docs: bring the 5.10.5 benchmark to main

README.md

@@ -240,20 +240,23 @@ sequenceDiagram
 
 Peak memory and per-operation latency against docker-compose and podman-compose,
 **all three driving the same rootless Podman**, same digest-pinned images,
-median of 10 measured runs (12 iterations, 2 warm-up discarded), on podup 5.7.1.
-podup is fastest in all 29 measured rows, though three teardown rows
-(`deep-chain`, `many-services` and `multi-healthcheck`) win by less than their
-own standard deviation and should be read as ties; the widest gaps are the ones
-with many services.
+median of 10 measured runs (12 iterations, 2 warm-up discarded), on podup 5.10.5.
+podup is fastest in 25 of 29 rows. Of the other four, two are ties with
+docker-compose, docker-compose leads `secrets up` by 17 ms, and podman-compose
+tears down five replicas faster (`scale down`, 0.42 s against 0.69 s). The widest gaps are the ones with many services.
 
 | | podup | docker-compose | podman-compose |
 |---|---|---|---|
-| memory per command | **5.7 MiB** | 29.0 MiB | 45.9 MiB |
-| `up`, 42 services | **1.11 s** | 1.58 s | 6.65 s |
-| `up`, 12 services | **0.37 s** | 0.47 s | 1.89 s |
-| `config` (parse only) | **13.0 ms** | 38.3 ms | 112.7 ms |
+| memory per command | **7.2 MiB** | 29.7 MiB | 51.5 MiB |
+| `up`, 42 services | **2.83 s** | 4.72 s | 8.51 s |
+| `up`, 12 services | **0.96 s** | 1.36 s | 2.56 s |
+| `config` (parse only) | **12.0 ms** | 33.8 ms | 538.3 ms |
 
-<img src="docs/assets/bench.svg" alt="Bar chart: podup uses about 6 MiB per command against 29 MiB for docker-compose and 46 MiB for podman-compose, and is faster in every measured scenario" width="760">
+On its own Docker engine, docker-compose wins the teardown of many containers and
+`exec`; that comparison, and why these latencies are not comparable with older
+releases' tables, is in the full page.
+
+<img src="docs/assets/bench.svg" alt="Bar chart: podup uses about 7 MiB per command against 30 MiB for docker-compose and 52 MiB for podman-compose, and is the fastest of the three on the three latency rows shown" width="760">
 
 Full tables and methodology: [docs/benchmarks.md](docs/benchmarks.md).
 

bench/chart.py

@@ -24,7 +24,13 @@
     ("up · 12 services", "many-services", "up", "seconds"),
     ("config · parse only", "config-heavy", "config", "seconds"),
 ]
-TOOLS = [("podup", "#3fb950"), ("docker-compose", "#58a6ff"), ("podman-compose", "#f778ba")]
+# (key in summary.json, label, colour). The chart shows the same-engine
+# comparison, so docker-compose is the variant pointed at the Podman socket.
+TOOLS = [
+    ("podup", "podup", "#3fb950"),
+    ("docker-compose-podman", "docker-compose", "#58a6ff"),
+    ("podman-compose", "podman-compose", "#f778ba"),
+]
 
 
 def value(summary, tool, scen, op, key):
@@ -43,7 +49,7 @@ def main() -> int:
 
     groups = []
     for label, scen, op, key in ROWS:
-        vals = [(t, value(summary, t, scen, op, key), c) for t, c in TOOLS]
+        vals = [(name, value(summary, t, scen, op, key), c) for t, name, c in TOOLS]
         if any(v is None for _, v, _ in vals):
             print(f"missing data for {scen}/{op}; skipping", file=sys.stderr)
             continue
@@ -60,8 +66,9 @@ def main() -> int:
 
     out = [
         f'<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 {width} {height}" '
-        f'width="{width}" role="img" aria-label="podup benchmark chart: podup leads every '
-        f'measured scenario in both memory and latency">',
+        f'width="{width}" role="img" aria-label="podup benchmark chart: memory per command '
+        f'and three latency rows, podup against docker-compose and podman-compose on the '
+        f'same rootless Podman">',
         "  <style>",
         "    .card { fill:#0d1117; stroke:#30363d; }",
         "    .t { fill:#e6edf3; font:600 13px ui-sans-serif,system-ui,sans-serif; }",