WAV: Fix one frame overflow at the end

[DeeJayLSP]

This fixes AudioStreamWAV decoding one additional frame after the last

---

Problem

Let's assume a PCM data with exactly 3000 frames as an example. This means the range should be 0-2999.

When you mix in AudioStreamPlaybackWAV, it considers you have to decode p_frames frames into p_buffer. There is a calculation right before the decoder is called (limit, aux and target) that is supposed to prevent the decoder (do_resample()) to get an index beyond the PCM data range and repeat the while loop in respect to the audio loop mode.

Assuming the audio has no loop points and goes from the very first frame to the last, here's the problem: when playing forward, the last frame (end_limit in the mix function) is assumed to be the length of the audio data (3000), not the last index (2999). Because of that, target ends up being calculated considering that additional frame, out of the range, and ends up making do_resample() get one more frame than it should.

The reason no errors happen when playing PCM audio data is because of the DATA_PAD (which reads a 0 value frame at that position, leading to a barely audible pop), and doesn't crash with QOA data because the resampling workaround doesn't allow the offset to be above the last frame index.

Solution

Do what should be done: subtract 1 from the length whenever you point to the last frame.

Even if the while loop forces the audio to stop when the offset is beyond the last frame, it doesn't prevent that one extra end frame request from being passed to the decoder function.

If merged, any audio imported with any enabled Loop Mode and Loop End set to -1 will need to be reimported to remove the extra frame from the count. Not sure if this count as compatibility breaking, but existing projects should not be affected.

Considerations

I found this by experimenting with #83483 (makes AudioStreamWAV use PlaybackResampled, removing the need for workarounds and data pad) with a QOA audio with 4160 frames (the number becomes the decode buffer size when instantiating playback) and it would crash for not being able to get index 4160.

[DeeJayLSP]

Force-push because I messed up the wrap on the importer.

And reworded the commit/PR to make it clear this is a fix.

[akien-mga]

Thanks!

[akien-mga]

Cherry-picked for 4.3.1.