[tignear]

Parameter binding and AST-to-IR conversion resolved a nonnumeric value twice: numeric fallback first produced an X/Z or wide literal and discarded it, then the caller resolved the same literal again for publication. Return the numeric result and fallback literal together so each binding resolves a literal at most once.

Direct numeric values still skip literal resolution. A number recovered from a fallback literal retains numeric precedence, while unknown/wide literals remain available to dependent declarations. Unresolved expressions, type-publication order, numeric formatting and the generate binder's existing fallback are preserved. The result is local to each binding; no cache crosses declarations or analyses. Public APIs are unchanged.

Owning-crate regressions compare against the former two-stage resolver for dependencies, selects, signed shifts, stale self-values, two-state conversion, wide values and unresolved names. They verify numeric recovery from an unknown dependency and the generate binder's unresolved fallback. Complete module AST/IR runs at 16/64/256 parameters count one literal evaluation per binding and per exported IR parameter, instead of two.

Performance

Base: synchronized master ef37b2c6c (#1170). A manual optimized heliodor-dev probe replays the same synthetic unknown prefix with both resolvers in one process, alternates their order, asserts every 32-bit X mask and reports seven-run medians:

Parameters	Old two-stage (ms)	Retained result (ms)	Literal calls before/after
32	0.570	0.361	64 / 32
128	6.296	1.548	256 / 128
512	28.168	10.178	1,024 / 512

The large replay sample is about 2.8x faster. This measures value resolution, metadata publication and mask checks, excluding parsing, enclosing-declaration discovery and backend/simulation work. Counts establish the eliminated second evaluation independently of timing variability.

Separate full AST runs do not establish an overall speedup on the shared machine: at 512 four-state module parameters, AST time was 285.531/481.522 ms before/after while parsing was 80.219/137.527 ms; reverse-generate AST was 3,961.605/8,534.573 ms while parsing was 85.630/209.199 ms. Preliminary generated size queries still dominate that path. Detailed phase timings and limitations are retained in docs/internals/sv-analysis-performance.md; no universal speedup or linearity is claimed.

Validation

Locked Nix shell, default managed Cargo targets:

• Formatting and cargo clippy --locked -p celox-sv-analyzer -p celox-frontend-sv --all-targets --all-features -- -D warnings.
• cargo test --locked -p celox-sv-analyzer -p celox-frontend-sv --all-features --quiet — 230 passed; doctests passed. The new manual timing probe is deliberately ignored in ordinary runs and was executed separately.
• cargo test --locked -p celox --features sv-dpi --test systemverilog --test sv_unpacked_array --test sv_dpi --test sv_packages --test interface --test advanced_interface --quiet — 1,617 SV, 52 DPI, 7 array, 37 package and 216 interface tests passed; 39 existing interface exclusions remain. Includes native/Cranelift/Wasm and applicable parallel/four-state variants.
• cargo test --locked -p celox-test-suite --features verilator,icarus --quiet — 52 library/integration tests and 2 doctests passed; 9 existing external oracles ignored.
• cargo doc --locked -p celox-sv-analyzer -p celox-frontend-sv --no-deps and pnpm docs:build.
• Manual paired probe: cargo test --locked -p celox-sv-analyzer --profile heliodor-dev --lib compare_retained_and_repeated_parameter_resolution -- --ignored --nocapture — passed.
• Existing four-state, numeric and ranged type_queries modes at 32/128/512 entries, asserting their IR outputs.

No shared executable cases, external adapters/emitter or retained proof/report artifacts change. External simulators were not rerun for this analyzer-only optimization under AGENTS.md/CONTRIBUTING.md. Default NAPI does not enable this frontend.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-10T12:38:45.646296Z	21eb07c	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.