Replace audio backend with miniaudio

[binary1248]

By popular demand.

Work in progress.

Here be dragons.

[ChrisThrasher]

master...miniaudio_rebase

I took a shot at rebasing and fixing merge conflicts. Some of the conflicts were easily fixed but I'm still not sure I correctly fixed the conflicts in src/SFML/Audio/Sound.cpp. I can't get it to compile.

[ChrisThrasher]

I have most of the build errors fixed and removed all the imgui and imgui-sfml source code. I still need to fix compiler warnings and clang-tidy errors but this PR is at least in a state where it passes most of CI which means more people should be able to test it out.

The iOS jobs fail for reasons I don't totally understand. It's hopefully a matter of ensuring that Obj-C headers are being compiled as Obj-C instead of C or C++. Hopefully miniaudio actually works on iOS or else this PR is going to have a pretty tough decision to make.

[codecov]

Codecov Report

Attention: Patch coverage is 3.61174% with 1281 lines in your changes are missing coverage. Please review.

Project coverage is 42.76%. Comparing base (6eaf300) to head (fe5e54d).
Report is 15 commits behind head on master.

❗ Current head fe5e54d differs from pull request most recent head 13490da. Consider uploading reports for the commit 13490da to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2749      +/-   ##
==========================================
- Coverage   43.65%   42.76%   -0.90%     
==========================================
  Files         253      253              
  Lines       20940    22019    +1079     
  Branches     5139     5440     +301     
==========================================
+ Hits         9142     9417     +275     
- Misses      10785    11717     +932     
+ Partials     1013      885     -128     

Files	Coverage Δ
include/SFML/Audio/InputSoundFile.hpp	71.42% <ø> (-3.58%)	⬇️
include/SFML/Audio/OutputSoundFile.hpp	0.00% <ø> (ø)
include/SFML/Audio/SoundFileFactory.inl	100.00% <ø> (ø)
include/SFML/Audio/SoundFileReader.hpp	100.00% <100.00%> (ø)
include/SFML/Audio/SoundFileWriter.hpp	100.00% <ø> (ø)
include/SFML/Audio/SoundRecorder.hpp	0.00% <ø> (ø)
src/SFML/Audio/SoundFileWriterFlac.hpp	100.00% <100.00%> (ø)
src/SFML/Audio/SoundFileWriterOgg.hpp	100.00% <100.00%> (ø)
src/SFML/Audio/SoundFileWriterWav.hpp	100.00% <100.00%> (ø)
include/SFML/Audio/AudioResource.hpp	0.00% <0.00%> (ø)
... and 24 more

... and 61 files with indirect coverage changes

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ebf916e...13490da. Read the comment docs.

[binary1248]

miniaudio officially supports iOS and building for it even has a dedicated subsection in its documentation.

[ChrisThrasher]

I think my hunch was right. This should be fixable by telling CMake what specific files need to be compiled as Objective-C++.

[ChrisThrasher]

I've fixed almost all of the compiler warnings. The last one to address is here where MinGW appears to have found a null dereference inside of mini audio. I'm not sure if this warning is triggered by code we wrote or what.

We'll probably just pragma around this but I wanted to highlight it in case someone had an idea for a different fix.

Lastly I still have to fix those iOS jobs and then'll have 100% green CI :)

[ChrisThrasher]

I was able to fix the iOS build errors and added and workaround to deal with that MinGW warning that I didn't otherwise know how to fix. CI should be 100% green now!

[ChrisThrasher]

#2640 made sure that a sound always had an associated sound buffer but that was merged after this PR's branch was originally started. We need to make sure that this new implementation maintains that same property. I'd expect the underlying pointers within the implementation to always be non-null no matter how a user may use the API.

[ChrisThrasher]

Rebased on master and applied a few changes recommended by Kimci.

[ChrisThrasher]

Rebased on master and fixed minor merge conflict.

[ChrisThrasher]

I reenabled the clang-tidy check which is catching this pure virtual function call bug so that we can get better feedback once we start trying to fix it.

[binary1248]

@danieljpetersen The pure virtual method call issue should be fixed now.

[ChrisThrasher]

The unit tests in #2927 are still crashing on the copy construction of sf::SoundSource when I rebase them on this PR. This time UBSan is stating a pure virtual function is being called.

[ChrisThrasher]

My Audio unit tests are now passing without any crashes. There is however this oddity in the sf::SoundSource tests

    SECTION("Set/get pitch")
    {
        SoundSource soundSource;
        soundSource.setPitch(42);
        CHECK(soundSource.getPitch() == 0); // Why not 42 here?
    }

    SECTION("Set/get volume")
    {
        SoundSource soundSource;
        soundSource.setVolume(0.5f);
        CHECK(soundSource.getVolume() == 0);
    }

    SECTION("Set/get position")
    {
        SoundSource soundSource;
        soundSource.setPosition({1, 2, 3});
        CHECK(soundSource.getPosition() == sf::Vector3f());
    }

    SECTION("Set/get relative to listener")
    {
        SoundSource soundSource;
        soundSource.setRelativeToListener(true);
        CHECK(!soundSource.isRelativeToListener());
    }

    SECTION("Set/get min distance")
    {
        SoundSource soundSource;
        soundSource.setMinDistance(12.34f);
        CHECK(soundSource.getMinDistance() == 0);
    }

    SECTION("Set/get attenuation")
    {
        SoundSource soundSource;
        soundSource.setAttenuation(10);
        CHECK(soundSource.getAttenuation() == 0);
    }

The setters and getters seem to just not work. Setting a value does not result in being about to get the old value. The setters and getters behave as one would expect with the OpenAL backend.

[binary1248]

SoundSource was downgraded from being a semi-concrete class managing the actual OpenAL sound source to being an abstract interface with common functionality that is used by Sound and SoundStream. If you work around the fact that you shouldn't even be able to instantiate a SoundSource object by defining an empty derived class you can't expect it to work properly because there is no backing implementation behind it. If you want to test the SoundSource interface instantiate a proper object (Sound or Music) and test through a SoundSource& to the object.

[ChrisThrasher]

CI is failing due to Homebrew updating clang-tidy which includes some new checks. I fixed that in #2953.

[danieljpetersen]

Thank you Binary! That did indeed fix it. And thank you Bambo, thank you Thrasher!

Yesterday I wrote a little sound interpolator for the audio section of my codebase, where you give a duration and an easing formula for any given music/sound property(s). I've got it working in my game now to transition music on scene changes, it's especially nice when going from high tempo to low tempo music.

I'm looking forward to whenever this gets merged into master.

[eXpl0it3r]

The assets of the new example need to be referenced in the asset_licenses.md file.

It's a really cool PR! 😎

Beyond getting rid or OpenAL and its LGPL implications, it's awesome to see the natural use of SFML 3 features like Angle and C++17(-ish). 🎉

[binary1248]

Rebased and updated.

[eXpl0it3r]

The asset_licenses.md file still needs updating, but we can do that later as well. 😊

[eXpl0it3r]

Big thanks to @binary1248 for taking on this work and everyone for reviewing! 🎉