libobs: bound volmeter plane index against MAX_AV_PLANES

[summeroff]

What

Adds an explicit plane_nr < MAX_AV_PLANES bound to the two plane-scan loops in libobs/obs-audio-controls.c (volmeter_process_peak and volmeter_process_magnitude).

- for (int plane_nr = 0; channel_nr < nr_channels; plane_nr++) {
+ for (int plane_nr = 0; channel_nr < nr_channels && plane_nr < MAX_AV_PLANES; plane_nr++) {

Why

Both loops index data->data[plane_nr] while plane_nr is advanced solely by the channel_nr < nr_channels condition — plane_nr itself is never checked against MAX_AV_PLANES.

It is safe today only because nr_channels is produced by get_nr_channels_from_audio_data(), which counts the non-NULL planes in the same data->data[] array. That guarantees the loop encounters enough planes before plane_nr leaves bounds, for any plane layout (contiguous, interleaved, or trailing-NULL).

The footgun: the safety lives in a different function than the unguarded access. If nr_channels were ever derived from a source's declared channel/speaker layout instead of from counting planes, the loop would read past data->data[MAX_AV_PLANES - 1]. Because MAX_AUDIO_CHANNELS == MAX_AV_PLANES == 8, the existing CLAMP in get_nr_channels_from_audio_data() would not catch that case either.

This is a behavior-preserving hardening change — under the current invariant the new guard is never the condition that ends the loop.

Testing

• Built the libobs target locally (VS 2022 / RelWithDebInfo) — compiles clean, no new warnings.
• Change is a no-op under current data flow; the guard only bounds the index.

Note on the clang-format CI check

The clang-format check is expected to come back red, and it is not caused by this change. CI runs clang-format (v19) over the entire changed file, and obs-audio-controls.c carries one pre-existing deviation unrelated to this PR (the obs_fader_add_callback signature, last formatted under the old 80-col config). That line was deliberately left untouched to keep this diff minimal and focused on the actual fix. Happy to fold in the one-line format normalization if preferred.

🤖 Generated with Claude Code