Build fixes: host portability, and 16 KB page alignment for the sink module

[Zum0DePapaya]

Four small build-script fixes. Three are host-portability issues that stop a fresh clone from building anywhere but the maintainer's machine; one is a page-alignment bug in the artifact that actually ships.

Discussed in the GameNative Discord #code-changes thread beforehand.

1. Derive the automake version from the host

main-build.sh patched the vendored libtool / libsndfileaclocal.m4 files with sed -i 's/1.15/1.18/g', which assumes automake 1.18 is installed. On any other host autoreconf then demands a version that isn't present and the build stops. Now derived from automake --version.

2. Seed .tarball-version so a fresh clone can configure

configure.ac derives the package version through ./git-version-gen, which needs either a reachable git tag or a .tarball-version file. The pulseaudio fork carries no tags and .tarball-version is gitignored, so on a fresh clone git-version-gen returns empty and configure.ac:34 aborts with git-version-gen failed.

Seeded with 13.0 only when absent. bootstrap.sh then rewrites it to 13.0-rebootstrapped, which is the exact version string the currently shipped binaries report — so this reproduces the existing artifacts rather than changing them.

3. 16 KB page-size linker flags for the module build

main-build.sh exports -Wl,-z,max-page-size=16384 -Wl,-z,common-page-size=16384 globally, and pulseaudio-module/CMakeLists.txt sets them too — but pulseaudio-module/build.sh invokes clang directly without them, and create-asset.sh copies from build64/, i.e. build.sh's output. So the module that actually ships is the one built without the flags.

Measured on the current pulseaudio-gamenative-20260612.tzst:

0x4000 modules/module-native-protocol-unix.so
0x4000 modules/libprotocol-native.so
0x1000 modules/module-aaudio-sink.so <- only this one

libpulsecore-13.0.so and the rest of jniLibs are 0x4000 as well. On a device booted with 16 KB pages the loader cannot map a 4 KB-aligned library, so dlopen of the sink module would fail; because the daemon runs with --fail=false it would stay up with no sink at all, which presents as "no audio on the pulseaudio driver" rather than an obvious crash.

Worth noting this isn't covered by Play's 16 KB compliance check, which scans lib/**/*.so inside the APK — all 27 of those are already correctly aligned. This module ships compressed inside assets/*.tzst and is extracted and dlopen'd at runtime, so the scanner never sees it.

Caveat: I don't have a 16 KB-page device, so I have not observed the dlopen failure in the field. What is verified is that building the module both ways from identical source gives 0x1000 without the flags and 0x4000 with them.

4. Use $HOME for the NDK path in the module build

pulseaudio-module/build.sh hardcoded /home/joshua/Android/Sdk/ndk/... while main-build.sh next to it already resolves the same NDK through $HOME. Without this, the module build fails immediately with aarch64-linux-android26-clang: No such file or directory for anyone else — including anyone trying to verify fix 3.

---

Verification

On Ubuntu 24.04 with NDK r27d (27.3.13750724), from a clean clone with only these four commits applied:

• main-build.sh -a arm64 completes, banner reports pulseaudio 13.0-rebootstrapped
• pulseaudio-module/build.sh produces module-aaudio-sink.so at LOAD align 0x4000
• The build reproduces the shipped artifacts byte-for-byte: pactl, libprotocol-native.so and module-native-protocol-unix.so from the .tzst, plus libpulsecore-13.0.so, libsndfile.so and libltdl.so from jniLibs

Only libpulse.so / libpulseaudio.so / libpulsecommon-13.0.so differ, purely because they embed the build prefix path — confirmed by building at two different prefix lengths and seeing the size delta scale accordingly.

No functional change to the built binaries beyond the alignment of the sink module. ac_cv_func_mkfifo=no is deliberately left untouched here — that's a separate discussion.