Single-player levels abort back to the main menu: CreateOpenALBuffer fails on Apple's OpenAL (macOS arm64, 0.12.0)

[DrHowarth]

Summary

On the experimental macOS arm64 package, single-player levels load completely —
AAS files load, Spawning entities runs — and are then torn down at the end of the
load by an OpenAL buffer allocation failure. The engine catches the error, runs
Game Map Shutdown, and returns to the main menu.

To a player this looks like "the game won't start" or "New Game just bounces me back
to the menu". It is neither a path problem nor a Gatekeeper problem: the map genuinely
loads first.

The macOS packages link Apple's deprecated OpenAL.framework and bundle no OpenAL Soft,
and Apple's implementation refuses to allocate the number of hardware buffers a stock
Quake 4 map needs.

Environment

Field	Value
openQ4 version	0.12.0
openq4_commit	eab43b9947e0777c19fbba189271e79fe5d423fe
openq4_game_commit	e479f914bedd9a324ac06f79ac199c456031746c
Package artifact	openq4-0.12.0-macos-arm64-opengl-unsigned.tar.gz (OpenGL package, not Metal bridge)
Launch path	/Applications/openQ4.app/Contents/MacOS/openQ4 from Terminal, openQ4.app dragged to /Applications from the unpacked archive
macOS	26.6 (build 25G72)
Hardware	MacBook Air Mac14,15, Apple M2, 8 GB RAM
Rosetta	Not involved — binary is arm64, lipo -archs reports arm64
Retail assets	GOG Quake 4 1.4.3 offline installer, extracted with innoextract; all 25 pak*.pk4, q4cmp_pak001.pk4 and the unsuffixed zpak_english.pk4 present and loading
Renderer	OpenGL (default). Vulkan not enabled. Renderer appears unrelated.

Symptom

loading maps/game/airdefense1.aas128
Entering doom_main()
Exiting doom_main()
Spawning entities
OpenAL Error: Invalid Value (0xa003), @ ../src/sound/OpenAL/AL_SoundSample.cpp 565
------------ Game Map Shutdown --------------
---------------------------------------------
********************
ERROR: idSoundSample_OpenAL::CreateOpenALBuffer: error generating OpenAL hardware buffer
********************
------------ Game Map Shutdown --------------
--------------- Game Shutdown ---------------

0xa003 is AL_INVALID_VALUE returned from the buffer generation call at
src/sound/OpenAL/AL_SoundSample.cpp:565.

OpenAL environment

OpenAL ALC version: 1.1
OpenAL vendor: Apple Computer Inc.
OpenAL renderer: Software
OpenAL version: 1.1
OpenAL extensions: AL_EXT_OFFSET AL_EXT_LINEAR_DISTANCE AL_EXT_EXPONENT_DISTANCE
                   AL_EXT_float32 AL_EXT_STATIC_BUFFER AL_EXT_SOURCE_NOTIFICATIONS
                   AL_EXT_SOURCE_SPATIALIZATION
OpenAL HRTF: unavailable
OpenAL output mode: unavailable
OpenAL EFX is not available in this build; reverb wet send disabled.

The package links Apple's system framework and ships no OpenAL Soft:

$ otool -L openQ4.app/Contents/MacOS/openQ4 | grep -i openal
	/System/Library/Frameworks/OpenAL.framework/Versions/A/OpenAL

$ ls openQ4.app/Contents/Frameworks/
game-mp_arm64.dylib   game-sp_arm64.dylib   libMoltenVK.dylib   renderer-vk_arm64.dylib

Apple's OpenAL has been deprecated since macOS 10.15, is software-only here, and exposes
no EFX. The AL_EXT_* list above is notably short compared with OpenAL Soft.

Reproduction matrix

s_maxSoundsPerShader reduces the number of sample variants loaded per sound shader, so
it is a proxy for total buffer count. Every run below is otherwise identical
(+set fs_basepath ... +map <map>), on the built-in speakers:

Map	s_maxSoundsPerShader	Result
game/airdefense1	0 (default)	Fails — CreateOpenALBuffer, returns to menu
game/airdefense1	3	Fails — same
game/airdefense1	1	Loads: 17747 msec to load game/airdefense1, clean exit
game/mcc_1 (campaign start)	1	Fails — same error
game/airdefense1	0, plus s_noSound 1	Loads: 15958 msec, Map media phases: ... sound=0

Two things follow:

1. Audio is the cause. With s_noSound 1 the identical map loads and stays loaded.
2. s_maxSoundsPerShader is not a workaround. It only moves the threshold. It is
   enough for airdefense1 but not for mcc_1, the first campaign map — so a player
   starting a New Game cannot get in at all.

Ruled out

• Audio device. Fails identically on a Bluetooth headset and on the built-in
  speakers. The device is not the variable.
• Missing or modified retail assets. All paks load with matching checksums; the
  fatal "media pk4 files are missing" path is a different failure that occurs only
  when fs_basepath is absent.
• Renderer. Occurs on the OpenGL package with the default renderer.
• Rosetta / architecture. Native arm64 throughout.
• Issue Crash on launch: SIGSEGV (11) in Renderer Upload Manager on macOS Tahoe (Apple Silicon M5) #73. That is a SIGSEGV in the renderer upload manager before the main
  menu. This is a caught idException after a successful map load, and this build
  logs Renderer upload manager: disabled legacy bridge.

Possible relationship to #94 and #104

This may be a packaging/build-provider question rather than a new engine bug.

• macOS: -Dmacos_openal_provider=system fails to build ('AL/al.h' file not found) #94 established that -Dmacos_openal_provider=system could silently fall back to
  Apple's OpenAL.framework, because Homebrew's openal-soft is keg-only and so
  missed by pkg-config.
• Resolve levelshot, macOS provider, dialogue media, and loadscreen issues #104 lists a fix to "make the macOS system OpenAL option require the intended
  pkg-config provider instead of silently selecting Apple's incompatible framework
  headers".

The published 0.12.0 arm64 artifact nevertheless links Apple's framework at runtime:

$ otool -L openQ4.app/Contents/MacOS/openQ4 | grep -i openal
	/System/Library/Frameworks/OpenAL.framework/Versions/A/OpenAL

and reports OpenAL vendor: Apple Computer Inc. with no EFX and no HRTF, with no
OpenAL Soft dylib present in Contents/Frameworks/.

So: are the published macOS packages built with the OpenAL Soft provider? If the
release pipeline still resolves to the Apple framework, then the runtime symptom above
is the user-visible consequence of exactly the problem #94/#104 addressed at build time,
and the shipped artifact would need rebuilding with OpenAL Soft linked and bundled.

Suggested fix

Link and bundle OpenAL Soft in the macOS packages. Beyond this crash it should also
restore EFX/reverb and HRTF, both currently reported unavailable.

Separately, and regardless of provider: a failed sound-buffer allocation currently
raises a fatal idException that tears down a fully-loaded map. Degrading gracefully —
a bounded buffer pool with reuse, or simply dropping the sample — would keep a level
playable instead of bouncing the player to the main menu with no on-screen explanation.

Log

Full openq4.log from a reproducing game/mcc_1 run and the
collect_macos_support_info.sh archive are available on request — happy to attach both
(redacted) if useful. Key sections inline:

Filesystem, renderer and OpenAL init

openQ4 0.12.0 (macos-arm64, Aug 18 2026 21:47:22)
Filesystem paths: fs_basepath='<retail root>' fs_game='baseoq4'
Setup OpenAL device and context... OpenAL output mode '5.1 surround' requested,
  but this build does not expose ALC_SOFT_output_mode; using the runtime default.
OpenAL ALC version: 1.1
OpenAL vendor: Apple Computer Inc.
OpenAL renderer: Software
OpenAL version: 1.1
OpenAL requested device: system default
OpenAL active device: MacBook Air Speakers
OpenAL HRTF: unavailable
OpenAL output mode: unavailable
OpenAL EFX is not available in this build; reverb wet send disabled.
sound system initialized.
Renderer bootstrap: selected=LegacyGL2Compat requested=auto bridge=ARB2
  caps={GL 2.1 unknown, VBO:0 PBO:1 UBO:0 VAO:0 MRT:1 FBO:1 ...}
Selected game module: logical='game_sp' binary='game-sp_arm64'

Map load completes, then the abort

loading maps/game/airdefense1.aas128
Entering doom_main()
Exiting doom_main()
Spawning entities
OpenAL Error: Invalid Value (0xa003), @ ../src/sound/OpenAL/AL_SoundSample.cpp 565
------------ Game Map Shutdown --------------
---------------------------------------------
********************
ERROR: idSoundSample_OpenAL::CreateOpenALBuffer: error generating OpenAL hardware buffer
********************
------------ Game Map Shutdown --------------
--------------- Game Shutdown ---------------

Same map with s_noSound 1 — loads and stays loaded

 15958 msec to load game/airdefense1
Map load phases: renderWorld=540 gameInit=3424 playerSpawn=407
                 mediaFinish=11565 settle=22 total=15958 msec
Map media phases: render=11565 sound=0 decl=0 bse=0 loadSize=0 ui=0 total=11565 msec

---

Unrelated note for other macOS reporters: collect_macos_support_info.sh prints a
review reminder, which is worth heeding — its system/hardware.txt includes the machine
serial number, Hardware UUID and Provisioning UDID, and system/uname.txt includes
the hostname. Easy to miss before attaching to a public issue.

[themuffinator]

Thank you for the exceptionally clear report and reproduction matrix — the s_maxSoundsPerShader threshold test and the s_noSound 1 control made the failure mechanism unambiguous.

I confirmed the release workflow was still selecting Apple's framework, so the published package did not contain OpenAL Soft despite the earlier migration work. PR #131 now:

• builds and bundles pinned OpenAL Soft 1.25.1 in the macOS app;
• verifies the package cannot silently retain Apple's framework or an absolute Homebrew/MacPorts dependency;
• includes signing, architecture, macOS-floor, licence-notice, and corresponding-source checks; and
• makes sample-buffer allocation/upload failure non-fatal, so resource exhaustion cannot tear down a map that already loaded successfully.

The local engine build and package-policy suite pass, and hosted macOS validation is next. I’ll keep this issue open until the fix is merged. A retest of game/mcc_1 and game/airdefense1 from the resulting arm64 artifact would be especially valuable afterward, including the OpenAL vendor, EFX, and HRTF lines from openq4.log.

[themuffinator]

Thank you again for the excellent reproduction details and Apple Silicon testing. The fix is now merged in #131.

The next arm64 package will use the bundled, checksum-pinned OpenAL Soft backend instead of Apple's legacy OpenAL framework. It also makes an individual OpenAL buffer allocation/upload failure non-fatal, so one exhausted sound can no longer abort the entire level back to the menu.

When a package containing merge commit dec0b8f6 is available, could you please retest both game/mcc_1 and game/airdefense1 on the same Mac? A complete logs/openq4.log would be especially helpful. Near startup it should identify OpenAL Soft and report the EFX/HRTF capability lines; please include those lines and note whether each map reaches gameplay and retains normal audio.

The automated arm64, Intel, universal2 OpenGL, and universal2 Metal build/package checks all passed. Real Apple-hardware confirmation is still the important final sign-off, so please reopen this issue if either map still returns to the menu or audio regresses.

[DrHowarth]

will keep an eye out for it and shall test when available Sent from my iPhoneOn 26 Aug 2026, at 21:27, Prof. Muffi Nator PhD ***@***.***> wrote:themuffinator left a comment (themuffinator/openQ4#122) Thank you again for the excellent reproduction details and Apple Silicon testing. The fix is now merged in #131. The next arm64 package will use the bundled, checksum-pinned OpenAL Soft backend instead of Apple's legacy OpenAL framework. It also makes an individual OpenAL buffer allocation/upload failure non-fatal, so one exhausted sound can no longer abort the entire level back to the menu. When a package containing merge commit dec0b8f is available, could you please retest both game/mcc_1 and game/airdefense1 on the same Mac? A complete logs/openq4.log would be especially helpful. Near startup it should identify OpenAL Soft and report the EFX/HRTF capability lines; please include those lines and note whether each map reaches gameplay and retains normal audio. The automated arm64, Intel, universal2 OpenGL, and universal2 Metal build/package checks all passed. Real Apple-hardware confirmation is still the important final sign-off, so please reopen this issue if either map still returns to the menu or audio regresses. —Reply to this email directly, view it on GitHub, or unsubscribe.Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS and Android. Download it today! You are receiving this because you authored the thread.Message ID: ***@***.***>