fix(t27c): read a dotted module name and use path whole

[gHashTag]

Summary

module sandbox.health; parsed as the module sandbox followed by a stray top-level StmtExpr .health with line 0, and use std.testing; as the import std followed by .testing. Twelve specs were affected. The t27b corpus run found it.

• Parser (Parser::parse, the module-header skip and use-path loop in parse_module_body):

  • A dotted module header now keeps the whole name verbatim (sandbox.health, port::trinity.src.tri.gen_image). That is the same string extract_module_name already gave the seal tool.
  • A dotted use path is stored as the :: path (sandbox::session_timeout), named after its last segment.
  • A . counts as part of the name only when an identifier follows it. A . inside a body is untouched.

• Names that become identifiers are sanitized:

  • C include guard (gen_c)
  • HIR Verilog module line (AstToHir::convert)
  • gen-testbench DUT (main.rs run_gen_testbench)

  The AST Verilog path already sanitized the name.

• use_resolve::use_targets now splits a dotted path the same way it splits a :: path.

• Zig: a use whose name is also bound as a parameter or local is no longer imported. This is needed because of the fix itself: once use sandbox.session; is named session, it collides with the session parameter in orphan_detection.t27, and Zig rejects a local that shadows a container @import.

Before / after (the 12 specs)

backend	before	after
Zig	header // Generated from t27 spec: sandbox; uses named sandbox/std	header sandbox.health; uses named session_timeout/testing; no shadowed @import
C	#ifndef SANDBOX_H	#ifndef SANDBOX_HEALTH_H
Verilog (AST)	module sandbox ( plus always @(*) begin health; session_timeout; session; end (not Verilog)	module sandbox_health (, no stray statements
Verilog (HIR)	module sandbox (	module sandbox_health (
testbench	DUT sandbox	DUT sandbox_health
Rust	unchanged	unchanged

t27c gen specs/sandbox/health.t27 (the path route, through use_resolve) now pulls Session, SessionStatus and Timestamp from session_timeout.t27. Before, it looked for specs/sandbox.session_timeout.t27 and pulled nothing. orphan_detection.t27 Zig passes zig test 6/6 before and after.

Corpus / seal / suite

• Generated-code hashes: compared master and branch across all 1187 specs/**.t27 and the 113 other .t27 files. Hashes changed for exactly 14 files:
  • the 12 specs above
  • specs/igla/coder/benchmark.t27 and compiler/cli/gen.t27: the Zig guard drops the same shadowed import (dataset, testgen). Both still fail zig test on unrelated undeclared identifiers, as they did before.
• Stray StmtExpr and parse failures:
  • Specs with a line-0 stray StmtExpr: 15 before, 3 after. The 3 left are unrelated.
  • Parse failures: 12 before, 12 after.
• Seals: 20 seal files resealed with t27c seal --save, no --force. The four specs/port specs have no seal.
  • tools/check_seal_currency.py: 0 stale.
  • tools/check_seal_coverage.py: OK (1449 seals).
• Suite: t27c suite --repo-root . reports TOTAL FAILURES 662 before and after, PRIMARY corpus 95 and 95, and an identical JSON summary. This was measured on the pre-rebase base.

Tests

bootstrap/tests/dotted_module_name.rs has 9 cases, including a field-access negative control and a qualified-call control for the import guard. Run against the unfixed binary, all 8 parser and backend cases fail. cargo test --release -p t27c before the rebase: 2816 passed, 0 failed.

Not done

The C guard of a :: module name (137 specs) still contains ::. Fixing it moves 137 seals, so it belongs in its own change.

Closes #5978. Part of #5905.

🤖 Generated with Claude Code