From 1f3e4b292be50ea6db4c189a8f1381cf3be40523 Mon Sep 17 00:00:00 2001
From: Sean Dolbec <sean@example.com>
Date: Sat, 31 Oct 2026 13:30:43 -0400
Subject: [PATCH] Upstream watch: store 12-character digests so the dashboard
 fits the issue body

Since 2026-10-28, 23 of the last 30 scheduled runs failed with 'Dashboard exceeds
GitHub issue size': 510 tracked sources (434 discovered branch heads) with
40-character SHAs, which barely compress, pushed issue #51 past 65,536
characters, so no new upstream changes were recorded.

change_marker now keeps the first 12 characters of sha, base_sha and sha256.
Stored full-length values shorten to the same prefix, so the first run after
this reports no false changes; legacy pending entries go through the same
marker. The live dashboard state migrates from 62.9k to 38.9k characters, and
a test covers 520 sources all pending.
---
 mseries/local-ai/github-watch.py      | 15 ++++++++++---
 mseries/local-ai/test_github_watch.py | 32 +++++++++++++++++++++++++++
 2 files changed, 44 insertions(+), 3 deletions(-)

diff --git a/mseries/local-ai/github-watch.py b/mseries/local-ai/github-watch.py
index a671444d8..7ac57c9f7 100644
--- a/mseries/local-ai/github-watch.py
+++ b/mseries/local-ai/github-watch.py
@@ -13,6 +13,8 @@
 HERE = Path(__file__).resolve().parent
 MARKER = '<!-- mseries-upstream-dashboard -->'
 TITLE = 'Upstream watch: changes awaiting review'
+DIGEST_FIELDS = ('sha', 'base_sha', 'sha256')
+DIGEST_CHARS = 12
 
 
 def api(path, method='GET', data=None):
@@ -35,8 +37,8 @@ def reconcile(previous, observations, errors, acknowledged=False):
     if not acknowledged:
         for key, change in previous.get('pending', {}).items():
             # Migrate legacy {from, to} entries without losing the first value.
-            pending[key] = (change['from'] if isinstance(change, dict)
-                            and 'from' in change and 'to' in change else change)
+            pending[key] = change_marker(change['from'] if isinstance(change, dict)
+                                         and 'from' in change and 'to' in change else change)
     for key, current in observations.items():
         current = change_marker(current)
         old = snapshots.get(key)
@@ -51,7 +53,14 @@ def change_marker(value):
     if not isinstance(value, dict):
         return value
     fields = ('sha', 'base_sha', 'branch', 'release', 'sha256', 'count', 'latest', 'pushed', 'ahead', 'behind', 'status')
-    return {k: value[k] for k in fields if k in value}
+    marker = {k: value[k] for k in fields if k in value}
+    # Hex digests barely compress, and 500+ tracked branch heads at 40 characters overflowed the
+    # 65,536-character issue body. Twelve characters still detect any move of the same ref, and
+    # stored full-length values shorten to the same prefix, so migration reports no false changes.
+    for k in DIGEST_FIELDS:
+        if isinstance(marker.get(k), str):
+            marker[k] = marker[k][:DIGEST_CHARS]
+    return marker
 
 
 def encode_state(state):
diff --git a/mseries/local-ai/test_github_watch.py b/mseries/local-ai/test_github_watch.py
index d3894b6ba..290af26b0 100644
--- a/mseries/local-ai/test_github_watch.py
+++ b/mseries/local-ai/test_github_watch.py
@@ -1,5 +1,6 @@
 #!/usr/bin/env python3
 """State migration and bounded dashboard rendering for github-watch.py."""
+import hashlib
 import importlib.util
 from pathlib import Path
 import unittest
@@ -13,6 +14,10 @@
 people_spec.loader.exec_module(people)
 
 
+def sha1(text):
+    return hashlib.sha1(text.encode()).hexdigest()
+
+
 class GitHubWatchState(unittest.TestCase):
     def test_legacy_pending_change_migrates_without_losing_latest_revision(self):
         key = 'branch:example/project:main'
@@ -137,6 +142,33 @@ def test_large_legacy_dashboard_fits_the_issue_body_limit_after_migration(self):
         self.assertEqual(len(state['pending']), 173)
         self.assertLess(len(body.encode()), 65000)
 
+    def test_full_length_digests_shorten_without_reporting_a_change(self):
+        key = 'branch:example/project:main'
+        full, base = 'a1b2c3d4e5f6' + '0' * 28, 'f' * 40
+        previous = {'snapshots': {key: {'sha': full, 'base_sha': base}},
+                    'pending': {'heads:example/project': {'from': {'sha256': 'e' * 64}, 'to': {'sha256': 'd' * 64}}}}
+
+        state = watch.reconcile(previous, {key: {'sha': full, 'base_sha': base}}, [])
+
+        self.assertNotIn(key, state['pending'])
+        self.assertEqual(state['snapshots'][key], {'sha': 'a1b2c3d4e5f6', 'base_sha': 'ffffffffffff'})
+        self.assertEqual(state['pending']['heads:example/project'], {'sha256': 'eeeeeeeeeeee'})
+
+    def test_every_source_pending_fits_with_520_full_length_heads(self):
+        # 2026-10-31: 510 tracked sources with 40-character SHAs overflowed the issue body.
+        observations, snapshots = {}, {}
+        for index in range(520):
+            key = f'branch:example/repository-{index % 24:02d}:feature/long-branch-name-{index:04d}'
+            snapshots[key] = {'branch': key.rsplit(':', 1)[1], 'sha': sha1(f'old-{index}'),
+                              'base_sha': sha1(f'base-{index}'), 'ahead': index, 'behind': index % 9,
+                              'status': 'diverged'}
+            observations[key] = dict(snapshots[key], sha=sha1(f'new-{index}'))
+
+        state = watch.reconcile({'snapshots': snapshots}, observations, [])
+
+        self.assertEqual(len(state['pending']), 520)
+        self.assertLess(len(watch.render(state).encode()), 65000)
+
 
 class PeopleWatchRetry(unittest.TestCase):
     class Response: