fix(perry-ffi): unbreak cargo test -p perry-ffi (undefined js_register_ffi_handle_exists_probe)

#5083 added a `js_register_ffi_handle_exists_probe(...)` call inside
`register_handle` (handle.rs) for the runtime's handle-vs-timer disambiguation
probe. That symbol is defined in perry-runtime, but perry-ffi's `runtime-link`
feature is off by default and CI runs `cargo test -p perry-ffi` per-package in
isolation (no --workspace feature unification). The handle-registry tests
(bare `#[cfg(test)]`) call `register_handle`, so the test binary retained a
reference to an unlinked symbol and failed at `cc` with `undefined reference`.

Gating the handle tests on `runtime-link` (the pattern other perry-ffi test
modules use) would drop them from CI entirely (the per-package loop never
enables the feature). Instead add a test-only no-op stub gated
`#[cfg(all(test, not(feature = "runtime-link")))]` so the registry tests keep
running and the binary links; the real extern declaration is gated out of that
exact config to avoid an E0428 name clash. The stub can never collide with
perry-runtime's real definition (it exists only in perry-ffi's own test binary
with runtime-link off).

Verified: `cargo test -p perry-ffi` links + passes (13); `cargo test -p
perry-ffi --features runtime-link` uses the real symbol + passes (35).

Author: Ralph Küpper

CHANGELOG.md

@@ -1,3 +1,33 @@
+## v0.5.1164 — fix(perry-ffi): unbreak `cargo test -p perry-ffi` (undefined `js_register_ffi_handle_exists_probe`)
+
+The `cargo-test` CI gate was red on `main`: #5083 added a
+`js_register_ffi_handle_exists_probe(...)` call inside `register_handle`
+(`crates/perry-ffi/src/handle.rs`) to wire up the runtime's handle-vs-timer
+disambiguation probe. That symbol is defined in `perry-runtime`, but
+`perry-ffi`'s `runtime-link` feature (which pulls `perry-runtime` into the
+Cargo graph) is **off** by default, and CI runs `cargo test -p perry-ffi`
+per-package in isolation — no `--workspace` feature unification. So the
+handle-registry tests (a bare `#[cfg(test)]` module that calls
+`register_handle`) retained a reference to a symbol that wasn't linked, and the
+test binary failed at `cc` with `undefined reference to
+js_register_ffi_handle_exists_probe`.
+
+All other `perry-ffi` test modules avoid this by gating on
+`#[cfg(all(test, feature = "runtime-link"))]`, but gating the handle tests the
+same way would drop them from CI entirely (the per-package loop never enables
+the feature). Instead, this adds a **test-only no-op stub** for the symbol,
+gated `#[cfg(all(test, not(feature = "runtime-link")))]`, so the registry tests
+keep running and the binary links. The real `extern` declaration is gated out
+of that exact configuration to avoid an `E0428` name clash. The stub can never
+collide with `perry-runtime`'s real definition: it exists only in `perry-ffi`'s
+own `test` binary with `runtime-link` off — not in `runtime-link`-on builds
+(real symbol used) nor in the lib builds that npm wrappers link against
+`libperry_runtime.a`.
+
+Verified: `cargo test -p perry-ffi` now links and passes (13 tests); `cargo test
+-p perry-ffi --features runtime-link` still uses the real perry-runtime symbol
+and passes (35 tests), no collision.
+
 ## v0.5.1163 — fix: chalk boolean style modifiers — `<Text dimColor>` no longer renders `[object Object]` (#5039)
 
 Fixes #5039 — ink's `<Text dimColor>` (and `bold`/`italic`/`underline`/…)

CLAUDE.md

@@ -8,7 +8,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 Perry is a native TypeScript compiler written in Rust that compiles TypeScript source code directly to native executables. It uses SWC for TypeScript parsing and LLVM for code generation.
 
-**Current Version:** 0.5.1163
+**Current Version:** 0.5.1164
 
 
 ## TypeScript Parity Status