fix: changeDevice() device selection and error reporting

[Colton127]

Description

Fixes two correctness defects in SoLoud.changeDevice(), plus three related ones found in the same code while making those correct.

changeDevice() could never select the default device

SoLoud.instance.changeDevice() with no argument is documented to select the system default output device, but it always threw SoLoudNoPlaybackDevicesFoundCppException. Dart passes -1, and Player::changeDevice() compared that int against devices.size():

if (devices.size() == 0 || deviceID >= devices.size())
    return noPlaybackDevicesFound;

-1 is promoted to SIZE_MAX by the usual arithmetic conversions, so the guard always fired. The device list is now consulted only for explicit (non-negative) IDs, and -1 passes nullptr straight through to deviceConfig.playback.pDeviceID, which is miniaudio's default-device selector. This mirrors what Player::init() already did. IDs below -1 are rejected as invalidParameter before any conversion to an unsigned type.

A failed device change reported success

if (result != SoLoud::SO_NO_ERROR)
    result = backendNotInited;   // assigns to a local
return noError;                  // ...then returns success regardless

If miniaudio could not initialize or start the replacement device, the caller was told the switch worked. It now returns audioDeviceFailedToStart, which Dart already maps to SoLoudAudioDeviceFailedToStartCppException. backendNotInited would be misleading here: the engine can stay initialized while its output device could not be replaced.

Note this PR only makes the failure visible — it deliberately does not add rollback, retry, or fallback-to-default.

Also fixed in the same code

• Use-after-free in Player::init(). With an explicit device ID, the device list was built inside an if block and playbackInfos_id pointed at one of its elements. The vector was destroyed at the end of that block, so soloud.init() handed miniaudio a pointer into freed heap. Reproduced under AddressSanitizer (details below).
• init() had the same signed/unsigned comparison, reporting IDs below -1 as noPlaybackDevicesFound. Now invalidParameter, matching changeDevice().
• listPlaybackDevices() leaked on every call. PlaybackDevice::name was a strdup()ed char * that nothing ever freed — one allocation per device per call, including the internal calls from init() and changeDevice(). It is now a std::string, which also lets the dangling pPlaybackInfos member become a local: it points at context-owned memory that ma_context_uninit() frees before the function returns.
• Out-of-bounds read in the device-name filter. bindings.cpp scanned the first 5 bytes of each name for control characters before checking strlen(name) <= 5, over-reading the buffer for short names. The length check now runs first; filtering behaviour is unchanged.
• Soloud::miniaudio_changeDevice() returned 0 when the audio thread mutex was null, reporting a successful device change when nothing had happened. It now returns UNKNOWN_ERROR (surfacing as audioDeviceFailedToStart), and NOT_IMPLEMENTED when built without the miniaudio backend.

Verification

The use-after-free and the leak were confirmed with a standalone AddressSanitizer harness built from the real plugin sources (miniaudio's null backend enumerates a device, so the explicit-device path is reachable without audio hardware).

Against the pre-fix code:

ERROR: AddressSanitizer: heap-use-after-free
READ of size 256 at 0x5120000001e8
  #1 ma_device_init            miniaudio.h:43662
  #4 SoLoud::miniaudio_init    soloud_miniaudio.cpp:540
  #5 SoLoud::Soloud::init      soloud.cpp:335
  #6 Player::init              src/player.cpp:273
freed by thread T0 here:
  #5 std::vector<PlaybackDevice>::~vector()
  #6 Player::init              src/player.cpp:269

...and 132 byte(s) leaked in 2 allocation(s), both strdup from Player::listPlaybackDevices(). The patched build is silent on both. Worth stressing that a passing run proves little on its own here — freed storage often still holds the right bytes, so the buggy version could return noError and play normally.

Tests

example/tests/tests/playback_devices.dart now covers the default selector, an enumerated device, rejected native selectors (-2 → invalidParameter, out-of-range → noPlaybackDevicesFound), and initializing on an explicit device — the use-after-free path, which nothing exercised before.

Two things to note

1. The WASM artifact has not been rebuilt. src/player.cpp is compiled into web/libflutter_soloud_plugin.wasm, so Web keeps the old behaviour until it is regenerated. I hit an unrelated failure building the Xiph libs locally (vorbis configure: must have Ogg installed! — --with-ogg="$OGG_DIR" expects an installed prefix layout, but the script builds ogg in-tree so the archive is at $OGG_DIR/src/.libs/). Since the artifact is normally yours to commit anyway, could you regenerate web/libflutter_soloud_plugin.{js,wasm} on merge? No exported signatures changed.

2. The changelog entry is filed under a placeholder ##### 4.1.7 (X Xxx 2026) heading — retitle or fold it in as you see fit. No version bump included.

No ABI, enum, exception, or binding signature changes.

Type of Change

• ✨ New feature (non-breaking change which adds functionality)
• 🛠️ Bug fix (non-breaking change which fixes an issue)
• ❌ Breaking change (fix or feature that would cause existing functionality to change)
• 🧹 Code refactor
• ✅ Build configuration change
• 📝 Documentation
• 🗑️ Chore

[alnitak]

Hi @Colton127, again, thank you very much for your support!

Since the artifact is normally yours to commit anyway, could you regenerate web/libflutter_soloud_plugin.{js,wasm} on merge?

Sure! I had no problems building wasm. For the record, I am using em++ version 4.0.22-git on macOS, but maybe this is not the problem.

Thanks again. Merging