gen-zig: an @extern body is written as @extern("sym"), which Zig refuses, so a host spec never runs its tests under t27c #11282

What

t27c gen writes a fn whose body is return @extern("getpid"); into Zig verbatim. Zig's @extern takes two arguments (a pointer type and an options struct), so the generated file does not compile, and t27c test-report reports the whole spec BLOCKED, its pure tests included.

Repro (t27c 0.5.2, master 1679c55)

module ExternRepro;

fn host_getpid() -> i32 { return @extern("getpid"); }

fn twice(x: i32) -> i32 { return x * 2; }

test twice_doubles {
    assert twice(2) == 4;
}

t27c test-report specs/repro/extern_repro.t27 --specs-dir specs prints BLOCKED does not compile: spec.zig:19:12: error: expected 2 arguments, found 1; line 19 of the generated Zig is return @extern("getpid");.

Where it bites

• Four T27 OS host specs are BLOCKED under t27c on master: specs/t27os/glass_window.t27 (34 test or invariant blocks), keys_host.t27, keys_probe.t27, pty_server.t27. Their tests run only under t27b, which lowers @extern itself (see t27b: lower @extern calls to libSystem (blocks the T27 OS PTY server) #10902).
• spec-guards scores a BLOCKED spec as 1000000 failing tests. A new host spec has no base copy, so it is scored against 0 and the job turns red: PR T27 OS: the PTY server reads the keys itself through /usr/bin/security at start and when a spawn finds none, so every pane gets a key without the window #11278 (which moved the CoreFoundation probe out of keys_host.t27 into the new keys_probe.t27) merged with spec-guards red for this reason alone. The three older host specs pass the guard only because their base copy is BLOCKED too.

What done looks like

• gen-zig lowers a fn whose body is a single @extern("sym") to a Zig extern declaration with the fn's own parameter and return types, called from the fn, so the module compiles and its pure tests run. Zig links an extern only when a reached call needs it.
• The repro above reports twice_doubles passed, vacuous passes 0.
• glass_window.t27, keys_host.t27, keys_probe.t27 and pty_server.t27 are no longer BLOCKED under t27c test-report on the Linux runner (a test that calls a Darwin-only extern would still fail to link there, which is the right report).

Boundary

The gen-zig emitter in bootstrap/src/compiler.rs (frozen core, the FROZEN_HASH rule applies) or its .t27 port. Related: #10295 (host fn declarations for gen-js and gen-ts), #10902 (t27b lowering of @extern), compiler defects tracked on #5980.

🤖 Generated with Claude Code