[gnosticgeek]

What and why

Two CI ratchets went red on the 0.12.0 release PR (#35) after the upstream merge: clang-tidy (lint-c) and the Mac host compiler-warnings check in Build check. The release shipped without these fixes, so main is red on both until this PR lands.

f983a79: clang-tidy. The LLVM 23.1.4 analyzer flags code that landed with the upstream merge.

• dm/gfx/nvn/nvn_mtl.m: a NOLINT, with the reason on the same line, on eight analyzer false positives:
  • BitwiseShift: the shift is by __builtin_ctz of a nonzero word.
  • ArrayBound (five cases): each index is bounded by the line before it (addressed ranges, vertex streams, g_mrb, g_mrbuf).
  • RetainCount (two cases): the objects are held by hm_put_id maps.
• t_depth16_clear, t_dlss_mfx, t_ndve_movie, t_pool_heap, t_upscaler_mfx: a NOLINT for StackAddressEscape on g_tex_live. mtl_tex_release reads the stored address through vm_read_overwrite and checks magic and serial, so a helper that returns its tex_t by value is safe.
• t_feedback: a NOLINT for bugprone-signed-char-misuse. The bf_key offsets are int8_t on purpose (-1 = none).
• t_pso_stall: a real fix. An archive that doesn't open is now left out instead of being added as nil.
• dm/tools/lint_c_baseline.tsv: regenerated with tidy --update. 11 file/check pairs now have fewer findings, and t_view_swizzle is back below main's count.

d27661c: host warnings. mtl_tex_init passed its const tex_t * to nvn_hm_put's void * value, which is a new -Wincompatible-pointer-types-discards-qualifiers warning (upstream stores (void *)1 at that point). The map only reads the address back, in mtl_tex_release as a const tex_t *, so an explicit cast is safe.

Checked

• clang-tidy ratchet: green in CI on f983a79. Before that it was run on an Apple silicon Mac the way CI runs it: LLVM 23.1.4, --against origin/main, rc 0, 220 findings against a baseline of 220 (down from 237), and 0 pairs above the baseline.
• Host warnings: the failing Build check on f983a79 (and on the release head 1550f95) reported exactly one new warning and 0 fixed-but-listed. That warning is at nvn_mtl.m:2373, which d27661c changes. A minimal clang repro gives the same warning without the cast and none with it. The rerun on d27661c is pending.
• Gfx tests for the touched files pass with GAME=0400486024DEC000:
  • depth16_clear, ndve_movie, pool_heap: pass.
  • dlss_mfx: 7 passed, 0 failed.
  • pso_stall: 41 passed, 0 failed.
  • upscaler_mfx: 12 passed, 0 failed.
  • view_swizzle: 554 passed, 0 failed.
  • feedback: 193 passed, 0 failed.
• t_feedback failed 1 of 193 on an earlier run where only a comment had changed, so it looks flaky. That failure isn't related to this change.
• No game was run. Apart from the t_pso_stall guard and the cast, the changes are comments and the baseline.

Docs

• No game files, keys or data extracted from a game in the change

🤖 Generated with Claude Code

https://claude.ai/code/session_016qteK29T18U8p1c9LsVY8g