Skip to content

Latest commit

 

History

History
98 lines (70 loc) · 12.4 KB

VALIDATION.md

File metadata and controls

98 lines (70 loc) · 12.4 KB

Validation record

Updated 2026-09-27 UTC. Results for the new portable rewrite and the previous installed edition are recorded separately.

Portable v1.0.0-rc.4

The portable download is camera-bridge-1.0.0-rc.4-win-x64.zip, containing only CameraBridge.exe and CameraBridge.cfg. The standalone executable is about 80 KB and needs no third-party binaries. It uses the system's .NET Framework; both cabinets have 4.8.1, registry Release 533509.

Version rc.3 captured correctly on cabinet A, but cabinet B could not connect its selected video format. Version rc.4 requests a supported video header and tries another format after a connection failure. The fix passed physical capture and three game-style EOF checks on each cabinet. Cabinet B's candidate started in 1.924 seconds and stopped cleanly. The user later confirmed GP2 local and remote portraits on both linked cabinets after the GP2 settings change recorded below. A fresh native GP1 test and repeated frontend start/exit tests remain pending.

Check Current result Scope
Native build and test-pattern startup Passed initial checks The app built and served the generated test image. This does not test the physical camera or a game.
Native server automated suite Passed: 93 assertions Exact JPEG bytes, original WSAPoll mask 0x300 with readable EOF, fragmented NUL request, stale-frame rejection, peer filtering, both listeners' Host checks, preview origin checks, and port-busy cleanup.
Native process suite Passed: 44 assertions Repeated and concurrent starts, mode/config mismatch handling, changed-config stop, invalid settings, missing-camera cleanup, and cancellation during startup.
Physical webcam on cabinet A Passed capture checks Baseline 320x240 JPEG, orientation/stride/mirror checks, three capture/release cycles, lower-rate throttling, and 50 startup-cancellation races. Current output was about 15 FPS, matching a separate current FFmpeg comparison on the same camera.
Hidden --start and --stop commands Passed with test pattern and physical webcam Startup waited for a fresh frame; both commands returned exit code 0. The live camera health check showed a 4 ms frame age. Actual frontend game sessions remain separate.
Capture/server code review Completed and fix tested Preview routes reject non-local Host names. Game requests remain compatible with the original hostless HTTP request.
Portable installation on cabinet A Passed standalone checks Installed EXE with local CFG; all eight existing frontend hooks migrated for four game entries, with XML backups and an idempotent verification. Exact start/stop commands succeeded with the physical webcam. Three LAN-bound probes passed JPEG dimensions, Content-Length, final EOI, urgent readiness and normal EOF, including a fragmented NUL request.
GP2 game camera tests on both cabinets Passed by user report Both local camera images worked with native rc.4. Remote portraits initially appeared black, then worked after both GP2 graphics settings below were disabled on both cabinets.
GP1 game camera test with the native build Pending The earlier edition's GP1 result does not establish a fresh native rc.4 result. Whether GP1 needs the same graphics settings is unconfirmed.
Cabinet B capture fallback candidate Passed standalone checks Candidate built from commit 6b725760f711b053b6c1d1607b1475026f88535a: live 320x240 capture, three game-style checks, and successful start/stop with no leftover processes or listeners. Official release-file installation and fresh game tests are recorded separately.
Official rc.4 release installed on both cabinets Passed standalone checks Both installed EXEs match SHA-256 1AAE296C6D0B8401D4041D8C915D1DEE612858D94F69FE2C74B9EC0E9ED91ECD. Cabinet A passed its final installed start/EOF/stop check. Cabinet B passed two physical start/stop cycles and all three game-style protocol checks. Both use their own configured webcam and LAN address; each has eight frontend hooks across four game entries and a desktop shortcut. Both bridges were initially left stopped, with no remaining processes or listeners, before the later user game test.
GP2 linked play with both portable bridges Camera portraits passed by user report The user confirmed both local and remote player photos after the combined GP2 settings change on both cabinets. This does not establish a GP1 result or repeated frontend lifecycle results.
LaunchBox repeated start/exit Pending Verify --start / --stop with real game sessions on each cabinet.

The user's linked-cabinet photo shows two visible camera images from the previous installed edition. Native rc.4's GP2 result above comes from the later user report, not that earlier photo.

GP2 remote portrait fix

With both games closed, the following GP2-only settings were added to User\GameSettings\GNLE82.ini on both cabinets after backup:

[Video_Hacks]
EFBToTextureEnable = False
DeferEFBCopies = False

The user confirmed that the other player's previously black portrait now showed its photo on both cabinets. The two settings were tested together. This does not prove that either setting alone is sufficient. Their need in GP1 is unconfirmed.

The saved GP2 game-settings file has SHA-256 F0E5CC79AB5EA5438A2C0553D56CD6620CCFF00597AC6E23B67AB96E5AC218F0. The existing guarded 20-second boot LAN-check patch remained in place. The emulator executable and cabinet IP table were unchanged for this trial. See the GP2 setup steps.

This change did not produce a new FPS measurement. The native capture rate recorded above remains about 15 FPS on cabinet A. Repeated real-game frontend start/exit tests remain pending.

Previous edition

The following records describe the earlier script-based cabinet installation and its renamed source candidate. Their tests and timing do not carry over automatically to the native rewrite.

Tested components

Component Observed version/setup
Emulator Original, unmodified installed executable.
Executable integrity Original installed executable used for successful second title testing; original digest verified.
Webcam One USB webcam per cabinet, with each local capture-device name configured separately.
Server runtime 24.17.0 on cabinet A; 24.21.0 on cabinet B.
Helper interpreter 3.14.6, standard library only.
Capture/encoding runtime 8.1.2, dshow input and mjpeg output.
Output 320 × 240 baseline JPEG; capture configured at 30 FPS.
Camera route Same-computer LAN-address listener with host EOF compatibility; cabinet seat mappings preserved.

Functional results

Check Result Evidence or remaining work
Second title boot camera check, original emulator Passed User confirmed successful camera check through the external helper.
Second title camera operation and improved smoothness Passed by user report Tested through the existing launcher after capture/helper optimization.
Second title linked cabinet session Passed by user report User reported the linked test worked with the camera active on cabinet A. This does not validate a webcam on cabinet B.
First title operation Passed by user report User reported the first title works well during the latest cabinet test. The bridge received camera requests without reported service errors. This records the installed cabinet version, not a new live test of the renamed public candidate.
Cabinet B standalone installation and live capture Passed Source archive verified; all 25 tests passed with zero skips. Valid 320x240 JPEGs, 30.12 output FPS over eight seconds, and the host EOF probe passed. Its camera destination was changed separately after backup; seat mappings, emulator executable, launcher profiles, and frontend files were preserved.
Both webcams active in a linked second title game Visible images confirmed by later photo The later user photo shows both local camera images. A separate completed paired-race result was not recorded here. This is evidence for the earlier installed edition only.
Frontend bridge startup and shutdown Observed on both cabinets After the user closed the previous game, cabinet A logged stop-request/shutdown at 00:02:00 UTC and cabinet B at 00:02:05 UTC on 2026-09-27. Their previous capture processes exited. New capture sessions began around 00:03:07 UTC for the next user test. Cabinet B's helper start command was confirmed exited while its new game/camera session continued. This verifies bridge lifecycle observations, not the pending in-game image and linked-race results.
Standalone setup helper Passed configuration checks Isolated checks covered dry run, unattended creation/defaults, invalid address and missing runtime rejection, existing-config protection, and verified backup/replacement. The live cabinet configuration remained unchanged. Setup does not change the emulator or frontend.
Local cabinet frontend hook configuration Passed on cabinets A and B Each cabinet has one Start and one Stop hook for each supported title entry under both configured platforms. The exact Start commands returned after fresh frames; the exact Stop commands released each bridge's server runtime, helper interpreter, and capture encoder processes. Cabinet A's camera listener also passed the host EOF check, and cabinet B's check confirmed both listeners closed. Installer reruns proposed no changes. Both bridges were initially left stopped for the user play-test; subsequent game-session shutdown and next-session startup were observed as recorded above. The local XML editing utility is excluded from the public package; public integration is documented in FRONTEND.md.
Original emulator and cabinet mappings Verified Running process used the original executable; only the camera destination changed among arcade IP mappings.

Timing observations

  • Webcam capture measured 29.87 FPS on cabinet A and 30.12 FPS on cabinet B with a 30 FPS setting, over eight-second samples. Small sampling differences around the configured rate are expected.
  • An isolated helper response benchmark improved from a 20.8 ms median to 0.46 ms after removing unnecessary delay. This measures that benchmark, not game rendering.
  • During live operation, observed response timing was approximately 34 ms, with the game requesting roughly 10 images per second. Capture rate, server response time, and game request/display rate are different measurements.
  • The user reported improved camera behavior and a working linked session after the changes. No in-game 30 FPS claim is made.

Protocol and automated evidence

The ordinary host close path was reproduced with native WSAPoll: data readiness was followed by POLLHUP alone, which the tested emulator did not expose as guest read readiness. The second title requires recv() == 0 before finalizing its image.

The synthetic urgent-byte probe exercised 24 cases across three payload sizes, two receive sizes, and two consumer delays. All 12 compatibility cases delivered the exact normal payload and EOF through the original read mask; all 12 ordinary-close controls reproduced the stall. Repeating the cases with an urgent byte value of zero gave the same result. An independent native probe against the actual helper interpreter helper also received the exact HTTP response, observed readable EOF, and verified graceful shutdown.

The renamed public candidate passed 25 automated tests with zero failures and zero skips. The test runner also reports a support module as a separate passing item. Its setup dry run passed without writing configuration. These checks are automated evidence only; the live game results above belong to the prior cabinet installation. Coverage includes HTTP/JPEG handling, capture, helper forwarding, NUL-terminated requests, source filtering, runtime resolution, and lifecycle behavior. Rerun node --test after release changes with the required runtimes available. Passing automated tests does not replace the pending game and cabinet checks above.

Reproduction tools:

  • Test-CameraEof.ps1: checks the running configured camera listener.
  • tools/Test-OobEof.ps1: standalone synthetic host EOF comparison.

No emulator executable patch is needed by the installed compatibility route. Runtime binaries, game files, local logs, and diagnostic backups are excluded from the intended source package.