test(verilog): terminate the r_ca_2 fixture's call statement

bootstrap/tests/verilog_array_literal_expr.rs

@@ -47,7 +47,7 @@ const SPEC: &str = r#"module RCA2Probe {
     }
 
     fn driver() {
-        consume([1, 2, 3, 4])
+        consume([1, 2, 3, 4]);
         return
     }
 }

docs/now/2026-08-22-rca2-terminator.md

@@ -0,0 +1,48 @@
+# NOW — the test died at parse and never reached its rule
+
+Last updated: 2026-08-22
+
+## Add the statement terminator the r_ca_2 fixture omits (Closes #2399)
+
+- Branch: `fix/2399-rca2-terminator`
+- Issue: #2399 · part of #2386
+
+### Что легло
+
+One character in `bootstrap/tests/verilog_array_literal_expr.rs` — `consume([1, 2, 3, 4]);`
+— plus one line pruned from `scripts/ci/test-baseline.txt` (373 → 372).
+
+**The test's name is wrong about the cause.** Four probes: `consume(7)` followed by
+`return` fails identically, and the same call with a `;` or as the final statement parses.
+The array literal is irrelevant; the condition is an expression statement without a
+terminator followed by another statement.
+
+With the terminator the test measures what it names: the emitter writes
+`consume(0 /* TODO: array literal [1,2,3,4] not yet lowered to Verilog */)` — a real
+argument **plus** a comment, which is what R-CA-2 requires.
+
+### Границы честности (BINDING)
+
+- **Array literals in call position are still not lowered.** The emitter says so and emits
+  `0`. R-CA-2 is the narrower rule that the argument must not be comment-only. This change
+  does not implement the feature and does not claim to.
+- **This is 6 of the 7 in #2386. One remains** (`array_param_index_is_element_part_select`).
+- Whether the parser should accept newline-terminated expression statements is a **language
+  decision**, raised in #2399 and not answered here. 28 of the 154 non-parsing specs fail
+  this way, including `specs/server/session.t27`.
+
+### A measurement I nearly got wrong
+
+Sweeping `t27c gen-verilog` over `specs/` reports **649 of 650 failing** — which sounds
+catastrophic and means almost nothing, because most specs are not hardware specs and that
+backend declines them legitimately. The right instrument is `t27c parse`: **496 parse, 154
+do not**, and 496 matches the repository's own recorded baseline.
+
+`t27c corpus`'s own help text warns about this exact class: *"a parser error count moves
+three orders of magnitude from one character and RISES when a real defect is fixed (T119)"*.
+
+### Evidence
+
+Mutant at `bootstrap/src/compiler.rs:15553` — the `0` dropped so the argument becomes
+comment-only — verified planted, with a local `FROZEN_HASH` reseal. Both tests FAILED;
+restored, `2 passed; 0 failed`. Neither `compiler.rs` nor `FROZEN_HASH` is in this diff.

scripts/ci/test-baseline.txt

@@ -1,7 +1,7 @@
 # Generated by scripts/ci/test_ratchet.py --emit-baseline.
 # Do not hand-edit: regenerate from a real --no-fail-fast log.
 # Each line is target<TAB>test for a test failing on master.
-# 73 targets ran; 373 failing tests on the Linux runner (pruned by #2389, #2391, #2395).
+# 73 targets ran; 372 failing tests on the Linux runner (pruned by #2389, #2391, #2395, #2399).
 # Four macOS-only failures pruned by #2387: the baseline describes the
 # platform that gates, and a CI-passing test left in it is invisible if it breaks.
 src/main.rs	compiler::tests_compiler_rejects::lowers_only_first_of_two_modules_characterization
@@ -375,5 +375,4 @@ tests/icarus_lowerable.rs	rejects_w534_negative_witnesses
 tests/icarus_lowerable.rs	rejects_w537_undefined_struct_witness
 tests/icarus_lowerable.rs	rejects_w543_nonlowerable_call_init_witness
 tests/icarus_lowerable.rs	rejects_w561_nonlowerable_struct_return_witnesses
-tests/verilog_array_literal_expr.rs	r_ca_2_synthetic_no_comment_only_call_argument
 tests/verilog_array_param_index.rs	array_param_index_is_element_part_select