pages: smoke-test what was actually repacked; index shows last-changed commit, source link, admin account

The boot smoke test was hardcoded to always boot dongfanggushi2, which
provides no regression coverage on the common case where that lib wasn't
touched this run. It now boot-checks the libs build_site.sh actually
repacked (all of them up to 10; an even sample of 10 with the skipped
set logged for a full rebuild; a known-good fallback lib on a pure cache
hit).

The index page now shows, per lib: the last commit that changed it
(linked to GitHub, tracked durably in lib-commits.json via the GitHub
commits API since the shallow CI clone has no usable history), a link
to the lib's source directory, and the pre-seeded admin account parsed
from each lib's own README (the AGENTS.md §1.5 convention is
fluffos/Mud@2026, but a few libs use a variant id or have no seeded
account).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VyQCUoTo1Z93Py9aVFHQi1

Author: thefallentree

.github/workflows/pages.yml

@@ -9,10 +9,15 @@
 #      via emscripten's file_packager, config rewritten to the in-memory
 #      mount, per-lib page patched to load the driver from ONE shared copy
 #      at site/_driver/ instead of duplicating ~3.5MB x 90 libs).
-#   3. scripts/gen_site_index.py renders the index (game names from each
-#      lib's README.md, playable/limited/noboot status derived from
-#      the checked-in scripts/wasm_status.json status mapping).
-#   4. Upload site/ as the Pages artifact and deploy.
+#   3. scripts/gen_site_index.py renders the index (game names, admin
+#      credentials from each lib's README.md, playable/limited/noboot
+#      status derived from the checked-in scripts/wasm_status.json status
+#      mapping, per-lib last-changed commit links from lib-commits.json --
+#      maintained by scripts/update_lib_commits.py via the GitHub commits
+#      API, because this shallow clone has no history to derive it from).
+#   4. Boot-check the bundles this run actually repacked (see that step's
+#      comment for the sampling policy), upload site/ as the Pages
+#      artifact, and deploy.
 #
 # INCREMENTAL packing -- two mechanisms with distinct roles:
 #
@@ -147,18 +152,60 @@ jobs:
         id: build
         env:
           LAST_PUBLISHED_SHA: ${{ steps.marker.outputs.sha }}
+          # for update_lib_commits.py's GitHub commits API queries
+          # (per-lib last-changed info on the index cards)
+          GH_TOKEN: ${{ github.token }}
         run: |
           scripts/build_site.sh "$RUNNER_TEMP/wasm-release/rel" \
               "${{ steps.driver.outputs.tag }}" \
               "$RUNNER_TEMP/packed-cache" site
 
-      # Boot one packed bundle inside the WASM driver under node -- proves
-      # the trimmed, packed image + release driver actually start a game
+      # Boot packed bundles inside the WASM driver under node -- proves the
+      # trimmed, packed images + release driver actually start a game
       # (catches over-trimming and packer/driver drift before deploying).
-      - name: Boot smoke test (packed image under node)
+      # Tested set = the libs THIS run actually repacked (a hardcoded lib
+      # proves nothing on the typical run where it was reused from cache):
+      #   - 1..10 repacked: boot-check every one of them.
+      #   - >10 (e.g. full rebuild after a driver/packer bump): boot-check
+      #     an even sample of 10 across the sorted list (first and last
+      #     always included) -- the regressions a full rebuild risks are
+      #     global (driver/packer drift), so any sample catches them, and
+      #     97 serial boots would cost up to ~1.5h of CI.  The skipped
+      #     slugs are logged explicitly (no silent caps).
+      #   - 0 repacked (pure cache reuse): nothing NEW to verify, but boot
+      #     one known-good lib anyway as a cheap sanity check that the
+      #     restored cached bundles still run under the freshly downloaded
+      #     driver (the driver is re-fetched every run even when no lib
+      #     changed).
+      - name: Boot smoke test (repacked images under node)
+        env:
+          REPACKED_SLUGS: ${{ steps.build.outputs.repacked_slugs }}
         run: |
-          node scripts/wasm_boot_check.js site/dongfanggushi2 site/_driver \
-              --timeout 60
+          set -euo pipefail
+          read -r -a all <<< "$REPACKED_SLUGS"
+          if [ "${#all[@]}" -eq 0 ]; then
+            fallback=dongfanggushi2
+            if [ ! -d "site/$fallback" ]; then
+              fallback=$(basename "$(ls -d site/*/ | grep -v /_driver/ | head -1)")
+            fi
+            echo "nothing repacked this run; sanity-booting known-good $fallback"
+            pick=("$fallback")
+          elif [ "${#all[@]}" -le 10 ]; then
+            echo "boot-checking all ${#all[@]} repacked libs: ${all[*]}"
+            pick=("${all[@]}")
+          else
+            mapfile -t pick < <(printf '%s\n' "${all[@]}" | python3 -c 'import sys; s = [l.strip() for l in sys.stdin if l.strip()]; n = len(s); k = 10; idx = sorted({round(i * (n - 1) / (k - 1)) for i in range(k)}); print("\n".join(s[i] for i in idx))')
+            skipped=$(comm -13 <(printf '%s\n' "${pick[@]}" | sort) \
+                               <(printf '%s\n' "${all[@]}" | sort) | tr '\n' ' ')
+            echo "boot-checking an even sample of ${#pick[@]} of the" \
+                 "${#all[@]} repacked libs: ${pick[*]}"
+            echo "NOT boot-checked this run: $skipped"
+          fi
+          for s in "${pick[@]}"; do
+            echo "== boot check: $s"
+            node scripts/wasm_boot_check.js "site/$s" site/_driver \
+                --timeout 60
+          done
 
       - name: Save packed-lib cache
         if: steps.build.outputs.changed != '0'

scripts/build_site.sh

@@ -51,9 +51,21 @@
 #   repacked.  Bundles for slugs that no longer exist (or turned noboot)
 #   are pruned.
 #
-# Outputs (for CI): appends "repacked=<n>" and "changed=<n>" to
-# $GITHUB_OUTPUT when set (changed = repacked + pruned, i.e. whether the
-# cache is worth re-saving).
+# The cache dir also carries lib-commits.json -- slug -> the last commit
+# that actually changed libs/<slug> (shown on the index cards), maintained
+# by scripts/update_lib_commits.py via the GitHub commits API because the
+# shallow CI clone has no usable history.  Same durable-anchor/disposable-
+# cache split as above: entries self-validate against the slug's current
+# tree hash, so a lost or evicted cache only costs API re-queries, never
+# wrong data (see that script's header for the full model).
+#
+# Outputs (for CI): appends "repacked=<n>", "changed=<n>" and
+# "repacked_slugs=<space-separated slugs>" to $GITHUB_OUTPUT when set
+# (changed = repacked + pruned, i.e. whether the cache is worth re-saving;
+# a lib-commits.json refresh alone deliberately does NOT count -- re-saving
+# the ~2GB cache to persist a ~10KB mapping costs more than the API
+# re-queries it would save.  repacked_slugs drives the boot smoke test:
+# CI boot-checks what was actually repacked this run, not a fixed lib).
 
 set -euo pipefail
 
@@ -78,6 +90,9 @@ mkdir -p "$CACHE_DIR/packed"
 MANIFEST="$CACHE_DIR/manifest.json"
 
 # --- 1. derive per-lib WASM status (also writes scripts/wasm_status.json) ---
+# (this early run exists for the status JSON, which the slug list below
+# needs; the index it renders is provisional -- step 7 re-renders it with
+# the freshly updated per-lib last-changed mapping)
 python3 "$SELF_DIR/gen_site_index.py" --out "$CACHE_DIR/index-staging"
 
 # Packable slugs: everything except noboot (a noboot lib gets no link on the
@@ -150,6 +165,7 @@ fi
 
 # --- 4. repack libs changed since last publish (or lacking a valid bundle) --
 REPACKED=0
+REPACKED_SLUGS=""
 REUSED=0
 NEW_MANIFEST_LIBS=$(mktemp)
 trap 'rm -f "$NEW_MANIFEST_LIBS"' EXIT
@@ -179,6 +195,7 @@ for slug in $SLUGS; do
   "$SELF_DIR/pack_lib_for_web.sh" "$slug" "$RELEASE_DIR" "$RELEASE_DIR" \
       "$CACHE_DIR/packed/$slug"
   REPACKED=$((REPACKED + 1))
+  REPACKED_SLUGS="$REPACKED_SLUGS$slug "
 done
 
 # --- 5. prune bundles whose slug is gone / no longer packable ---------------
@@ -206,7 +223,20 @@ out = {'driver_tag': sys.argv[2], 'packer': sys.argv[3], 'libs': libs}
 json.dump(out, open(sys.argv[1], 'w'), indent=1, sort_keys=True)
 PYEOF
 
-# --- 7. assemble the site ----------------------------------------------------
+# --- 7. per-lib "last changed" mapping + final index render -----------------
+# Refresh lib-commits.json (slug -> last commit that changed libs/<slug>,
+# via the GitHub commits API pinned to HEAD -- see update_lib_commits.py's
+# header for why the shallow clone forces this and why a lost cache is
+# harmless), then re-render the index with that info on the cards.  The
+# update never fails the build: last-changed info is cosmetic, and a card
+# without it beats a blocked deploy.
+python3 "$SELF_DIR/update_lib_commits.py" \
+    --mapping "$CACHE_DIR/lib-commits.json" \
+    --head "$(git -C "$REPO" rev-parse HEAD)"
+python3 "$SELF_DIR/gen_site_index.py" --out "$CACHE_DIR/index-staging" \
+    --commits "$CACHE_DIR/lib-commits.json"
+
+# --- 8. assemble the site ----------------------------------------------------
 rm -rf "$SITE_DIR"
 mkdir -p "$SITE_DIR/_driver"
 cp "$RELEASE_DIR/fluffos.js" "$RELEASE_DIR/fluffos.wasm" \
@@ -220,7 +250,7 @@ done
 cp "$CACHE_DIR/index-staging/index.html" "$SITE_DIR/index.html"
 touch "$SITE_DIR/.nojekyll"
 
-# --- 8. summary --------------------------------------------------------------
+# --- 9. summary --------------------------------------------------------------
 echo
 echo "== per-lib packed sizes =="
 du -sm "$SITE_DIR"/*/ | sort -n
@@ -231,5 +261,6 @@ if [ -n "${GITHUB_OUTPUT:-}" ]; then
   {
     echo "repacked=$REPACKED"
     echo "changed=$((REPACKED + PRUNED))"
+    echo "repacked_slugs=${REPACKED_SLUGS% }"
   } >> "$GITHUB_OUTPUT"
 fi