Settings -> Enable Custom Bitmaps was a static bool with no backing store:
it defaulted to off on every launch, so an operator had to re-tick it each
session and any custom plasma art was silently ignored until they did. New
ConsoleSettings persists machine-level menu toggles as XML in
%ProgramData%\Tesla Console\console.settings, alongside RPDefaults.rpd /
BTDefaults.btd / local.siteconfig. It is loaded once from Main and never
from a static initializer: the differential suite drives PlasmaBitmaps
directly and must keep seeing the original defaults rather than whatever
this machine has saved. A missing file is the first-run case; a corrupt one
is ignored and rewritten by the next toggle, because losing a menu setting
must never stop the console starting.
Custom art is now version-controlled and rolls with the release. New
Console\Plasma Images\ is copied into the package, and the lookup searches
%ProgramData%\Tesla Console\Plasma Images first and the exe-relative folder
second, so a release can never clobber a site's own name bitmaps.
install.bat creates the data-dir folder before the icacls grant so an
unelevated operator can write it. Ships with three 128x32 name bitmaps
(Deadmeat, Muerte, Phrogg); Muerte arrived as "Muerte_128x32-2.bmp", a name
the lookup can never build, so it is renamed to match its pilot.
Two fixes fell out of making the flag sticky. Path.Combine ran on the raw
participant name outside the try block, so with the option on a pilot named
"A:B" threw ArgumentException straight out of egg generation; names that
cannot be a Windows file name now just render procedurally. And the art was
loaded with Image.FromFile, which keeps the bitmap backed by the file and
locked for its whole lifetime — it is copied out through a stream now, so
art can be swapped between missions without restarting the console.
Verified against the built net40 exe: all three shipped bitmaps resolve by
pilot name, ProgramData wins over the shipped copy, a wrong-size file is
ignored, an illegal-character name does not throw, the file is not left
locked, and the settings round-trip and corrupt-file tolerance both hold.
Diff suite 106/106.
Version bumped 4.11.4.4 -> 4.11.4.5 across Console, Launcher, vPOD, the
install/build banners, the diff-suite version assertion and the README's
latest-release pointer.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The eight internet pod rows are furniture the operator builds once in
Manage Site — the slot-to-IP map is frozen — but which of those slots a
human actually claimed changes every session. New SessionRoster reads the
roster TeslaLobby writes at the state=launching flip, and both game panes
grow a session strip above Mission Properties: a banner (session key, game,
slots claimed, waiting, written-at), Apply Session, and the Reset Pods that
until now existed only as a right-click on a Go button that is disabled
exactly when the reset is wanted.
Two properties shape all of it. Only claimed slots appear in the file, so
absence is the unclaimed signal and every failure path — truncated, stale,
unreadable, refused — degrades to "no roster", which is byte-for-byte
today's arcade behaviour; museums run this software and a bad JSON file
must never stop a mission that would otherwise run by hand. And enabled
implies claimed, not the reverse: the operator may always sit a pilot out,
never add one, because an enabled row nobody claimed puts a dead IP in the
egg and the pods then wait on a peer that will never boot. Roster issues
join the pane's existing issue text, so the Go button is still the gate.
The roster is one file per launch generation and deliberately not a live
view: pod peer tables are boot-static, so a player whose lobby crashed is
still in every pod's table and still playable, and live tracking would
evict that working pod mid-session. Poll runs at ~1Hz off the existing
network timer with its own deadline, and the strips are built at runtime —
InitializeComponent is decompiled 1995 designer output that the
differential tests compare literally.
Go/Load also re-checks CheckAllValues at the click instead of trusting the
last status tick: a pod that died in that gap went straight into the
mission, and the post-Load barrier in NetworkScan then waited forever for a
WaitingForLaunch that never arrives.
vPOD gains -bind <ip>. Both listeners defaulted to IPAddress.Any, so a
second vPOD on the machine lost the port and the console could only ever
see one fake pod; binding each instance to its own address runs a whole
eight-pod session side by side with no cockpits. Bind all of them or none —
Windows lets IPAddress.Any take a port that specific addresses already
hold, and the unbound instance then answers for every slot nobody claimed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The provisioning text on the pod's plasma panel came up garbled — stale
content bleeding through, new text overlapping at the wrong position.
Root cause recovered from the dumped PD01D221 controller firmware
(vrio/PlasmaNew): PlasmaWriter.ClearAll() sent ESC J, which on this
controller is NOT a clear — it toggles an orientation/mode bit — so the
panel never cleared and the cursor never homed.
Fix: ClearAll() now sends the real commands ESC @ (clear active buffer,
reset text state) + ESC L (home to 0,0), and the writer hides the cursor
once at open with ESC G 0 — exactly what the game and the ROM's own demo
do. Verified by running the launcher's actual byte streams through the
firmware-modeled vPLASMA emulator: the old ESC J path left stale pixels
(732 lit vs 404); the new path renders byte-identical to a freshly
cleared panel (404 lit, cursor hidden).
Version bumped 4.11.4.3 -> 4.11.4.4 across Launcher, Console, vPOD, the
install/build banners, and the diff-suite version assertion. This cuts a
clean release boundary that also carries the earlier field fixes
(process-tree kill on Stop, Win10 install.bat icacls quoting, real system
volume + pre-uninstall opt-in, and the volume menu check-mark fix).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two launcher behaviors vPOD only simulated are now available for real,
matching the pod exactly:
- "Actually set system volume": set_VolumeLevel drives this machine's
master volume through the launcher's own chain — nircmd.exe in the
games root, else Core Audio (Vista+), else winmm. The chain moved out
of TeslaLauncher.cs into Launcher/VolumeControl.cs and is compiled
into both apps as linked source (the MiniZip pattern); launcher
behavior is unchanged. Off by default: the value is stored/echoed
only, as before.
- pre-uninstall.bat now runs before the product directory is deleted on
UninstallApp (working dir, hidden window, 120 s wait, exit code
logged — mirrors CleanupProductDirectory). Gated behind the renamed
"Run package install/uninstall scripts" checkbox (was "Run
postinstall.bat after install"; RunPostInstall -> RunPackageScripts),
closing the asymmetry where install scripts had an opt-in but
uninstall scripts silently never ran.
Verified: 106/106 diff tests; live master-volume set/restore through
vPOD's build of VolumeControl; functional probe of UninstallApp against
an isolated games root with the flag off (script skipped, dir removed)
and on (script ran, then dir removed).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The Launcher's XP11 port (8730b9b) now extends to everything: one net40
flavor across Console, vPOD, Contract, and SecureConfig (Newtonsoft.Json
everywhere; the net48/System.Text.Json legs and their #if splits are gone
since nothing consumed them).
Console (net40, single TFM like the Launcher):
- The ~31 BinaryFormatter bitmap blobs in the .resx files became raw
embedded files under assets/icons/ (extracted byte-faithfully via a
serialization surrogate — the animated square_throbber.gif survives),
loaded by Properties.Resources.EmbeddedBitmap/EmbeddedIcon. Reason:
System.Resources.Extensions' DeserializingResourceReader is net461+
and cannot load on net40. Strings stay in the .resx.
- IReadOnlyList -> IList in AppRegistry (net45+ interface).
vPOD (net40, single TFM):
- Zip extraction now shares the Launcher's MiniZip.cs (linked source), so
the diff-test install round-trip exercises it against ZipArchive zips.
- RPC args as JTokens; LaunchApps.json persistence via Newtonsoft;
Thread.VolatileRead instead of Volatile.Read.
Contract/SecureConfig: net40-only; Client/** (PodManagerConnection) now
ships in the one build. The Launcher package gains
TeslaSecureConfiguration.dll as a dependency of the client half.
Tests: the net48 xunit host loads the net40 assemblies (both CLR4), so
the suite exercises exactly what ships — 106/106 green. Also verified
live: net40 console provisioned, managed, and ran a full RP mission
against net40 vPOD (beacon/passphrase/RSA, 53290 RPC, egg load,
Run/Stop Mission).
Version: 4.11.4.3 across Launcher, Console, and vPOD (vPOD joins the
suite version line; was 1.0.0). Ship the dotNetFx40 redistributable in
Launcher/assets for XP-era pods.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Adopt the launch-key convention (documented in the Apps.xml header):
fresh Guid = product id, first Launch key reuses it, each extra entry
increments the last hex digit (wrapping F->0). Never a -1 suffix:
keys parse as System.Guid and silently collapse to Guid.Empty.
- Rewrite BT411 LC/MR + RIOJoy keys to the convention. RP4.11 LC/MR stay
pinned to the original console's hardcoded Guids (SiteManagement
constants + diff tests).
- Remove vPOD from the shipped catalog: dev tool, never a console-deployed
product (README documents the ad-hoc Add Product path instead).
- Add TeslaRel410 (DOSBox-X preservation pods): six entries, BT/RP 4.10 x
GameClient/LC/MR, all C:\Games\TeslaPod410\pod-launch.exe with mode
bt/rp. LC/MR boot identically (role assigned via egg hostType) and no
{res} token (output size fixed per rig at postinstall).
- CatalogTests: now 5 products / 14 entries; full diff suite 106/106.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two additions to the virtual launcher's real-process mode:
- Auto-restart watchdog. Replaces the poll-on-query PruneExitedProcesses
with a per-process watcher thread (StartWatcher): when a real-launched
app exits on its own -- not via a Kill*/Uninstall, which untrack it
first -- it is dropped from the running list and, if its LaunchData has
AutoRestart and the "Auto-restart after the app exits (watchdog)"
toggle is on, relaunched after the Agent's 2 s delay. A watchdog
generation counter cancels pending restarts when the pod goes dark
(power off / reboot / reprovision / WipeApps); the console's KillAllApps
leaves them pending, matching the real Agent's race.
- postinstall.bat toggle. A "Run postinstall.bat after install" checkbox
(above "Actually launch apps", off by default) makes an install execute
a packaged postinstall.bat via cmd /c (waited up to 5 min) before
deleting it, like the real service. Off, it is logged and removed unrun
as before -- it runs package script code on the host.
Both are opt-in from the vPOD window. Verified against the real
LauncherRpcServer over a loopback socket: the watchdog test relaunches an
exited ping.exe with a new PID and stops once toggled off; a crafted
package's postinstall.bat runs (and is removed) only when enabled. Full
differential suite 103/103.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- InstallProduct now extracts into the real C:\Games (the launcher's
GAMES_DIR) so deployed products land where their catalog launch entries
point; uninstall removes the real product folder. Tests pass an isolated
games root through the VirtualLauncher ctor.
- New "Actually launch apps (real processes)" toggle (off by default =
simulated PIDs): LaunchApp starts the entry's exe exactly like the Agent
(same start info, same registered-but-not-installed error), Kill*/
Uninstall/Wipe terminate the real processes (kill before folder delete),
and self-exited apps are pruned from GetLaunchedApps/FullUpdate. Real
processes die with the machine: power off, reboot, or closing vPOD. The
Agent's autoRestart watchdog is deliberately not emulated.
- Pod Power and Mimicking Game groups swapped (game left, power right).
- Provisioning round-trip test now skips as inconclusive when a running
TeslaConsole holds UDP 53291 instead of failing the suite; two new tests
cover real launch/kill (ping.exe) and the missing-exe error path.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
vPOD has outgrown its home inside the console's folder: it now emulates both
halves of a pod (Munga game client + TeslaLauncher service / provisioning),
so it lives at the repo root beside Console/, Launcher/, Contract/ and
SecureConfig/, like the peer it has become.
Accompanying changes: project references rebased (Contract, SecureConfig,
the console's vendored Munga Net.dll), solution + DiffTests reference paths,
the console csproj's now-obsolete vPOD source exclusion removed, and the
root README / Apps.xml / vPOD README path mentions updated. pack.ps1 is
self-relative and now emits vPOD/dist/vPOD.zip.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CydandClaude Opus 5