Skip to content

Decisions

Choices made on the maintainer's behalf, dated, with the reason. Overturn any of them by editing the row and the code.

Date Decision Why Alternatives considered
2026-09-25 Python 3.14 with uv, src layout, PyInstaller --onefile for the Windows binary the knowledge base's conventions (07-project-conventions.md); the repository already carried a Python .gitignore C# console app beside the launcher (would share types, but adds a .NET toolchain to a tool that mostly polls HTTP and watches files)
2026-09-25 Drive the official launcher instead of starting the client ourselves the client needs the account token, the content database and the engine signature that only the launcher holds (reference §8) re-implementing the launcher's Connector in Python (large, fragile, and a security concern around the token)
2026-09-25 Primary way to send a connect: run bin_x64\SS14.Launcher.exe ss14://… with DOTNET_ROOT=<install>\dotnet_x64; the named pipe is an optimisation to add once the exe path is verified the exe path works whether or not the launcher is running (it forwards over the pipe or becomes the launcher and queues the connect); the bootstrap exe drops arguments so it cannot be used; there is no ss14:// protocol registration on Windows os.startfile("ss14://…") (no handler registered); writing the pipe directly (needs the pipe name and a running launcher; keep as optimisation)
2026-09-25 Detect the join outcome from client.stdout.log, with the launcher log for command acceptance and PID, and the process list for cleanup the launcher does not know the outcome; the client stays open on failure; the client logs every connection phase at Debug level and the server's reason text (reference §9, §10) player-count change in /status (ambiguous), window title (constant), UI automation of the Retry button (fragile)
2026-09-25 On failure, terminate the failed client and relaunch through the launcher rather than pressing Retry in the client no programmatic way to press Retry; a fresh client is not bound by the 30 s delay the UI shows UI automation (fragile, focus-stealing)
2026-09-25 Default poll interval 3 s, offline back-off 10 s, attempt timeout 90 s, cooldown 2 s after a "full" failure, stop on non-retryable reasons status requests are cheap and the launcher itself uses a 5 s timeout; a content download on first connect can take a minute; retrying a ban or whitelist denial is pointless honouring the server's delay between attempts (only enforced in the client UI, and each attempt already costs a client start-up)
2026-09-25 Stop after a successful join by default; --rejoin opt-in the task is to get in; a player who was in the round bypasses the cap on reconnect, so automatic rejoin adds little and could fight the player who closed the game on purpose always rejoin
2026-09-25 Restarting a stuck launcher (errored overlay) is opt-in it kills a process the user may be looking at; the common cause (server /info unreachable) is rare when /status was just reachable always restart; never restart
2026-09-25 Standard library urllib for HTTP; psutil for processes; no other runtime dependencies for now keeps the PyInstaller binary small and the frozen import list short; psutil has wheels for Windows and Linux httpx/requests (not needed for two GET endpoints)
2026-09-25 The cloud session's uv was upgraded through pip (0.8.17 → 0.12.19) to install Python 3.14.7 the container's uv only knew 3.14.0rc2; uv self update is blocked by an unauthenticated GitHub API rate limit pinning the rc (works, but tests should run on the release)
2026-09-26 Proceed with the Steam launcher build as if it were the open-source one (same layout, same log lines, --cvar branding.steam=true added), pending verification item 2 the reference launcher log shows the Steam install at <Steam library>\steamapps\common\Space Station 14 Playtest; the private additions are Steam branding and multi-window, not the command path stop until the Steam source is known (not available)
2026-09-26 Default target is Wizard's Den Lizard, ss14s://lizard.spacestation14.io/server it is the server in every log the maintainer sent Leviathan (also favourited)
2026-09-26 The app gets a Tkinter window with a settings panel (launcher folder, server, interval, options) backed by a TOML file; the GUI moves from optional to milestone M3 the maintainer asked for a configuration panel; Tkinter needs no extra dependency and keeps the PyInstaller build small a CLI-only tool with a config file; a web UI (heavier, no benefit for one user)
2026-09-26 Launcher discovery order: configured path, running launcher process, launcher log, Steam libraries (registry, Program Files, libraryfolders.vdf), common standalone folders the maintainer asked for a matrix of common locations checked before asking; the log and process hints are exact on any machine that has run the launcher once asking every time; scanning whole drives
2026-09-26 Accept any dotnet* folder with dotnet.exe as the runtime, preferring dotnet_x64 a first report said dotnet_x86; the dir later showed dotnet_x64, so this is now just tolerance requiring dotnet_x64
2026-09-26 Sending a connect is always bin_x64\SS14.Launcher.exe <uri> with DOTNET_ROOT, whether or not the launcher runs; the tool never waits on that process verified both cases on the reference Windows machine; the process is the launcher itself in the cold case named pipe (unverified, unnecessary)
2026-09-26 Settings live in %APPDATA%\ss14-autojoin\config.toml (SS14_AUTOJOIN_CONFIG overrides), a flat table written by the app; command-line flags win over the file one user, one machine; TOML reads with the standard library; a flat table needs no writer dependency the launcher's own data folder (not ours to write); INI; JSON
2026-09-26 The window is one screen: settings on top (folder with Browse/Detect, server, numbers, options), Start/Stop and Save below, a scrolling log underneath; the joiner runs in a thread and reports through a queue thin GUI over the existing loop, nothing to test beyond form parsing; matches the "configuration panel" request tabs, a tray icon, a separate settings dialog
2026-09-26 Two PyInstaller builds: SS14AutoJoin.exe (window, no console) and ss14-autojoin.exe (console tool); CI builds both on Windows on every push and uploads them the window is what the maintainer will click; the console tool is what he pastes logs from one exe with a --gui switch (a console window would flash for the GUI)
2026-09-26 The join sound is synthesized in the repository (tools/make_fanfare.py, C5 E5 G5 C6 brass "ta-da", 1.5 s) and dedicated to the public domain, rather than a downloaded clip the maintainer asked for a free trumpet success sound; a generated file has clean provenance and can be tuned; no new dependency a Freesound CC0 clip (licence tracking, attribution habits vary); Windows system sounds
2026-09-26 The app icon is generated (tools/make_icon.py: ring station, lit hub, solar panels; PNG + multi-size ICO) and public domain, like the sound the maintainer asked for a free SS14-themed icon; the container reaches only GitHub, where icon sets are MIT/CC-BY with attribution duties; a drawn icon has clean provenance and can be tweaked Tabler/Lucide "satellite" or "planet" icons (MIT/ISC); a Wikimedia public-domain image (unreachable here)
2026-09-26 Keep both icon paths in the window: Tk's iconbitmap (also sets the default for later toplevels) and a WM_SETICON through ctypes after the window is mapped on Windows with Python 3.14's Tk 9.0.4, iconbitmap returned success but the title bar kept the default icon; the Windows API call fixed it (verified 11:28) Tk only (fails on Tk 9); Windows API only (message boxes would lose the default icon)
2026-11-01 The winget generator declares the newest manifest schema version at the time of writing (1.28.0) and the tests pin a floor; an open submission is fixed by editing the three files on its branch, not by a new pull request winget-pkgs deprecates schema versions and its bot rejects older ones even on a pull request that already passed validation; the package's fields exist unchanged in the new schema resubmitting through wingetcreate (opens a duplicate pull request that moderators close); staying on the old version (blocked)