[facelessdevhack]

What does this PR do?

Clang reports warnings in debug and lightly optimized amalgamation builds, so the existing warning check skips Clang. This change makes those builds warning-free and enables the same -O0/-O1 check for every selected compiler.

• Spell the opcode-name prefix offset as &(#o)[3].
• Retain ten optional inline helpers and mark each individually as intentionally unused.
• Initialize the six borrowed numeric views once before the entity loops, preserving their ownership and avoiding per-entity clearing.
• Explicitly initialize the guarded VM index-write temporary, matching the already initialized helper path and clearing a GCC 13 -O1 diagnostic.
• Remove the compiler-specific warning-check gate and SKIP.

Fixes #1872.

Warning classification is posted in the issue comment. Per optimization level, Linux Clang 18 reports 97 deliberate prefix offsets and 10 optional unused helpers: 107 per invocation, 214 across both modes. Apple Clang 21 additionally reports six view-initialization diagnostics: 113 per invocation. All sites were audited; no runtime or concatenation bug was found. Each numeric view's consulted fields are written by slot_view before use. No warning class is globally suppressed, and optional APIs remain available.

Validation

• Enabled warning check against original production source: 5 passed, 2 failed, rejecting the real Clang warnings at both optimization levels. Fixed source: 7 passed, 0 failed with Apple Clang 21 and GNU GCC 15.2, including embedded evaluation and zero-live-allocation teardown.

• Linux GCC 13 -O1 also reproduces a pre-existing guarded INDEX_SET temporary warning on unmodified main. The corrected explicit initialization passes the full amalgamation gate (7/7); optimized native Clang assembly is byte-identical to baseline.

• All 256 possible uint8_t opcode-name inputs produce byte-identical results before/after; indexed spelling compiles with -Wall -Wextra -Werror on both native compilers at both optimization levels.

• Full unsharded Linux arm64 GCC 13 and Clang 18 release suites: 4292/4292 passed each, 0 failed, 12 declared skips each (disabled extensions, unavailable optional tooling/assets, and build/architecture-specific cases).

• make precheck: 19 passed, 0 failed, 0 skipped.

• make test-changed: 224/224 passed, 0 failed, 0 skipped. Runtime sources and the direct CI-invoked amalgamation script are explicitly reported outside the selected-section coverage.

• Focused index/error-path, container API/escape-view control and builtin direct/indirect checks: 55/55 passed; boolean behavior: 165/165 passed in each of default, interpreter, and non-strict modes.

• Numeric-read gate: 190 reads, 72 reviewed functions, 0 violations/stale entries. Bash 3.2 syntax and git diff --check: passed.

• Additional Linux x86-64 compile-only check with JIT enabled: GCC 13.3 and Clang 18 are warning-free at both -O0 and -O1; correct target headers and all four ELF64 x86-64 objects verified.

Hosted validation at 606006c503b0e1bbdc6ab3abb7e87e8ecfe9c005: 25 successful checks, six conditional skips, zero failures. CI run completed successfully, including the hosted GCC suite and x86 JIT differential. CodeQL, documentation/WebAssembly and CodSpeed workflows also succeeded.

Prepared with OpenAI Codex assistance and independently reviewed by a separate Codex agent.

Checklist

• Complete local suite passes on GCC and Clang (make plus unsharded tests/run_all_tests.sh; counts and skips above).
• No new builtins, library functions, or examples; their signature/doc/header requirements do not apply.
• Internal changelog fragment added; CHANGELOG.md is unchanged.

[codspeed]

Merging this PR will not alter performance

✅ 13 untouched benchmarks

---

_{Comparing facelessdevhack:fix/clang-amalgamation-warnings (606006c) with main (5a7b703)}

[InauguralPhysicist]

Thank you, @facelessdevhack. This is exactly the fix #1872 needed, and the review was easy because of how you did it. In particular:

• &(#o)[3] keeps op_name derived from the enum spelling, which is Opcode metadata tables have already drifted: OP_TRAJECTORY_SLOT missing from the untrusted-chunk verifier, 4 opcodes disassemble as "???", 15 desync the disassembler #737's whole point. You checked all 256 uint8_t inputs byte for byte before and after, instead of trusting that it's equivalent.
• The per-site audit. You classified all 107 (Linux Clang) and 113 (Apple Clang) warnings, found no real concatenation or runtime bug, and suppressed nothing globally.
• The hoisted EigsViews. I re-read both loops: each view is rewritten by slot_view before its pointer is used in the same iteration, so the hoist is safe and the per-entity cost stays at zero.
• Removing the clang SKIP. That was the gap I had to leave open in Fix warning-free debug amalgamation builds #1799. The -O0/-O1 warning check now runs for every compiler, and you showed it rejecting the unfixed source (5 passed / 2 failed) before passing the fixed one.

A heads-up only, nothing to do: several of the helpers you marked __attribute__((unused)) (list_get_num, dict_value_get_num, dict_value_set_num, buffer_read_byte, slot_is_numeric) are already scheduled for deletion by the Phase 2 numbers work (#1758, PR #1776). Annotating them was the right call for main today, since deleting them here would collide with that branch. I've added "delete every helper left without a caller" to Phase 2's final sub-round.

Merging through the queue now. If you'd like another one in the same territory, #1876 (the suite's failure excerpt for long sanitizer reports shows the tail but loses the head) and #1877 (strict_shape_contract.py has no pin-update mode) are both open.