Skip to content

Merge testing into main - #650

Closed
chestahh wants to merge 796 commits into
OpenCommunityEdition:mainfrom
chestahh:testing
Closed

chestahh wants to merge 796 commits into
OpenCommunityEdition:mainfrom
chestahh:testing

Conversation

@chestahh

Copy link
Copy Markdown

No description provided.

Promote the remaining fixture-only skill enum correction after all 122 bot harness checks pass. Production behavior is unchanged from the validated enhanced bot fix. Main CI reruns the full release workflow.
The main-menu UI map has a live player but its root widgets belong to NONE,
so the per-player widget check incorrectly treated it as gameplay. Keep
physical navigation there while retaining per-player pause ownership.

Give custom layouts a separate Cycle Camera action in every gameplay camera
context. Keep timing and release guards in the remapper and pass the result
through the director's existing selection and client-permission checks.
Save v3 layouts with 26 actions; accept older layouts without overwriting
existing debug bindings. Default layouts and network contracts are unchanged.

Add production-code regression coverage for menu ownership, camera entry and
exit, rebinding, clearing, migration, timing, per-port state and lifecycle.
Keep current keyboard/mouse camera fixes and update their harness stub.
Register the integration suite in CI and update the controls documentation.

Reviewed the combined implementation, interfaces, state lifetime and tests.
On the current testing base: 42 pytest cases passed, one POSIX-only socket
case skipped on Windows; all 12 startup/camera harness cases and negative
controls passed with 32-bit clang. Diff whitespace checks passed.
Before transfer: Windows portable release built; 22 Android/web units
cross-compiled; hidden Xbox/CE/MCC gameplay smoke tests and menu screenshot
passed. Full Android/web builds were blocked by Windows shell setup, so
complete multi-platform builds remain for CI. No physical-controller test.

Player-Note: Remapped controllers keep D-pad and Back working in the main menu.
Player-Note: Bind Cycle Camera to enter and leave flying-camera mode with custom controls.
Promote the isolated main-menu navigation fix, remappable camera-cycle
action and regression coverage from testing. The merge adds only the
controller change; rendering and stutter investigations remain uncommitted
and untouched in their original checkout.

The combined controller implementation was reviewed and locally validated:
42 pytest cases passed (one POSIX-only case skipped on Windows), plus 12
startup/camera harness cases and negative controls. Windows, Android and
MCC validation CI passed on testing; Linux/browser jobs were still running
at merge time and are being monitored along with main's full pipeline.
Revert OpenCE PR OpenCommunityEdition#440 (d79ea95,
mainline 1, including cb64385),
and its follow-ups 88edf51 and
62d4154.

Players report inconsistent device detection across phone families,
including disappearing controls on Samsung devices without peripherals.
Restore the pre-PR controller-only visibility policy, scene flags,
manual Android mouse capture and SDL relative-mouse patch. Preserve the
separately merged video/frame-pacing changes and GulchCE touch features.
No simulation, network protocol or map-family boundaries change.

Reviewed the combined seven-file rollback, callers, lifecycle and docs.
Validation: touch layout and gyroscope Java checks; 36 web page tests;
27 web/controller pytest cases passed, one POSIX socket case skipped;
C touch-menu checks and three gyro cases passed using Windows-compatible
compiler invocations without -lm. Whitespace checks passed.
Android/web Ninja builds attempted: blocked by Windows shell/toolchain
invocations. Full platform builds and tests will run in testing CI before
main promotion. Physical Samsung/Google device verification remains needed.

Player-Note: Restore Android touch controls by reverting unreliable automatic input-device detection.
Promote testing 6303ff6 after its Build and MCC map validation workflows
passed. Restore the input behavior before OpenCE PR OpenCommunityEdition#440 and its two
follow-ups, preserving unrelated Android and GulchCE changes.

The merge is conflict-free and its tree matches testing exactly.
Validation: testing Build 40993742606 passed Linux, Windows, Android APK,
web/browser checks and regression tests; MCC validation 40993742605 passed
reader/validator and Windows adapter checks. Local checks and limitations
are recorded in the revert commit. Physical-device confirmation is pending.
Fix issue #84: point emission was tick-batched but state aging ran every
render frame. Short bright states therefore expired sooner at high FPS.
Use the emission interval for state aging, retain per-frame motion and
texture animation, and reset the existing clock at map boundaries.

This is a shared effect fix, with no tag conversion, map-family dispatch,
projectile simulation or wire-format change. No protocol bump is needed.

Self-reviewed the complete implementation, private caller, lifecycle,
renderer state contract, regression coverage and reading documentation.
Checked: 203 tests passed, 5 skipped; Windows portable release build;
real Xbox b40 Shade captures before/after at 30 and 120 FPS; CE Death
Island Ghost capture at 120 FPS; Android arm64_32 compilation/assembly
and wasm32 compilation of the changed source.
Full Android/web builds require unavailable Unix shell utilities here.
MCC Digsite 1/10 tag loading succeeded, but gameplay smoke checks stopped
at an unsupported debug_camera_load script. Linux and real map-family
transitions remain unverified. See docs/contrail_timing.md for limits.

Player-Note: Plasma projectile trails keep their brightness and thickness at high frame rates.
Fix #86: cinematic_render used the rectangle midpoint for every widescreen
translation. Xbox b30 authors its insertion title right-aligned at column
597; the old translation put it at 695 on an 852-column canvas instead of
794. Translate using the left, right or center anchor matching the text
justification, while preserving box dimensions and all 640-column output.

This fixes existing shared rendering behavior. MCC's guarded clipping
expansion still runs after positioning; loaders, tag data, timing and
state ownership are unchanged. No network agreement or packets change,
so the network version remains unchanged.

Extend the real-code harness for three alignments, four canvas widths,
HUD-default bounds, resolution changes and MCC-to-legacy isolation. The
new cases fail before the fix; a midpoint negative control also fails.
Update the reader guide. Final combined review found no need for broader
refactoring or additional state, and no temporary instrumentation remains.

Validation: 380 cinematic-title, Unicode-layout and MCC HUD/layout tests
passed; full portable release Windows build passed. Hidden b30 captures
verify 16:9 placement and pixel-identical 4:3 title output; 21:9 also works
with high-resolution text disabled. A hidden process traversed Warthog
Rally, b30, CE Death Island and Warthog Rally again, with complete trophy
titles before/after reload and corrected b30 positioning. These are
bounded rendering checks, not full playthroughs.

Android guest and wasm32 compilation of cinematics.c passed. Full ninja
android/web attempts are blocked by Unix shell build rules on this
Windows host; full links and Linux/device/browser runs remain unverified.
Guide links/symbols and git diff --check passed. Temporary maps, caches and
builds removed; local evidence and verified archives of test saves retained.

Player-Note: Chapter titles now stay correctly positioned on widescreen displays.
An existing map cache can skip the loading draws that previously created
the back-buffer read FBO. Creating it inside Present binds both GL targets,
overwriting the window draw binding and causing a self-blit. Restore the
window binding after the lookup. Keep cache reuse and map-family paths intact.

Direct reset comparisons isolated this startup error: shader, smart-state,
texture-memory and GL-cache resets did not remove it; a loading draw did,
even with its hardware reset disabled. Investigated #80, but the sustained
Android geometry failure remains unreproduced and is not claimed fixed.

Validation: the production-code presentation harness failed 4 of 5 cases
before the fix and passes all 5 after. Final targeted suites: 112 passed,
2 platform-dependent skips. Native Windows release builds; the changed
renderer compiles for Android ILP32/AArch64 and WebAssembly. Full Android
and web builds hit Windows NDK/SDL invocation and cp tool failures.

Hidden native GLES gameplay with Android capacities passed fresh and reused
cache runs: no GL errors, shader mismatches or sampled stale textures, with
all disk-cache payloads matching independent decompression. This is Windows
GLES validation, not an Android device test. No network contract changes.

Player-Note: Fix presentation when a cached map skips the loading screen.
Retain upstream history through 899219c. Its touch revert matches the rollback already in testing; retain the Android lighting-availability section while accepting the upstream touch documentation. Include the Discord LINK action correction and its regression tests. Reviewed the merge and unchanged touch implementation; combined integration validation follows with the requested issue fixes.

Player-Note: Discord Account LINK now opens the selected master when confirmed.
Integrate the tested framebuffer restoration from the issue 80 investigation. Persistent Android geometry loss remains unresolved. No map-family or network contract changes. Reviewed the three-line renderer fix and regression coverage; combined integration validation follows.
Integrate issue 84 with its lifecycle regression coverage and rendering guide. Preserve tick-based emission and per-frame movement while matching aging to emission. No simulation or wire-format changes. Combined integration validation follows.
Integrate issue 86 alongside issue 84, the framebuffer fix from the issue 80
investigation, and OpenCE through 899219c. Preserve the authored title
alignment and MCC clipping guard. Resolve the upstream README conflict by
retaining the fork's Android lighting section; the final Android documentation
and touch implementation match the existing testing branch.

Self-reviewed the complete combined implementation, private callers, contrail
map lifecycle, framebuffer creation/presentation, title bounds, MCC activation,
LINK action, regression tests and guides. No further code changes or unrelated
refactoring were needed. No simulation, wire format or network-version change.

Checked the resulting tree: portable release Windows build (PGO/LTO off),
280 pytest cases passed and five skipped, 36 web page tests passed, local guide
links and git diff --check passed. The additional unchanged video-border test
hit missing Windows sanitizer libraries; its native 32-bit assertion probe
passed without sanitizers. Harness tests used an isolated Windows compiler
adapter, without altering their assertions or checked-in sources.

Android and Web builds were attempted with installed cross-toolchains but
stopped on Unix cp/sed/ln/echo shell rules unavailable on this Windows host.
Linux runtime is unavailable. Full supported-platform validation remains for
CI. Prior issue-branch Xbox/CE visual checks and issue 86 MCC/Xbox/CE transition
evidence apply to unchanged source; no new gameplay run is claimed here.
Issue 80's persistent Android geometry loss remains unresolved.
Investigate #83 from origin/testing fc9c57c.
That base already contains OpenCE 521c8c8's NaN-crossing correction,
which addresses a CPU portal visibility failure consistent with the video.
Preserve the existing production fix and add successive-intersection,
rejection, metadata and in-place clipping coverage plus a negative control.
Record the evidence and remaining reproduction limits in the reader's guide.

Self-reviewed the clipping implementation, visibility caller, tests and docs.
No production, map-family, simulation or network-contract changes; no
network-version increment or additional production refactoring is needed.

Checked: 31 focused tests pass, including mutations without the NaN guard;
Windows release build (--pgo=off --lto=off), with existing Expat warnings;
hidden RTX 5070 Ti/1280x720 Blood Gulch run and solo c10 -> CE deathisland ->
MCC DigsiteMP/1 -> Xbox bloodgulch transitions, both normal exits; sampled
captures; documentation links/symbols and git diff --check. c10 coverage is
its opening cinematic. The MCC map retains its unsupported debug_camera_load
script rejection. Exact reported camera path and AMD/Intel/Linux behavior
remain unverified. Android/web code is unchanged and was not rebuilt.

Removed task-owned builds, map copies, caches and temporary dependencies.
Retained about 47 MiB of logs, captures and verified archived test saves.
Leave integration into testing to the coordinating chat.
The first-person distance cutoff mistakes stock rocket melee and reload
motion for a skeleton replacement, holding the entire pose at tick rate.
Track the full weapon handle and animation graph instead, retaining node
count, consecutive-tick and non-finite-input checks. Invalidate history
when interpolation is disabled or a pose is unavailable.

Stock Xbox Sidewinder also exposes an authored 180-degree tube reset after
the first shot. Blending it invents another half-turn. Treat near-half-turn
rotations as per-node discontinuities; other nodes and ordinary rotation
still blend. This changes rendered poses only, not animation timing,
ammunition, simulation or network agreement; no protocol bump is needed.

The existing first-person caller must provide identity because matrices
alone cannot distinguish weapon replacement from rapid motion. Its small
port-marked addition preserves Xbox, CE and MCC loading paths and guards.
Update the rendering guide and add production-code regression coverage.
Self-review covered state ownership, lifecycle, bounds, integration,
documentation and tests; no unrelated refactoring was needed.

Validation on the combined final implementation:
- Windows portable release build, PGO/LTO off, passed (existing Expat
  LNK4217 warnings remain).
- 17 focused harness tests passed, including two negative controls that
  restore the distance cutoff and erroneous chamber-reset blending.
- Both changed C translation units compiled for Android arm64_32 and wasm.
- Full ninja android and ninja web were attempted but blocked by Windows
  build/bootstrap limitations: NDK compiler launcher CreateProcess failure
  and shell-dependent SDL bootstrap/missing SDL3-LICENSE.txt. No WSL/Linux
  environment was available; full Android/web/Linux builds are unverified.
- Hidden stock Xbox and CE rocket runs reproduced the motion cutoff and
  verified fractional output. Xbox Sidewinder at 240 FPS reproduced the
  chamber reset; corrected output held the reset across eight subframes.
- Final production binary transitioned Xbox Blood Gulch -> CE Death Island
  -> MCC Mercury Falling -> Xbox and exited 0. Xbox/CE gameplay and MCC's
  opening cinematic were observed; this is not full campaign or all-map
  validation. An earlier refused stock MCC load also recovered to Xbox.
- Rendering-guide local links and named symbols checked; git diff --check.

Temporary instrumentation removed. Disposable maps, screenshots, caches
and test build outputs cleaned; inactive test saves archived losslessly
with original paths and verified hashes. Keep integration into testing in
the separate coordinating chat.

Fixes #85
Player-Note: Rocket launcher melee, chambering and reload animations stay smooth at high frame rates.
Retain the existing upstream visibility correction and add the issue investigation, successive clipping tests and negative control. Reviewed the production contract and evidence limits; no runtime behavior or network change. Combined integration checks follow with issue 85.
Combine 5f23f41 with issue 83's 14f7b6f regression coverage on testing
fc9c57c. Both branches merge without conflicts; production files exactly
match the reviewed issue 85 commit. Keep both rendering-guide additions.

Self-reviewed the complete combined changes, caller identity, per-player
history, reset/disabled paths, bounds, half-turn guard, clipping regression
coverage and documentation. No additional implementation changes were needed.
Existing Xbox/CE/MCC loading paths remain intact. This changes rendered poses,
not simulation or network agreement; no protocol increment is required.

Validation: portable Windows release build (PGO/LTO off); 90 regression tests
passed and one platform-dependent test skipped, covering both issue fixes,
visibility readback, geometry bounds, shadows, prior title/contrail/presentation
fixes and web build contracts. Guide links/symbols and git diff --check pass.
The harness uses an isolated Windows compiler adapter without assertion changes.

Android and Web builds attempted with installed cross-toolchains; both stopped
on unavailable Unix shell commands on this Windows host. Android's initial
sandbox download failure was resolved by the permitted network retry. Full
platform validation remains for CI; no new gameplay run is claimed. The issue
branches' bounded Xbox/CE/MCC observations apply to unchanged production code.
Issue 83's exact reported camera path and AMD/Intel/Linux reproduction remain
unverified. Retain the original commits and their evidence limitations.
Ruby Rebalanced's flamethrower uses separate world and first-person particle
attachments. The native emitter only queried world markers and ignored the
particle types' view exclusions, leaving the visible flame below the barrel.

Resolve first-person-only markers and per-view visibility in mcc_particles.
Keep the engine integration to two port-marked hooks because these private
emission/render operations have no registration interface. Preserve native
Xbox and Custom Edition results while MCC is inactive. No persistent state,
projectile changes, damage changes or network-version bump: particle emission
uses local visual randomness and does not change deterministic simulation.
Document the evidence, isolation boundary, lifecycle and validation limits.

Reviewed the complete implementation, integration, documentation and tests.
Validation:
- Release portable Windows build passed (existing Expat linker warnings).
- Focused MCC/rendering/lifecycle/web graph tests: 149 passed, 2 skipped
  (optional map fixture and POSIX-only socket harness).
- Regression controls reject missing FP markers and ignored view exclusions.
- Ruby C20 before/after hidden OpenGL captures confirm the barrel stream;
  private starting-weapon fixture and firing instrumentation are not shipped.
- Final production binary: MCC Ruby C20 -> Xbox a10 -> CE deathisland -> MCC
  Ruby C20, responding after each load and exiting normally.
- Both changed C units compiled with Android guest and web target flags.
  Full Android/web builds attempted but blocked by generated Unix commands
  and the Android NDK wrapper on Windows; WSL unavailable. Full platform
  integration, multiplayer and device visual parity remain unverified.
- Added documentation links/symbols and git diff --check passed.

Temporary fixtures, caches and builds removed. Useful logs/captures and a
verified lossless archive of inactive test saves remain locally (6 MiB).

Fixes #33
Player-Note: MCC flamethrower fire now starts at the barrel in first person.
Add MCC-only abs_integer, salted executing-thread handles and kill_thread
with parameter reference cleanup. Recover Ruined Halo's incomplete named
animation hierarchy only from agreeing associated models. Adapt its
short jet physics blocks with an explicit zero third constant, preserving
complete blocks and checking allocation ownership.

Implement Mission Z9's rasterizer_model_lighting_ambient writes instead
of refusing or discarding them. Capture and restore the prior renderer
value across load/unload and script restart, checkpoints and host/client
snapshots. Campaign state v6 is 772 bytes; checkpoint footer v3 stores
the extension in reserved space so saved thread pointers never move.
MCC protocol 63 rejects older MCC peers for the larger snapshot. Keep
HALO_PORT_NETWORK_VERSION at the pinned OpenCE 41 contract for Xbox/CE.
All compatibility logic remains MCC-owned; no shared source changes.

Based on origin/testing 75130bc.
Final design/code/docs/test review completed. Windows debug portable
build (PGO/LTO off) passed; 535 regression tests passed, five skipped.
Reproduced all three original refusals. Ruined Halo passed 90 seconds
of rendered simulation beyond the later particle assertion. Z9 passed
live save/revert, restart and Xbox/CE/MCC transitions with clean lighting.
Two Halo RPG peers exited normally after 100 seconds; read-only process
inspection verified both 0.1 and 0 lighting values on the client. The
full log comparator flags 21 seconds of cinematic aim disagreement,
also seen before the lighting change; the 45 seconds after the intro
pass with zero drift. Existing Ruined Halo shader warnings remain.

Android/web targets and WSL are unavailable on this Windows host.
Browser transport tests are blocked by their POSIX Windows-path handling
and missing web-site/broker/Xvfb prerequisites. Full campaign and visual
parity remain unverified. Details and evidence: docs/mcc_maps.md.
Temporary caches, fixtures and duplicate binaries removed; inactive
saves archived losslessly and verified. No integration into testing.

Player-Note: Ruined Halo, Mission Z9 and Halo RPG now pass their reported script and tag load failures.
Integrate the MCC-owned script, tag, checkpoint and synchronized lighting changes. Preserve OpenCE 41 for Xbox/CE and MCC protocol 63 for the extended campaign snapshot. Reviewed isolation and lifecycle; combined integration checks follow with the first-person particle fix.
Integrate 7bfbdc5 with the issue 89 MCC script, tag-validation and model-
lighting changes already merged from 6d3da9d. Preserve both original
commits, authors and player notes. Resolve the sole docs/mcc_maps.md
append conflict by retaining both complete evidence sections.

Combined review found no code conflict or further simplification needed.
The particle adapter owns no persistent state and preserves the native
Xbox/CE path when MCC is inactive. Issue 89 keeps its campaign/checkpoint
layout handling and MCC protocol 63; OpenCE protocol 41 is unchanged.
All 28 non-conflicting files match their reviewed source branches exactly.
Both MCC documentation additions and 150 local link targets were checked.

Validation on the combined tree: Windows portable debug build passed;
540 regression tests passed, 5 skipped, covering MCC scripts, threads,
parameters, validation, checkpoints, saves, campaign/lifecycle, protocol,
networking, particles, cache formats and web graph. The Windows harness
adapter changes compiler invocation only, keeping test assertions intact.
The source branches' runtime/map-family/co-op evidence remains applicable
to unchanged code; no additional live game run was performed here.

Android and web full builds were attempted but blocked by generated Unix
shell commands unavailable on this Windows host. The browser/native WebRTC
test stopped at its Windows URL-path handling before compilation; internet
map testing lacks a built web site, mosquitto and Xvfb. These checks require
a supported environment and remain validation limits, not passing results.
Issue 89 raised MCC_PROTOCOL_VERSION for the model-lighting campaign
snapshot but left the generated wire manifest at version 62. Regenerate
it from the current sources. Only the manifest version and its recorded
MCC_PROTOCOL_VERSION change; the pinned OpenCE 41 contract is unchanged.

Reproduced the failing CI assertion locally, reviewed the two-line change,
and then passed 76 checks covering the wire manifest, pinned OpenCE/MCC
protocol, network, campaign snapshots and checkpoints. No runtime code or
network behavior changes in this follow-up.
Promote testing b8d365f, retaining the reviewed issue 80/83/84/85/86/89/33
work and OpenCE changes. The merge is conflict-free and its complete tree
matches testing. No further implementation changes or refactoring needed.

Testing 128f8bc passed all four platform builds, the web browser checks,
and both MCC validation jobs. Its Linux test failure was the stale MCC
wire manifest; b8d365f records version 63 and passes the reproduced check
plus 75 protocol/network/campaign/checkpoint checks. OpenCE remains 41.
Prior combined local Windows build and 540-pass regression evidence remains
applicable. Main's full debug/release and publication workflows will be
monitored through completion; platform CI results are not full map parity.
Integrate testing b8d365f with OpenCE 41 and MCC 63 while preserving
the camera modules and host authority. Reserve camera join kind 101 after
upstream voting and MCC admission; expand the bounded join for link passes.
Move camera policy to bits 3-4 to preserve the dedicated co-op flag.
Temporary isolated camera protocols become 32710/32711 because the merged
wire and simulation contract differs from prior camera builds. Crossplay
remains disabled until the following session compatibility change.

Resolve input/menu generation additively; retain incoming controller startup
and UI guards, co-op wipe settings and camera checkpoint recovery.
Checked camera, controller, wire and checkpoint regressions; 16 focused
policy/admission/baseline checks pass after conflict resolution. Windows
release compilation reached linking; full integration checks follow.

Player-Note: Update the third-person branch with current testing fixes and features.
Default host setup to OpenCE Crossplay and lock the optional camera mode
when opening the lobby. Bind client permission to the selected host, clear
it at teardown, preserve saved preferences and reject camera policy bits
in ordinary sessions. Keep the upstream 41 relay unchanged in crossplay;
only camera sessions send the shoulder byte or use camera checkpoint fixes.

Keep MCC admission independently at version 63. Camera agreement versions
32712/32713 distinguish the new selectable contract from the merge baseline.
The camera envelope uses kind 101 and a bounded 520-byte buffer around
upstream joins or MCC kind 100. Host token, ban, slot and damage authority
remain unchanged. Lifecycle and codec hooks are necessary where the engine
owns connection creation, admission and relay serialization. No stencil
shadows or new camera controller bindings are included.

Reviewed combined module/state ownership, default paths, input allocation,
settings generation, documentation and upstream IDs. Focused production-code
regressions pass (58); broader web/comparator run passed 91 with 6 Windows
skips. The broad Linux harness needs Linux libm and is deferred to CI; its
Windows failure is retained. Native crossplay and final platform checks
follow this commit and will be recorded with the delivery.

Player-Note: Choose OpenCE Crossplay or GulchCE Third Person when creating a session.
Player-Note: Third person is automatically disabled when joining an OpenCE-compatible session.
Reset the feature-owned client capability when a connection fails or is
reset. Preserve the selected host capability only for the existing
same-host Discord pass retry, whose temporary advertisement has no metadata.
Keep host state independent so creating/resetting its loopback client does
not change the session contract.

Reviewed state ownership and cleanup. Windows portable release build passed;
16 camera admission, policy and MCC protocol tests passed, including real
client reset execution with and without socket teardown and a local host.
No wire or simulation contract changes.
The merged director and settings functions reference the existing optional
camera entry point and policy. Add those external service declarations to
the imported fixtures; keep their production code, assertions and negative
controls unchanged. The combined controller context suite passes locally
(4 tests). Linux CI executes the full harness because its runner links libm.
The first full run had 2134 passing tests; its 13 failures were these missing
fixture declarations, not simulation or assertion failures.
Use the existing spinner mouse handler, matching the other host controls.
Keep full-height host rows and verify both labels and mouse registration in
the generator regression. Match the surrounding integration indentation and
clarify that checkpoint recovery extensions run only in camera sessions.

Reviewed the combined session state, policy, admission, relay and menu paths.
Seven host-policy/menu tests pass; the broader local suite passed 95 tests
with 6 environment skips before this menu-only correction. Windows release
rebuild and rendered menu inspection are in progress.

Player-Note: [MULTIPLAYER] Choose OpenCE Crossplay or GulchCE Third Person before opening a lobby.
Keep the existing host rows, full-size selection artwork and all six game
rule links in their original positions. Replace the camera row with a
Session Options submenu containing the compatibility and camera policy.
Reuse the existing host menu dispatch without reinitializing its gametype
or listing when entering the submenu. Save only accepted host policy edits.

The generated spinner renderer ignores text_y. Match its bounds plus child
offset to the label baseline instead; both new controls use the standard
large menu font, 512 by 28 row and symmetric spinner arrows. Rendered the
restored host screen and crossplay submenu in hidden Windows instances.
All help text is visible. No shared renderer changes were necessary.

Combined review removed the abandoned Game Options page and reflow logic.
95 local tests passed, 6 skipped. Fix the Linux test fixture's linkage to
match its header and assert cleanup state/counters instead of leaving them
write-only under stricter warnings. Windows release build passed.

Player-Note: [MULTIPLAYER] Host compatibility and camera policy now have their own Session Options submenu.
At the user's choice, keep the remaining host controls at their original
size and spacing, with the invite link on the session submenu. Twelve rows
now leave a dedicated 40-unit hint area above the footer. Align it with the
row labels and retain the original header and full selection outlines.
The submenu reuses the existing invite-copy handler and shows its live link.

Reviewed the resulting menu flow and removed the abandoned reflow/page
approaches. Host-menu geometry tests pass; Windows release build and hidden
rendered host/submenu checks pass. Final full regression and browser/native
runs follow on this source. No wire or simulation changes.

Player-Note: [INTERFACE] Host help has room for two lines; Invite Link is now in Session Options.
The engine opens admission before setup. Lock the feature-owned contract
at lobby entry, remote admission, or match start instead; preserve the
host loopback join and all existing admission checks. Display a copy
prompt for the long invite link rather than clipping its value.

Reviewed lifecycle hooks, local admission, host authority, and MCC
boundaries. Focused regression: 131 passed, 6 platform skips. Windows
release compiled; actual browser host flow reproduced the premature lock.
All-platform CI and final native/browser checks follow this commit.

Player-Note: Session Options stays editable during setup and the invite link fits its row.
Record the Sentinel animation aim limit and shooter collision during projectile spawn safety checks. Keep independent experimental controls, map data evidence, exclusions and implementation boundaries explicit; no gameplay changes are retained.

Checked all 75 installed Digsite maps, hidden Windows captures at 2560x1440, source routes and document links. Archived and verified local evidence and test saves; removed temporary probes, map caches and diagnostic builds. No cross-platform or network compatibility claim.
Construct eligible player rays from the simulation camera and the authored
trigger offset, with cover checks and convergence separate from aim assist.
Keep the original-ray flag and existing Xbox/CE and excluded-unit paths.
Share the nominal ray with first- and third-person previews. MCC agreement
advances to 65/32714 because projectile simulation changes; Xbox/CE remains
41/32712.

Remove the unproven trail-to-marker experiment. The firing effect has its
own authored particle offset; a bare marker is not the flash centre. Do not
change trail rendering, fading, timing or instability during rapid aim.

Checked: 74 focused tests passed, one Windows-incompatible skip; Windows
release and full web builds passed; six affected Android guest units compiled.
Earlier unchanged simulation checks passed Xbox and MCC net_compare tolerances
with both host assignments (not perfect MCC determinism). Fixed-pose Sentinel
and Rocket captures cover level/steep aim; an unobstructed Sentinel capture
shows reticle convergence and flash overlap. Exact native visual parity is
not established. Full Android/Linux and browser-network checks remain limited
by the Windows environment, as recorded in docs/digsite_projectile_offsets.md.

Player-Note: Correct MCC weapon projectile offsets and aiming without changing beam trail behavior.
Record the user's acceptance of the Prisoner capture with the experimental trail adjustment disabled. Retain the camera-relative firing implementation and remove all follow-up visual experiments and runtime probes.

Reviewed the combined change. Rechecked 74 focused tests (one POSIX skip), clean Windows and full web builds, and six wasm objects. Prior Android object and network results apply to identical retained engine sources. Verified documentation links and preserved diagnostic evidence while cleaning disposable test outputs.
Record the Sentinel animation aim limit and shooter collision during projectile spawn safety checks. Keep independent experimental controls, map data evidence, exclusions and implementation boundaries explicit; no gameplay changes are retained.

Checked all 75 installed Digsite maps, hidden Windows captures at 2560x1440, source routes and document links. Archived and verified local evidence and test saves; removed temporary probes, map caches and diagnostic builds. No cross-platform or network compatibility claim.
Construct eligible player rays from the simulation camera and the authored
trigger offset, with cover checks and convergence separate from aim assist.
Keep the original-ray flag and existing Xbox/CE and excluded-unit paths.
Share the nominal ray with first- and third-person previews. MCC agreement
advances to 65/32714 because projectile simulation changes; Xbox/CE remains
41/32712.

Remove the unproven trail-to-marker experiment. The firing effect has its
own authored particle offset; a bare marker is not the flash centre. Do not
change trail rendering, fading, timing or instability during rapid aim.

Checked: 74 focused tests passed, one Windows-incompatible skip; Windows
release and full web builds passed; six affected Android guest units compiled.
Earlier unchanged simulation checks passed Xbox and MCC net_compare tolerances
with both host assignments (not perfect MCC determinism). Fixed-pose Sentinel
and Rocket captures cover level/steep aim; an unobstructed Sentinel capture
shows reticle convergence and flash overlap. Exact native visual parity is
not established. Full Android/Linux and browser-network checks remain limited
by the Windows environment, as recorded in docs/digsite_projectile_offsets.md.

Player-Note: Correct MCC weapon projectile offsets and aiming without changing beam trail behavior.
Record the user's acceptance of the Prisoner capture with the experimental trail adjustment disabled. Retain the camera-relative firing implementation and remove all follow-up visual experiments and runtime probes.

Reviewed the combined change. Rechecked 74 focused tests (one POSIX skip), clean Windows and full web builds, and six wasm objects. Prior Android object and network results apply to identical retained engine sources. Verified documentation links and preserved diagnostic evidence while cleaning disposable test outputs.
Register the real-code projectile tests in native CI and link libm for the standalone Linux test executable. No gameplay changes.
Register the real-code projectile tests in native CI and link libm for the standalone Linux test executable. No gameplay changes.
Update the wire-format snapshot for MCC 65 and its shoulder agreement; keep OpenCE 41 unchanged. The network format check now passes.
Update the recorded format and documentation for MCC protocol 73 and shoulder protocol 32742. Shared Xbox and Custom Edition agreement remains unchanged. Verified tools/network_format.py.
The fixed 64-entry item buffer rejected an entire valid 75-entry Halo 1
collection before discovery could list any maps. Match the existing 1024-map
catalog bound for all multiplayer packages, retain JSON/path validation,
and report oversized lists. No content identity exceptions are introduced.

Keep campaign progression at its independent 64-mission bound. Move enlarged
metadata scratch buffers off the stack and free them on every exit. Xbox/CE
paths and network formats/simulation remain unchanged; no version bump.

Reviewed the combined implementation, callers, allocation/failure lifecycle,
platform boundaries, documentation and regressions before final checks.
Validated Windows portable release (PGO/LTO off); 319 focused tests passed,
with 10 Linux/POSIX/JDK-specific skips. Seven new regressions failed before
the fix while the 64-entry control passed. Real Digsite 1.10 metadata now
returns all 75 authored titles; its local maps folder is empty, so collection
gameplay is not claimed. Its indexed-color PNGs remain outside the existing
RGB/RGBA preview decoder's support.

Hidden two-instance comparisons passed Xbox Blood Gulch (47 seconds), CE
Death Island (51 seconds), and MCC Tiny Halo (45 seconds), with clean exits
and no comparator divergences. Temporary maps/caches were removed; logs and
saves retained. Cross-platform build and sanitizer validation runs in CI.

Player-Note: Workshop collections with more than 64 maps now appear correctly.
(cherry picked from commit db1697c)
Negative periods and phases are accepted by the existing shader callers.
Using fmod's negative remainder as an interpolation weight extrapolated
past table samples, and the slide branch could return a negative alpha.
Deathtrap's wet materials exposed this with a -0.4 animation period.
Sample the preceding interval for negative fractional positions, leaving
nonnegative time and integer sample positions unchanged for every map family.

Keep the existing protocol for the coordinated testing-build rollout. No map-family
adapter, map-name exception, or alpha-assert suppression is introduced.

Validation: 22 focused evaluator/object-function checks, including negative
controls and bitwise positive-time comparison; Windows debug and web builds;
Deathtrap rendered gameplay with no assertions; matching Windows debug
clients on Blood Gulch compared for 63 seconds with no mismatched seconds.
Android's local build is blocked by Windows-host build-script assumptions;
full Linux/Android/Windows/web CI is checked separately.

Player-Note: Fix crashes from materials animated with negative periods.
(cherry picked from commit 44925e0)

GulchCE integration retains protocol 41. The engine source is identical to
the validated OpenCE base before this patch; map-family loaders are unchanged.
Adapt the HRTF, reverb, resampling and level models from OpenH1/OpenH1
6750db5 under AGPL-3.0-or-later.
Document the hardware research, retail-image reconstruction and recorded
measurements, and distinguish these approximations from bit-exact MCPX
emulation. No SDK headers, original filter bytes or DSP image are included.

Enable HRTF on every platform and supported output layout by default;
retain authored 2D routing and the positioned-stereo extension. Keep the
GulchCE packet queue, callback ownership, cut fades and loading buffers.
Use the measured direct/send resamplers, 24 kHz room network, level floor,
Doppler law, 64-frame parameter ramps and surround/LFE behavior. Preserve
saved resampler preferences; new configurations select the Xbox kernel.

Keep processing in audio-owned includes, publish settings under the mixer
lock, and leave Xbox/CE/MCC loading and network simulation unchanged. Review
also corrected a negative wrapped reverb index and guarded non-finite HRTF
table angles. Preserve upstream attribution and measured regression data.

Checked: 297 audio, lifecycle, listener, decoder and menu checks; local
Windows, Android and web builds (temporary Windows shell adapters for the
latter two); hidden Xbox -> CE -> MCC -> Xbox transitions with 5.1 dummy
output; modern and CE settings at 1920x1080; documentation links and diff
checks. Hardware listening and Android/web device playback remain untested.
Linux and the broader CI suites remain pending remote push authorization.

Player-Note: [AUDIO] Restore Xbox-style 3D sound and environmental effects, with HRTF on by default.
Player-Note: [AUDIO] Add an Xbox resampler mode; new configurations use it by default.
Linux Clang rejects the repeated LPDIRECTSOUNDSTREAM typedef in the C99 test fixture under warnings-as-errors. Keep its original declaration and remove the duplicate; game sources and the Windows playtest executable are unchanged.

Reviewed the one-line fixture correction. All 140 mixer, Xbox mixer, HRTF and reverb tests passed locally, including negative controls. The prior Linux CI log traced all 75 failures to this duplicate declaration; full platform CI will rerun after push.
Web browser setup failed and MCC validation timed out while the hosted runner repeatedly failed to resolve azure.archive.ubuntu.com. Remove only that entry from the runner's existing mirror list before installing dependencies in the affected jobs, retaining its official Ubuntu archive and security fallbacks and package signature verification. See actions/runner-images#14594 for the matching hosted-runner issue.

Reviewed both workflow changes. Both shell snippets parse; a fixture using the runner mirror list confirms only the Azure entry is removed and repeated application preserves both fallback entries. Web build checks: 11 passed, 1 existing toolchain-dependent skip. Remote CI will validate actual package downloads and all platform jobs.
Skipping the Azure archive did not resolve dependency setup: the fresh runner also failed to resolve archive.ubuntu.com and timed out before MCC tests. Remove the ineffective workaround and retain the runner's standard fallback configuration. Preserve the logs identifying the external DNS failure; no game code or test assertions change.

Reviewed the inverse workflow diff and verified it restores both workflows exactly to 78f599e, whose Linux, Windows, Android and Windows MCC checks passed. Web and Linux MCC dependency installation remain blocked by runner DNS failures.
Integrate the validated OpenH1-derived Xbox DSP, positional audio and default HRTF processing, retaining the feature history and existing Xbox, Custom Edition and MCC loading boundaries.

Reviewed the conflict-free integration: the resulting tree exactly matches xbox-audio-parity at 8d41b23. Its Linux, Windows, Android, web browser and both MCC validation jobs passed; 140 focused audio checks and Windows map-family transitions also passed. Testing CI will validate this merge commit before main promotion.
Combine refactor lifecycle, sharing, checkpoint and browser work with testing audio, controller and MCC fixes. Preserve both sets of regression coverage. This is the integration baseline for the authoritative OpenCE merge; final platform and runtime validation follows on the combined result.
Keep upstream MCC, checkpoint, vehicle-riding and engine changes authoritative.
Remove the obsolete MCC capability envelope and use shared protocol 43 for
Xbox, Custom Edition and MCC maps. Preserve content identity and load checks.

Keep GulchCE-only third-person sessions on camera protocol 32744. This version
accounts for the upstream agreement and moves camera policy to bits 4-5 so
upstream vehicle riding remains independent.

Validation in progress: Windows release and web builds pass. Blood Gulch and
Custom Edition Death Island crossplay against upstream 14ff64c pass with
either host and no measured simulation drift. MCC co-op is running. Focused
admission and camera checks pass; broad regression fixture updates and full
platform CI are pending before advancing testing.

Player-Note: OpenCE crossplay now supports MCC; GulchCE-only sessions retain third-person play.
Use the authoritative upstream cloth clock in GulchCE's checkpoint/unload
cleanup hook. The automatic merge had retained a second, unused clock, so
cleanup reset no longer affected the simulation. Keep upstream's tick cap.

Update production-code fixtures for upstream menu dispatch, shader allocation,
vehicle deck state, bounded worker waits and optional scripted-pose validation.
Retain malformed-packet, lifecycle, map-family isolation and negative controls.

Checked focused camera/admission and MCC adapters, 128 UI/scripted-state cases,
shader-cache, riding-camera and server-teardown cases, and 37 worker/render
lifecycle checks. Windows-only limitations of POSIX harness cases are covered
by Linux CI. Full platform CI and the final combined suite remain in progress.

Player-Note: Checkpoint restores and map changes correctly reset cloth timing after the upstream merge.
Regenerate Modern menus after the upstream co-op and audio additions using
Pillow 12.3.0, matching CI. Recognize upstream's initialize-only campaign list
so Modern geometry and scrolling remain attached. Keep Crossplay and GulchCE
Only captions consistent across session menus. Remove a duplicate stereo
Doppler config entry and let the touch harness use Windows CRT math.

Validation: 2625 feature regressions passed, 12 platform skips; 34 focused
frontend and touch checks passed. Windows and web builds passed before the
menu regeneration; final rebuild and platform CI follow. Xbox, CE and MCC
crossplay passed against upstream 14ff64c in both hosting directions;
GulchCE-only choice/on/off, late join and co-op camera policy checks passed.

Player-Note: [MENUS] Keep campaign scrolling and session compatibility choices working in Modern menus.
The merged CE XML lost OpenCE's solo map list update callback. Restore the
complete authoritative upstream file, regenerate Modern menus, and remove
the unnecessary initialize-only list workaround. Add a regression requiring
both native list updates and Modern scrolling. Correct reader-guide links
from the removed MCC guard to the camera-only admission module.

Review: all 146 upstream MCC-owned modules match upstream 14ff64c exactly;
no duplicate config entries remain. Preserve map-family lifecycle and
GulchCE camera isolation, with shared OpenCE 43 crossplay for all map families.
Validation: 46 focused frontend tests pass, final Windows/web builds pass.
Classic and Modern Session Options were inspected at 2560x1440; campaign
capture exposed this missing callback. Guide links and diff whitespace checked.
Earlier final feature run: 2625 passed, 12 platform skips. Live upstream
crossplay and GulchCE-only camera session results are retained in local logs.

Player-Note: [MENUS] Restore campaign mission rows and previews in Classic and Modern menus.
Remove redundant pre-refactor hooks left beside OpenCE's equivalent hooks
in cache read/unload, script validation/compilation/assignment, grenade HUD
and grenade gameplay, and session disposal/restart. Retain upstream ordering,
validation and port comments. GulchCE's bot command hook remains intact.
This removes duplicate invocation and unreachable branches without changing
shared protocol 43 or camera protocol 32744.

Review compared the affected implementations to upstream 14ff64c and
scanned the remaining feature/MCC hooks for nearby duplicate calls.
Validation: 307 focused lifecycle, grenade, script, parameter, syntax,
UI-network and server teardown tests passed. Final Windows/web rebuild and
MCC crossplay recheck follow before testing promotion.
Keep upstream's vehicle-riding placement, generated-asset and six mutation checks, while accounting for GulchCE's existing Session Options help entries and 28-unit host rows. The original compact upstream layout is not the merged product layout.

Validation: 34 vehicle-riding and network-format checks passed, including all six deliberate layout faults. No production code or assets changed.
Find the Modern controller editor by its stable gamepad key instead of its former list position. Match the upstream five-argument controller_glyph_draw API with a typed stub, preserving the fallback and disposal assertions. No production code or generated assets change.

Validation: all 52 tests in the complete eight-file controller, keyboard glyph, menu glyph and HUD CI group passed locally on Windows with x86 clang. Reviewed the diff and checked whitespace. CI monitoring intentionally omitted at the user's request.
Merge OpenCE updates and MCC refactor into testing
@chestahh

Copy link
Copy Markdown
Author

me when i open my pr against the wrong repo

@chestahh chestahh closed this Nov 12, 2026
Adam-AtlasSoftware pushed a commit to Adam-AtlasSoftware/MendicantCE that referenced this pull request Nov 12, 2026
…-band

Fog the water where the world ends under it, not the sky behind it
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants