evidence, and which are only guards Reference the checks that cannot fail without the fix are named where they live: the 10 acceptance guards in tests/server-live.test.py (including "no Origin is still accepted") and the "it still works" halves of the login and boot checks. They catch a regression in the fix; they do not detect the absent feature. Nothing else in the suite is in that category
N2 The service user must own /var/roomcad itself, not just rooms.db Reference SQLite in WAL mode creates rooms.db-wal and rooms.db-shm beside the database, so a root-owned directory fails at boot on PRAGMA journal_mode=WAL with "unable to open database file". deploy.sh's install -d -o roomcadapp /var/roomcad is what prevents it — do not "simplify" that into a chown of the three files
N3 A fresh query string does not re-evaluate an already-imported ES module Reference the first version of login.test.mjs was built on the opposite assumption and was silently inert: every scenario after the first shared the first one's result, and the checks "passed" against a module that had not run. It loads a fresh copy at a fresh path per scenario instead
N4 A test that configures its own preconditions can hide the bug it is for Reference T3 passed for months because installDOM() defaulted dpr to 1, making CSS and device pixels the same number. The fix is not only the right comparison but asserting that the two spaces differ, so the check cannot quietly go vacuous again

Clone this wiki locally