Five gen-verilog features have live emitter code and failing tests: seven guards that never ran

[gHashTag]

What

Seven tests across five files fail on master on both macOS and the Linux runner. Each
file has exactly one commit — the one that added the test alongside the feature it
tests — and none has been touched since:

file	added	commit subject (truncated)
verilog_r_si_1.rs	2026-05-22	wave-27: t27c gen-verilog emits __mul_noop(a, b) helper
verilog_initial_decl.rs	2026-05-23	wave-29: t27c gen-verilog hoist bench integer counters
verilog_translate_off.rs	2026-05-23	wave-30: t27c gen-verilog emit standalone synthesis …
verilog_array_literal_expr.rs	2026-05-23	wave-31: t27c gen-verilog ExprArrayLiteral in expression …
verilog_array_param_index.rs	2026-08-05	fix(gen-verilog): array-param element index -> part-select

The failing tests:

verilog_array_literal_expr.rs  r_ca_2_synthetic_no_comment_only_call_argument
verilog_array_param_index.rs   array_param_index_is_element_part_select
verilog_initial_decl.rs        r_vd_1_synthetic_no_integer_decl_inside_initial
verilog_initial_decl.rs        r_vd_1_real_uart_spec_no_integer_decl_inside_initial
verilog_r_si_1.rs              r_si_1_emitter_injects_mul_noop_helper
verilog_translate_off.rs       r_tr_1_synthetic_no_inline_translate_marker
verilog_translate_off.rs       r_tr_1_real_uart_spec_no_inline_translate_marker

Messages:

R-SI-1 helper missing: `__mul_noop` function declaration not found.
Expected exactly 2 module-scope `_bench_<name>_cycles` counters … got [].
array-param index did not lower to a 64-bit part-select
Error: Compile error: parse error in fn 'driver' near line 8:
       unexpected token after expression statement: KwReturn

The emitter code is still there

This is the part that makes it a defect rather than a cleanup. Checked on origin/master:

• __mul_noop — 9 hits in bootstrap/src/compiler.rs, including the emission itself at
  :11842: function [63:0] __mul_noop; // t27#1886: 64-bit, u64 products no longer truncate
• _bench_ — 6 hits · translate_off — 4 hits · part-select — 9 hits

So the features were not removed. The emitter still knows how to emit them and no longer
reaches that path for these inputs. That is a conditional regression, not a deletion, and
it will not show up in a diff of the feature's own code.

r_ca_2 is different in kind and possibly the most interesting: it fails at parse, not
emission — unexpected token after expression statement: KwReturn. Its fixture no longer
parses at all.

Why nobody noticed

cargo test -p t27c --tests stops after the first failing target; these are targets 43–73
of 73, so they never ran. That is #2382. The ratchet landed in #2383 now runs all 73 and has
these seven baselined, so they will not regress further unnoticed — but baselining is not
fixing, and this issue is the fixing.

What is NOT established

• Whether these ever passed. Each was added in the same commit as an implementation whose
  message claims the feature works, which makes "passed, then regressed" the likely reading —
  but I did not check out those commits and run them, so I am not asserting it.
  Born-failing and regressed need different responses, and the distinction is one bisect away
  for anyone who wants it.
• Whether the seven share one cause. Five files, two shapes (missing emission vs parse
  failure). They may be one regression or five.
• Nothing here is attributed to any particular commit. In particular this is not a claim
  about fix(compiler): dead-store elimination deleted every write to module state (Closes #2363) #2380, which touched the optimiser today — the baseline that surfaced these was taken
  after it, so it cannot distinguish.

Separately: four tests are baselined that CI passes

spec_first_layer2_packs_two_neuron_trits, spec_first_mlp2_…, spec_first_mlp3_… and
spec_first_neuron_n_accumulates_and_quantizes fail on macOS and pass on the Linux
runner. They are in the baseline, which means a genuine Linux regression in any of them
would be invisible to the ratchet. Being pruned separately.

Related: #2382, #2383, #2384.