fix(main/openal-soft): disable dlopen() of opensles for luanti

[robertkirkman]

• One of two potential solutions to [Bug]: after openal-soft bump to 1.24.2, sound stopped working in x11/luanti #23148

• compared to fix(x11/luanti): force-link to libOpenSLES.so #23185

  • Potential advantages: would not require editing any other reverse dependency of openal-soft that could in the future be found to be affected by the same issue
  • Potential disadvantages: requires editing the openal-soft package

• mpv --ao=openal and mplayer -ao openal continue to work with or without this reversion, so it seems ok to me. however I do not understand why this is only required for luanti, and I have not found any other reverse dependency of openal-soft that was negatively affected by the bump to openal-soft 1.24.2. MRE created and submitted upstream

[Biswa96]

Instead of reverting the patches, would it be possible to disable the HAVE_DYNLOAD constant in alc/backends/opensl.cpp file?

[robertkirkman]

Thank you, I did not really notice that earlier, but yes, adding -DHAVE_DLFCN_H=OFF to TERMUX_PKG_EXTRA_CONFIGURE_ARGS appears to have an identical effect, and it also solves the problem, while being much fewer lines, so I have changed it to that.

[Biswa96]

I did not mean to add that option globally. There are many source files which depend on that option. So, I am not sure if -DHAVE_DLFCN_H=OFF breaks any existing stuff.

[robertkirkman]

Well, I did not think that other stuff will be affected because it doesn't seem used in the Termux build of this package,

but I do not fully understand the problem, since I have no idea why only one app is affected by the problem and not other apps,

and yes you are correct, so I will write this exactly how you suggested. Also, the exact way you suggested does also work to fix the problem.

[robertkirkman]

@Biswa96 something still seemed confusing to me, even more so than this problem already is, so I carefully double checked everything, and unfortunately, I have to report the following:

• my original bisection and whole commit reversion was correct, and -DHAVE_DLFCN_H=OFF or -DALSOFT_DLOPEN=OFF by themselves were both also able to work
• but, very unfortunately, I am regretfully saddened to have to report that when you sent me your suggestion about HAVE_DYNLOAD, I made a mistake while testing it and I did not test it correctly, so I did not realize that it didn't work as written. When I double check the exact method you suggested, unfortunately it results in this error:

CANNOT LINK EXECUTABLE "mplayer": cannot locate symbol "SL_IID_ENGINE" referenced by "/data/data/com.termux/files/usr/lib/libopenal.so"...

I have now thoroughly tested this, and it appears that at the very least, in order for the HAVE_DYNLOAD reversion to work, this single line in the CMakeLists.txt must also be reverted. I have just pushed a version that is currently able to fix the problem for me.

[robertkirkman]

After continuing to debug this problem for a while, I have managed to create:

• A possible way to solve the problem by editing the luanti package instead of the openal-soft package, which I have put in a separate PR, and I have written an outline of potential advantages and disadvantages of each method in both PRs
• An MRE which I have submitted upstream here ALSOFT_DLOPEN=ON ALSOFT_REQUIRE_OPENSL=ON builds not working at runtime inside Termux on random Android devices kcat/openal-soft#1111