parallel-make-check: warn when a job runtime drifts >50% from "minutes"

The "minutes" field is only a scheduling estimate; when it goes stale it
just packs the schedule a little worse, and there was no signal that a
value needed updating. Emit a non-fatal warning when a config that
explicitly sets "minutes" finishes more than 50% above or below it (a
GitHub ::warning:: annotation in CI, a plain line locally) and flag the
row in the step-summary table with the value to copy over.

Configs that omit "minutes" keep riding the 1.0 default and are left
alone. The warning never touches the exit status, so it cannot fail the
job.

Author: julek-wolfssl

.github/scripts/parallel-make-check.py

@@ -19,7 +19,9 @@
 #   minutes    expected duration, from the Minutes column of a previous
 #              run's summary (default 1.0). Schedule weight only - configs
 #              run longest-first and --shard balances shards by it; a stale
-#              value just packs the schedule a little worse.
+#              value just packs the schedule a little worse, but one that
+#              drifts past +/-50% of the measured time draws a warning
+#              (never a failure) so it is easy to spot and update.
 #   user_settings  header staged as <builddir>/user_settings.h before
 #              configure (path relative to the source root); pair it with
 #              --enable-usersettings in "configure"
@@ -88,6 +90,9 @@ class Config:
     check: bool = True
     prepare: list[list[str]] = field(default_factory=list)
     run: list[list[str]] = field(default_factory=list)
+    # Whether "minutes" was given in the JSON (vs the 1.0 default); only an
+    # explicit estimate is checked for >50% drift against the real time.
+    minutes_provided: bool = False
 
 SRCDIR = Path(__file__).resolve().parents[2]
 ON_GITHUB = os.environ.get("GITHUB_ACTIONS") == "true"
@@ -208,7 +213,8 @@ def load_configs(opts: argparse.Namespace,
                               entry.get("ldflags", opts.ldflags),
                               float(minutes), user_settings, check,
                               list(entry.get("prepare", [])),
-                              list(entry.get("run", []))))
+                              list(entry.get("run", [])),
+                              minutes_provided="minutes" in entry))
     if not configs:
         error(f"{opts.json}: no configs")
     return configs
@@ -238,6 +244,23 @@ def dump(title: str, path: Path) -> None:
     sys.stdout.flush()
 
 
+def warn(msg: str) -> None:
+    # GitHub surfaces ::warning:: as an annotation at the top of the run;
+    # locally it is just a line. Informational only - never fails the run.
+    print(f"::warning::{msg}" if ON_GITHUB else f"WARNING: {msg}")
+
+
+def stale_estimate(cfg: Config, minutes: float) -> bool:
+    # "minutes" is only a scheduling estimate (configs run longest-first;
+    # --shard balances by it), never a pass/fail bound. Flag a finished
+    # config whose real time drifted past +/-50% of an explicitly given
+    # estimate so stale values - which pack the schedule worse - are easy
+    # to find and update. Configs that omit "minutes" ride the 1.0 default
+    # placeholder and are left alone.
+    return (cfg.minutes_provided
+            and not 0.5 * cfg.minutes <= minutes <= 1.5 * cfg.minutes)
+
+
 def run_config(cfg: Config, opts: argparse.Namespace) -> tuple[str | None,
                                                                float]:
     if opts.fail_fast and stop_event.is_set():
@@ -344,6 +367,10 @@ def record_failure(step: str) -> str:
             # One line per passing config; the full logs would bloat the CI
             # log (they stay in build-<name>/make-check.log).
             print(f"{cfg.name}: pass [{minutes:.1f} min]")
+            if stale_estimate(cfg, minutes):
+                warn(f"{cfg.name}: ran {minutes:.1f} min but \"minutes\" "
+                     f"says {cfg.minutes:g} (>50% off) - update it in the "
+                     f"config JSON")
             sys.stdout.flush()
         else:
             dump(f"{cfg.name}: FAIL ({failed}) [{minutes:.1f} min]", log)
@@ -364,6 +391,11 @@ def summarize(results: list[tuple[Config, str | None, float]],
             ok = f":x: FAIL ({failed})"
         else:
             ok = ":white_check_mark: pass"
+            if stale_estimate(cfg, minutes):
+                # Non-fatal nudge mirroring the per-config warning, kept in
+                # the summary next to the Minutes value to copy over.
+                ok += (f' :warning: "minutes" {cfg.minutes:g} is >50% off, '
+                       f"update to ~{minutes:.1f}")
         lines.append(f"| {cfg.name} | {ok} | {minutes:.1f} |")
     # Two views of how efficiently the pool used the machine: thread
     # occupancy is the time the workers spent running configs out of the