Context
The 2026-05-07 audit found that home check reports error(DebugAllocator) ... leaked on exactly 2 home-os kernel files:
- `kernel/src/drivers/wifi/broadcom.home`
- `kernel/src/drivers/wifi/iwlwifi.home`
These are the only checker-side memory leaks in the corpus. Both files trigger consistently and in isolation, making them easy repros.
Reproduction
```bash ./zig-out/bin/home-debug check /Users/glennmichaeltorregosa/Documents/Projects/home-os/kernel/src/drivers/wifi/broadcom.home ```
Should emit a leak trail from the DebugAllocator (the in-process allocator used in debug builds).
Hypothesis
Some specific code path in either the parser or typechecker, exercised only by these 2 files, allocates without freeing on the success path. Could be:
- A specific node type with a custom allocation that's missed in deinit
- An error path that leaks before recovering
- An interaction between recent changes (#28 unified struct-body, #38 alias tracking) and a kernel-syntax shape these 2 files use
Approach
- Reproduce both
- Use the leak trail's stack to identify the alloc site
- Add corresponding deinit/free
- Verify with both files plus a fuzz pass
Why this matters
Compiler hygiene — these leaks make the debug build technically broken on these inputs. Likely small fix (~5-30 lines). Won't directly affect kernel pass-rate but signals checker correctness.
Reference
- 2026-05-07 audit: `~/Desktop/home-os-kernel-audit-2026-05-07.tsv`