4.0 KiB
Codex final verification (RV17, 2026-10-12)
At b244567, full parity is not yet established: 6 P3 findings; no verified P1/P2 findings.
RV16’s four register omissions are resolved. Its FreeType-only backend finding is resolved on Linux, but remains incomplete for Apple targets. No incorrect claim in the existing deviation register was verified.
| Severity | Verified finding | Location |
|---|---|---|
| P3 | FreeType-only backend remains unavailable on Apple. Rust explicitly ignores the feature there; Zig permits .freetype, including discovery-free fallback handling on Darwin. Unregistered. |
build.rs:50 |
| P3 | Z19 introduces a feature-unification test defect. Enabling ghostty-font/font-backend-freetype without the CLI forwarding feature selects no discovery, but the CLI test asserts discovery exists. The assertion reproduces compiler error E0080. Conversely, enabling the forwarding feature on Apple selects the wrong assertion. |
list_fonts.rs:140 |
| P3 | Inspector font rendering differs. Rust uses 16 × display scale and stb_truetype; Zig uses 32 pixels and FreeType. At scale 1, font size is halved. Neither difference is registered; VP’s scenarios do not exercise the Inspector. |
inspector.rs:32 |
| P3 | Inspector DCS descriptions differ. Rust formats fields and slices using its own debug syntax; Zig’s Parser.Action.format delegates to Zig 0.15.2’s formatting rules. The displayed/filterable description differs. Unregistered. |
termio.rs:269 |
| P3 | Inspector demo appears in release builds with assertions. Rust gates it on debug_assertions, enabled by this repository’s release profile; Zig gates it strictly on .Debug, excluding .ReleaseSafe. Unregistered. |
surface.rs:137 |
| P3 | Metal buffer allocation correction is unregistered. Rust allocates twice the required bytes; Zig multiplies that quantity by sizeof(T) again. This changes GPU-memory allocation. Preserve the correction and register it. |
buffer.rs:128 |
Verification performed:
- 461 Zig files inventoried and interface-screened; 2,801 public declarations screened, with targeted implementation comparisons. This is a breadth audit, not exhaustive behavioral-equivalence proof.
- 1,980 config comparisons across 198 keys, plus 166 CLI comparisons: zero mismatches.
- 2,000 fresh terminal differential cases: 1,970 identical results, 30 Zig aborts, zero Rust aborts or defined-result mismatches. The documented 390k campaign was not rerun.
- 2,517 successful unit-test executions, including both font variants; seven ignored executions. Fixture-writing tests were excluded.
- Existing three visual captures and 36 sprite sheets remain pixel-identical. WASM Node smoke test passed.
- Linux FreeType-only CLI behavior, argument ordering, fallback gates, and the maximum-length
NOTIFY_SOCKETregression were checked.
Inherently unavailable for native verification on Linux: Apple Metal/IOSurface/CoreText, Apple input APIs, Mach handling and packaging; Windows ConPTY/Win32 behavior; native FreeBSD behavior. Swift was excluded.
Other verification limits: existing binaries were used; no fresh full build, GUI captures, live portal/D-Bus integration, real-browser rendering, or native io_uring execution was performed. Platform findings above are source-verified.
No files were modified; the worktree remains clean.