feat: bench.py --outer kick mode for kick-in-kick benchmarks

bench.py

@@ -7,10 +7,16 @@ use) and reports timing + resource numbers as a human-readable table,
 optionally also as JSON.
 
 Usage:
-    ./bench.py [--mode auto|nested|standalone] [--repeat N] [--cycles N]
+    ./bench.py [--mode auto|nested|standalone] [--outer host|kick]
+               [--repeat N] [--cycles N]
                [--tabs N] [--filter SUBSTR] [--list] [--json OUT.json]
                [--quick]
 
+--outer selects what the measured nested compositor nests into: the
+ambient host (labwc etc.) or a standalone-headless kick booted by the
+script (kick-in-kick via _spawn_nested_in). Only valid with
+--mode nested.
+
 Benchmarks (all wall-clock unless noted):
     startup_1tab_ms   spawn -> first tab mapped+active (1 wltest tab)
     startup_2tab_ms   spawn -> both tabs mapped (2 wltest tabs)
@@ -56,6 +62,41 @@ class BenchError(RuntimeError):
     pass
 
 
+# Shared standalone-headless outer for --outer kick (kick-in-kick):
+# booted once per run, kept alive across benchmarks; each measured
+# session is a fresh inner nested into it. Closed at the end of main().
+_OUTER = None
+
+
+def _ensure_outer():
+    """Boot the shared standalone outer (with one keeper tab) if needed."""
+    global _OUTER
+    if _OUTER is not None and _OUTER.proc.poll() is None:
+        return _OUTER
+    if _OUTER is not None:
+        _OUTER.close()
+        _OUTER = None
+    try:
+        outer = T._spawn_standalone({}, _tabs(1)[0])
+    except BaseException as e:
+        raise BenchError(f"outer spawn failed: {e}") from e
+    try:
+        _wait_for(outer, r"active app=\d+", 30.0)
+        _wait_for(outer, re.escape(MAP_RE + "0"), 30.0)
+    except (BenchError, TimeoutError) as e:
+        outer.close()
+        raise BenchError(f"outer never settled: {e}") from e
+    _OUTER = outer
+    return _OUTER
+
+
+def _close_outer():
+    global _OUTER
+    if _OUTER is not None:
+        _OUTER.close()
+        _OUTER = None
+
+
 def _tabs(n):
     return ["%s '%s'" % (T.WL, COLORS[i % len(COLORS)]) for i in range(n)]
 
@@ -74,9 +115,11 @@ def _resolve_mode(mode):
     return mode
 
 
-def _spawn(mode, tabs):
+def _spawn(mode, args, tabs):
     try:
         if mode == "nested":
+            if args.outer == "kick":
+                return T._spawn_nested_in(_ensure_outer(), *tabs)
             return T._nested_session(*tabs)
         return T._spawn_standalone({}, *tabs)
     except BaseException as e:  # pytest.skip/fail raise BaseException subclasses
@@ -140,7 +183,7 @@ def _assert_alive(s):
 def bench_startup_1tab(mode, args):
     tabs = _tabs(1)
     t0 = time.perf_counter()
-    s = _spawn(mode, tabs)
+    s = _spawn(mode, args, tabs)
     try:
         _wait_for(s, r"active app=0\b", 30.0)
         _wait_for(s, re.escape(MAP_RE + "0"), 30.0)
@@ -154,7 +197,7 @@ def bench_startup_1tab(mode, args):
 def bench_startup_2tab(mode, args):
     tabs = _tabs(2)
     t0 = time.perf_counter()
-    s = _spawn(mode, tabs)
+    s = _spawn(mode, args, tabs)
     try:
         _wait_for(s, r"active app=\d+", 30.0)
         _wait_for(s, re.escape(MAP_RE + "0"), 30.0)
@@ -170,7 +213,7 @@ def bench_startup_8tab(mode, args):
     n = args.tabs
     tabs = _tabs(n)
     t0 = time.perf_counter()
-    s = _spawn(mode, tabs)
+    s = _spawn(mode, args, tabs)
     try:
         _wait_for(s, r"active app=\d+", 30.0)
         for i in range(n):
@@ -182,8 +225,8 @@ def bench_startup_8tab(mode, args):
         s.close()
 
 
-def _boot_2tab_settled(mode):
+def _boot_2tab_settled(mode, args):
-    s = _spawn(mode, _tabs(2))
+    s = _spawn(mode, args, _tabs(2))
     _wait_for(s, r"active app=\d+", 30.0)
     _wait_for(s, re.escape(MAP_RE + "0"), 30.0)
     _wait_for(s, re.escape(MAP_RE + "1"), 30.0)
@@ -193,7 +236,7 @@ def _boot_2tab_settled(mode):
 
 def bench_tab_cycle(mode, args):
     samples = []
-    s = _boot_2tab_settled(mode)
+    s = _boot_2tab_settled(mode, args)
     try:
         for i in range(args.cycles):
             cur_m = ACTIVE_RE.findall(s.read_log())
@@ -211,7 +254,7 @@ def bench_tab_cycle(mode, args):
 
 def bench_input_ack(mode, args):
     samples = []
-    s = _boot_2tab_settled(mode)
+    s = _boot_2tab_settled(mode, args)
     try:
         for _ in range(args.cycles):
             pos = len(s.read_log())
@@ -228,7 +271,7 @@ def bench_input_ack(mode, args):
 
 def bench_rapid_cycle(mode, args):
     n = 20
-    s = _boot_2tab_settled(mode)
+    s = _boot_2tab_settled(mode, args)
     try:
         cur_m = ACTIVE_RE.findall(s.read_log())
         cur = int(cur_m[-1]) if cur_m else 1
@@ -247,7 +290,7 @@ def bench_rapid_cycle(mode, args):
 
 def bench_frame_dump(mode, args):
     samples = []
-    s = _boot_2tab_settled(mode)
+    s = _boot_2tab_settled(mode, args)
     try:
         for _ in range(3):
             t0 = time.perf_counter()
@@ -263,7 +306,7 @@ def bench_frame_dump(mode, args):
 
 
 def bench_rss_2tab(mode, args):
-    s = _spawn(mode, _tabs(2))
+    s = _spawn(mode, args, _tabs(2))
     try:
         _wait_for(s, r"active app=\d+", 30.0)
         _wait_for(s, re.escape(MAP_RE + "0"), 30.0)
@@ -279,7 +322,7 @@ def bench_rss_2tab(mode, args):
 
 def bench_rss_8tab(mode, args):
     n = args.tabs
-    s = _spawn(mode, _tabs(n))
+    s = _spawn(mode, args, _tabs(n))
     try:
         _wait_for(s, r"active app=\d+", 30.0)
         for i in range(n):
@@ -329,6 +372,10 @@ def main(argv=None):
     ap.add_argument("--mode", default="auto",
                     choices=["auto", "nested", "standalone"],
                     help="compositor mode (auto: nested with a host, else headless standalone)")
+    ap.add_argument("--outer", default="host",
+                    choices=["host", "kick"],
+                    help="what a nested session nests into: the ambient host "
+                         "or a standalone-headless kick (kick-in-kick)")
     ap.add_argument("--repeat", type=int, default=3,
                     help="outer repeats for boot/static benchmarks (default 3)")
     ap.add_argument("--cycles", type=int, default=10,
@@ -356,20 +403,23 @@ def main(argv=None):
 
     _check_binaries()
     mode = _resolve_mode(args.mode)
-    print(f"bench: mode={mode} repeat={args.repeat} "
+    if args.outer == "kick" and mode != "nested":
+        raise BenchError("--outer kick needs --mode nested")
+    print(f"bench: mode={mode} outer={args.outer} repeat={args.repeat} "
           f"cycles={args.cycles} tabs={args.tabs}", flush=True)
 
     results = {}
     failed = 0
+    try:
         for name, unit, desc, fn, kind in BENCHMARKS:
             if args.filter and args.filter not in name:
                 continue
             # Static benchmarks measure once; boot benchmarks repeat per spawn;
             # session benchmarks gather inner samples per spawn, repeated.
-        outer = 1 if kind == "static" else args.repeat
+            reps = 1 if kind == "static" else args.repeat
             samples, extras, err = [], {}, None
             t0 = time.perf_counter()
-        for _ in range(outer):
+            for _ in range(reps):
                 try:
                     s, e = fn(mode, args)
                 except (BenchError, TimeoutError) as ex:
@@ -392,6 +442,8 @@ def main(argv=None):
             print(f"  {name:20s} mean={summary['mean']:10.2f} {unit:9s} "
                   f"(median={summary['median']:.2f} min={summary['min']:.2f} "
                   f"max={summary['max']:.2f} n={summary['runs']})", flush=True)
+    finally:
+        _close_outer()
 
     # Human table.
     print("\n%-20s %5s %12s %12s %12s %12s %-9s" %
@@ -409,7 +461,8 @@ def main(argv=None):
                    r["min"], r["max"], unit))
 
     if args.json:
-        payload = {"mode": mode, "repeat": args.repeat,
+        payload = {"mode": mode, "outer": args.outer,
+                   "repeat": args.repeat,
                    "cycles": args.cycles, "tabs": args.tabs,
                    "results": results}
         with open(args.json, "w") as f: