ci: clamp the duration accumulator and truncate section bodies structurally

Round-6 review fixes:

- test.yml: the serialised duration is clamped too — per-case isfinite keeps each addend and the slowest list finite, but a sum of finite doubles (two 1e308s) still overflows to inf, and round(inf,1) would emit bare Infinity into the JSON. Failed test-case entries with neither fullname nor name fall back to "(unnamed)" so sorted(set(...)) cannot hit a None/str TypeError.
- upsert-ci-status.sh: the 20k cap now guards both body paths (env and file), and truncation closes any code fence or <details> the cut severed — an unterminated construct would render the rest of the comment inside it, visually eating the neighbouring sections.

Committed via API because repository rules require verified signatures.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: eordano

.github/actions/ci-status-comment/upsert-ci-status.sh

@@ -33,14 +33,28 @@ set -euo pipefail
 #                       find it, spawning duplicates).
 if [ -n "${SECTION_BODY_FILE:-}" ]; then
   SECTION_BODY="$(cat "$SECTION_BODY_FILE")"
-  # GitHub caps an issue comment at 65536 chars across every section; keep one
-  # writer from consuming the whole budget and failing an unrelated section's
-  # PATCH with an opaque 422. Truncation is fine for a status section that
-  # already links out to the full report.
-  if [ "${#SECTION_BODY}" -gt 20000 ]; then
-    echo "::warning::Section body is ${#SECTION_BODY} chars; truncating to 20000."
-    SECTION_BODY="${SECTION_BODY:0:20000}"$'\n\n'"_…truncated; see the linked run for the full report._"
+fi
+
+# GitHub caps an issue comment at 65536 chars across every section; keep one
+# writer — whichever path its body arrived by — from consuming the whole budget
+# and failing an unrelated section's PATCH with an opaque 422. Truncation is
+# fine for a status section that already links out to the full report.
+if [ "${#SECTION_BODY}" -gt 20000 ]; then
+  echo "::warning::Section body is ${#SECTION_BODY} chars; truncating to 20000."
+  SECTION_BODY="${SECTION_BODY:0:20000}"
+  # Close constructs the cut may have severed — an unterminated code fence or
+  # <details> makes GitHub render everything after it in this comment inside
+  # the open block, visually eating the neighbouring sections.
+  if [ $(( $(grep -c '^```' <<< "$SECTION_BODY") % 2 )) -ne 0 ]; then
+    SECTION_BODY="$SECTION_BODY"$'\n''```'
   fi
+  opens=$(grep -oi '<details' <<< "$SECTION_BODY" | wc -l || true)
+  closes=$(grep -oi '</details' <<< "$SECTION_BODY" | wc -l || true)
+  while [ "${opens:-0}" -gt "${closes:-0}" ]; do
+    SECTION_BODY="$SECTION_BODY"$'\n</details>'
+    closes=$((closes + 1))
+  done
+  SECTION_BODY="$SECTION_BODY"$'\n\n'"_…truncated; see the linked run for the full report._"
 fi
 
 # Fail fast on a section name outside the fence set — an unknown name would

.github/workflows/test.yml

@@ -841,14 +841,19 @@ jobs:
                   if case_result == "Passed":
                       passed += 1
                   elif case_result == "Failed":
-                      failed.append(test_case.get("fullname") or test_case.get("name"))
+                      # "(unnamed)" keeps the set homogeneous — one None among
+                      # strings makes sorted() raise and kills the whole file.
+                      failed.append(test_case.get("fullname") or test_case.get("name") or "(unnamed)")
 
           result = {
               "hasResults": len(xml_files) > 0,
               "total": total,
               "passed": passed,
               "failed": sorted(set(failed)),
-              "duration": round(duration, 1),
+              # The per-case clamp keeps each addend (and the slowest list)
+              # finite, but a sum of finite doubles can still overflow to inf —
+              # clamp again at the one point the accumulator is serialised.
+              "duration": round(duration, 1) if math.isfinite(duration) else 0.0,
               "slowest": [
                   {"name": name, "seconds": round(seconds, 1)}
                   for seconds, name in sorted(timings, key=lambda t: -t[0])[:10]