fix(tests): extend the spawn pin to benches; correct the pin's cost note

benches/alias.rs

@@ -48,7 +48,7 @@ fn wt_cmd(binary: &Path, repo: &Path, user_config: &Path, args: &[&str]) -> Comm
 
 fn bench_dispatch(c: &mut Criterion) {
     let mut group = c.benchmark_group("dispatch");
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
 
     // Startup floor: `wt --version` exits before any repo discovery, so the
     // delta between this and the scaling rows is the parent-side dispatch

benches/completion.rs

@@ -73,7 +73,7 @@ fn assert_completion_candidates(binary: &Path, repo_path: &Path, expected: &BTre
 
 fn bench_completion_switch(c: &mut Criterion) {
     let mut group = c.benchmark_group("completion_switch");
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
     let (linked_worktrees, branchless_branches) = (24, 120);
 
     group.bench_function("full_surface", |b| {

benches/list.rs

@@ -58,7 +58,7 @@ fn run_benchmark(
 
 fn bench_skeleton(c: &mut Criterion) {
     let mut group = c.benchmark_group("skeleton");
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
 
     for total_worktrees in WORKTREE_COUNTS {
         for cache in CacheState::WARM_AND_COLD {
@@ -85,7 +85,7 @@ fn bench_skeleton(c: &mut Criterion) {
 
 fn bench_worktree_scaling(c: &mut Criterion) {
     let mut group = c.benchmark_group("worktree_scaling");
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
 
     for total_worktrees in WORKTREE_COUNTS {
         for cache in CacheState::WARM_AND_COLD {
@@ -120,7 +120,7 @@ fn bench_large_repository(c: &mut Criterion) {
     group.measurement_time(std::time::Duration::from_secs(20));
     group.sample_size(10);
 
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
     let fixture = OnceCell::new();
 
     // Cold goes last because it removes the commit graph. Each warm row has
@@ -153,7 +153,7 @@ fn bench_divergent_branches(c: &mut Criterion) {
     group.measurement_time(std::time::Duration::from_secs(30));
     group.sample_size(10);
 
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
 
     for cache in CacheState::WARM_AND_COLD {
         group.bench_function(cache.label(), |b| {
@@ -218,7 +218,7 @@ fn bench_full(c: &mut Criterion) {
     group.measurement_time(std::time::Duration::from_secs(20));
     group.sample_size(10);
 
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
     let (linked_worktrees, branchless_branches) = (24usize, 120usize);
 
     for cache in CacheState::WARM_AND_COLD {

benches/picker_preview.rs

@@ -45,8 +45,6 @@ fn main() {
 #[cfg(unix)]
 use criterion::{BenchmarkId, Criterion, criterion_group, criterion_main};
 #[cfg(unix)]
-use std::path::Path;
-#[cfg(unix)]
 use wt_perf::{CacheState, FixtureRecipe, bench_wt, wt_command};
 
 #[cfg(unix)]
@@ -57,7 +55,7 @@ fn bench_picker_preview(c: &mut Criterion) {
     group.sample_size(10);
     group.measurement_time(std::time::Duration::from_secs(35));
 
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
     let total_worktrees = 8;
 
     for cache in CacheState::WARM_AND_COLD {

benches/prune.rs

@@ -93,7 +93,7 @@ fn generated_prune_fixture() -> FixtureRepo {
 
 /// Build the `wt <args>` command for `repo`.
 fn wt_cmd(repo: &Path, args: &[&str]) -> Command {
-    let mut cmd = wt_command(Path::new(env!("CARGO_BIN_EXE_wt")), repo, None);
+    let mut cmd = wt_command(&worktrunk::testing::wt_bin(), repo, None);
     cmd.args(args);
     cmd
 }

benches/remove.rs

@@ -194,7 +194,7 @@ fn bench_variant(
     expect_hooks: bool,
     cache: CacheState,
 ) {
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
 
     group.bench_function(name, |b| {
         b.iter_custom(|iterations| {

benches/time_to_first_output.rs

@@ -14,14 +14,13 @@
 //   cargo bench --bench time_to_first_output -- switch  # Just switch
 
 use criterion::{Criterion, criterion_group, criterion_main};
-use std::path::Path;
 use wt_perf::{
     CacheState, FixtureRecipe, bench_wt, run_and_check, standard_benchmark_profile, wt_command,
 };
 
 fn bench_first_output(c: &mut Criterion) {
     let mut group = c.benchmark_group("first_output");
-    let binary = Path::new(env!("CARGO_BIN_EXE_wt"));
+    let binary = &worktrunk::testing::wt_bin();
 
     let fixture = FixtureRecipe::generated(3).create();
 

src/testing/mod.rs

@@ -108,9 +108,14 @@ pub fn wt_bin() -> PathBuf {
 /// by the observed binary's identity, keeping `src`'s basename. The hardlink
 /// pins the inode: cargo's uplift only unlinks the *path*, so the pinned entry
 /// keeps serving the observed binary through any number of concurrent
-/// rebuilds. A link costs no space while the `deps/` artifact it shares an
-/// inode with exists, and everything under `wt-test-bin/` dies with
-/// `cargo clean`, so nothing sweeps it.
+/// rebuilds. An entry's marginal disk cost is near zero — where cargo uplifts
+/// by hardlink (Linux) the pin shares the `deps/` artifact's inode outright,
+/// and where it uplifts by copy-on-write clone (macOS/APFS) the pin keeps the
+/// clone, whose blocks stay shared with that artifact (measured: cloning the
+/// 70 MB binary consumes 8 KB) — the bytes belong to `deps/`, which cargo
+/// already retains. Everything under `wt-test-bin/` dies with `cargo clean`,
+/// so nothing sweeps it; a sweeper could unlink a generation another live
+/// suite pinned, re-creating the very window this exists to close.
 ///
 /// Two runs observing the same binary converge on the same entry (`link`
 /// returning `AlreadyExists` is success — the key names the content); a run

tests/common/mod.rs

@@ -1419,7 +1419,7 @@ mod tests {
         let needle = ["CARGO_BIN_EXE_", "wt\""].concat();
         let root = Path::new(env!("CARGO_MANIFEST_DIR"));
         let mut offenders = Vec::new();
-        for dir in ["src", "tests"] {
+        for dir in ["src", "tests", "benches"] {
             scan_for_needle(&root.join(dir), &needle, root, &mut offenders);
         }
         assert_eq!(