perf(runner): create cold-start scratch dirs in one mkdir fork

CHANGELOG.md

@@ -13,6 +13,7 @@
 - `--jobs auto` / `-j auto` caps parallel concurrency at the CPU core count (portable across Linux/macOS/BSD); the default stays unlimited (#766)
 
 ### Changed
+- Faster cold start: the run-output scratch directory and the shared temp directory are created with a single `mkdir -p` call instead of one fork each (2 -> 1 `mkdir` forks per cold start). No behaviour change
 - Faster test runs: the runner detects a failed source (syntax error / unexpected EOF) and discovery detects the `.bash` pattern variant with shell `case` matches instead of `grep` forks (2 -> 0 `grep` forks per sourced test file). No behaviour change
 - Faster cold start: the base64 `-w` capability is detected with a shell `case` on `base64 --help` instead of piping into a `grep` fork (1 -> 0 `grep` forks per cold start). Same detection, no behaviour change
 - Faster cold start: selecting the `perl` clock reads the time once instead of an empty `Time::HiRes` probe followed by a separate first read (2 -> 1 `perl` fork per cold start on bash 3.2 / macOS). No behaviour change

src/env.sh

@@ -494,7 +494,6 @@ MKTEMP="$(command -v mktemp)"
 # not be pre-created. The random suffix keeps the directory unique across
 # recursive and parallel invocations, matching TEMP_DIR_PARALLEL_TEST_SUITE.
 _BASHUNIT_RUN_OUTPUT_DIR="${TMPDIR:-/tmp}/bashunit/run/${_BASHUNIT_OS:-Unknown}/$(bashunit::random_str 8)"
-mkdir -p "$_BASHUNIT_RUN_OUTPUT_DIR" 2>/dev/null || true
 FAILURES_OUTPUT_PATH="$_BASHUNIT_RUN_OUTPUT_DIR/failures"
 SKIPPED_OUTPUT_PATH="$_BASHUNIT_RUN_OUTPUT_DIR/skipped"
 INCOMPLETE_OUTPUT_PATH="$_BASHUNIT_RUN_OUTPUT_DIR/incomplete"
@@ -504,9 +503,11 @@ PROFILE_OUTPUT_PATH="$_BASHUNIT_RUN_OUTPUT_DIR/profile"
 # next --rerun-failed can replay just those. Shared across parallel subshells.
 RERUN_FAILED_OUTPUT_PATH="$_BASHUNIT_RUN_OUTPUT_DIR/rerun-failed"
 
-# Initialize temp directory once at startup for performance
+# Shared temp directory, initialized once at startup for performance.
 BASHUNIT_TEMP_DIR="${TMPDIR:-/tmp}/bashunit/tmp"
-mkdir -p "$BASHUNIT_TEMP_DIR" 2>/dev/null || true
+
+# Create both scratch directories in a single `mkdir -p` fork.
+mkdir -p "$_BASHUNIT_RUN_OUTPUT_DIR" "$BASHUNIT_TEMP_DIR" 2>/dev/null || true
 
 if bashunit::env::is_dev_mode_enabled; then
   bashunit::internal_log "info" "Dev log enabled" "file:$BASHUNIT_DEV_LOG"

tests/acceptance/bashunit_coldstart_forks_test.sh

@@ -65,3 +65,20 @@ function test_coldstart_does_not_fork_grep_to_detect_base64_wrap() {
 
   assert_equals 0 "$grep_forks"
 }
+
+# Regression guard: cold start creates its two scratch directories (the run
+# output dir and the shared temp dir) with a single `mkdir -p` call instead of
+# one fork each.
+function test_coldstart_creates_scratch_dirs_with_one_mkdir() {
+  if bashunit::check_os::is_windows; then
+    bashunit::skip "process tracing is unreliable under Git Bash" && return
+  fi
+
+  local trace
+  trace="$(PS4='+ ' bash -x ./bashunit --version 2>&1 >/dev/null)"
+
+  local mkdir_forks
+  mkdir_forks="$(printf '%s\n' "$trace" | grep -cE '^\++ +/?[a-z/]*mkdir ' || true)"
+
+  assert_less_or_equal_than 1 "$mkdir_forks"
+}