fix: three crashes found while packaging

[MrDavid5465]

Split out of #53 at your request — these are bugs the packaging work exposed, not packaging changes. None of them are needed for that PR. Three commits, five files, no build-system or CI changes.

sounddev_init never returns

sounddev_init() is declared int and its last statement was a bare call to usb_generic_shaker_init(...). Falling off the end of a non-void function is undefined behaviour, so new_sound_device() read whatever happened to be in the return register and treated most initialised devices as failures. The PulseAudio streams had in fact connected — visible in qpwgraph — but the devices were freed and never fed telemetry, so nothing shook. Observed as 22 of 24 configured shakers reporting "Could not initialize Sound Device".

One word: return.

--help frees uninitialised pointers

monocoque.c and monocoque-cli.c malloc both Parameters and MonocoqueSettings, then --help / --version / bad parameters goto cleanup_final, which is:

cleanup_final:
    monocoquesettingsfree(ms);
    free(ms);
    freeparams(p);
    free(p);

Every pointer in both structs is freed before any of them was assigned, so malloc leaves them pointing at whatever was in the heap block. Found by installing the released debian-latest .deb on Debian forky, where monocoque --help aborts with free(): invalid pointer; gdb put it in monocoquesettingsfree() called straight from main(). It is undefined behaviour everywhere — forky's glibc is simply the first to check, which is why the same source looks healthy on Debian stable and both Fedoras.

calloc in all three entry points, including monocoque-gui.c, where freeparams() walks the same struct on its early-exit paths.

gmonocoque segfaults on launch for common configs

Three related problems in mgui/uiconfighelper.c, the first of which is a behaviour change worth your judgement:

find_default_config() required sim, api and car to all be present, continue-ing past any entry missing one. conf/monocoque.config — the example the README points at — sets sim and car and never mentions api, so every entry was skipped and the function returned -1. This now treats a missing key as unconstrained rather than disqualifying (absent behaves as "default"). That's the interpretation the shipped example config implies, but if you intended the stricter reading I'll rework it.

The other two are straightforward guards: config_num < 0 now falls back to entry 0 rather than passing -1 to config_setting_get_elem, and selectedconfig / config_devices are NULL-checked, because config_setting_lookup dereferences its argument — which is where it actually died, inside libconfig, before the window was ever drawn.

---

Verified: builds clean against current master, and monocoque --help runs under MALLOC_CHECK_=3. The original abort was reproduced on Debian forky from the released .deb.

🤖 Generated with Claude Code

[Spacefreak18]

i might remove some of the comments but this is good.