bench(interpreter): add fork_for_serving benchmark

.github/workflows/ci.yml

@@ -36,6 +36,15 @@ jobs:
       # Broad suppressions (-A dead-code, -A unused-variables, etc.) removed.
       - run: cargo clippy --all-targets -- -A clippy::approx_constant -A clippy::result_large_err -A clippy::only_used_in_recursion -A clippy::len_zero
 
+  fork-benchmark:
+    name: fork_for_serving benchmark
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@stable
+      - uses: Swatinem/rust-cache@v2
+      - run: cargo bench --bench fork_for_serving -- --warm-up-time 3 --measurement-time 5 --sample-size 50
+
   fmt:
     name: Format
     runs-on: ubuntu-latest
@@ -78,7 +87,7 @@ jobs:
       - name: Run Forge test suite
         run: |
           echo "Running all .fg tests in tests/ directory..."
-          ./target/release/forge test tests/ 2>&1 | tee /tmp/forge-test-out.txt
+          ./target/release/forge --allow-run test tests/ 2>&1 | tee /tmp/forge-test-out.txt
           # Check for any failures in the summary line
           if grep -qE "[1-9][0-9]* failed" /tmp/forge-test-out.txt; then
             echo "FAILURE: Some Forge tests failed"

.planning/fork-for-serving-benchmark.plan.md

@@ -0,0 +1,35 @@
+# Fork For Serving Benchmark Plan
+
+## Goal
+
+Close issue #112 by adding a real Criterion benchmark for `Interpreter::fork_for_serving()` and wiring it into CI for visibility.
+
+## Current State
+
+- There is no `benches/` directory or Criterion setup.
+- `src/interpreter/tests.rs` has a diagnostic `fork_for_serving_is_under_50ms` test, but its threshold is intentionally loose and not a useful performance signal.
+- `fork_for_serving()` is public and benchmarkable from an external bench target through the `forge_lang` library crate.
+
+## Implementation
+
+1. Add `criterion` as a dev-dependency using Cargo.
+2. Add `[[bench]] name = "fork_for_serving" harness = false` to `Cargo.toml`.
+3. Add an explicit `[profile.bench] opt-level = 3` so benchmark builds are release-like even if future profile edits change defaults.
+4. Add `benches/fork_for_serving.rs`:
+   - build an `empty` fixture from `Interpreter::new()` to show the lower bound,
+   - build a `with_closures` fixture by lexing/parsing/running a small representative program that creates top-level objects, arrays, functions, and captured lambdas,
+   - benchmark `criterion::black_box(interp.fork_for_serving())` for each fixture.
+5. Add a lightweight CI job on Ubuntu that runs:
+   - `cargo bench --bench fork_for_serving -- --warm-up-time 3 --measurement-time 5 --sample-size 50`
+6. Keep the CI job visibility-only: do not parse timing or fail on noisy performance thresholds. Compilation and successful benchmark execution are the gate.
+7. Remove the existing diagnostic wall-clock test in `src/interpreter/tests.rs`; Criterion replaces it with a statistically meaningful benchmark.
+
+## Tests
+
+- `cargo fmt`
+- `cargo bench --bench fork_for_serving -- --warm-up-time 3 --measurement-time 5 --sample-size 50`
+- `cargo test`
+
+## Rollback
+
+Remove the bench target, Criterion dev-dependency, CI job, and this plan file.