t27c gen (Zig): test bindings shadow module vars, redeclare locals, and inherit param renames #6295

Refs #6063

Found while lowering module-level vars in t27b (#6294). The Zig backend of t27c gen emits code that does not compile for three test-block shapes. t27c test-report therefore reports these specs as BLOCKED.

1. A top-level assignment to a module var in a test becomes a shadowing const

specs/fpga/testbench/spi_tb.t27, test test_cs_assert_during_transfer:

spi_start = true;          // spi_start is a module-level var

This is emitted as var spi_start = true;, and Zig refuses it: "local variable shadows declaration of 'spi_start'".

The test-binding pass (compiler.rs, the fresh_binding loop in the Zig test emitter) declares the first assignment to any identifier. It does not check whether the name is a module declaration.

Affected: apb_bridge_tb, bridge_tb, dft_tb, formal_tb, integration_tb, spi_tb, vcd_trace_tb (specs/fpga/testbench/).

2. A local declared earlier in the test is declared again

specs/fpga/testbench/hir_tb.t27, test test_depth_limit:

var ok : bool = check_depth_limit(10);
ok = check_depth_limit(20);

This is emitted as var ok ...; const ok = ...;, and Zig refuses it: "redeclaration of local variable 'ok'".

The Zig bound set is not seeded with the block's StmtLocal names. The C path fixed the same defect in #2834.

Affected: hir_tb, axi4_tb, apb_bridge_tb.

3. Parameter renames leak into the following test

var counter: u32 = 0;
fn f(counter: u32) -> u32 { return counter + 1; }
test t { assert(counter == 0); }

counter in the test is emitted as counter_arg ("use of undeclared identifier"). param_renames (W735/W736) is cleared only at the start of the next fn, so it is still in force for every test, bench or invariant emitted after f. This also affects module consts and fns, not only vars.

t27b refuses all three shapes with named blockers, so there is no mismatch:

• StmtAssign(module var in test)
• StmtLocal(shadows module var)
• ExprIdentifier(renamed module var)

For shape 2 (a plain local), t27b lowers the reassignment as a reassignment. That is the meaning of the spec, so hir_tb runs in t27b while the reference is BLOCKED. With the reassignment written as a fresh local, both report 7 pass and 1 fail (test_tick).

Boundary

• specs/fpga/testbench/hir_tb.t27
• specs/fpga/testbench/spi_tb.t27