Fix perpetual re-regeneration of projects with up-to-date caches

claude_code_log/converter.py

@@ -18,6 +18,8 @@
     from .cache import CacheManager
 
 from .utils import (
+    coalesce_trunk_session_id,
+    collect_trunk_session_ids,
     format_timestamp_range,
     get_parent_session_id,
     get_project_display_name,
@@ -1431,8 +1433,8 @@ def compute_session_data(
         ):
             continue
 
-        session_id = get_parent_session_id(getattr(message, "sessionId", ""))
-        if not session_id or session_id in warmup:
+        session_id = coalesce_trunk_session_id(message, warmup)
+        if not session_id:
             continue
 
         if session_id not in result:
@@ -1689,10 +1691,19 @@ def _generate_paginated_html(
         total_output_tokens = 0
         total_cache_creation_tokens = 0
         total_cache_read_tokens = 0
+        # The cached count must be the sessions-table sum, because
+        # is_page_stale() compares it against SUM(sessions.message_count)
+        # on the next run. len(page_messages) counts what's rendered
+        # (including Summary/AiTitle/Attachment entries that
+        # compute_session_data skips), so caching it left any page where
+        # the two rules diverge permanently "message_count_changed" —
+        # regenerating on every single run.
+        page_sessions_message_count = 0
 
         for session_id in page_session_ids:
             if session_id in session_data:
                 s = session_data[session_id]
+                page_sessions_message_count += s.message_count
                 if s.first_timestamp and (
                     first_timestamp is None or s.first_timestamp < first_timestamp
                 ):
@@ -1778,7 +1789,7 @@ def _generate_paginated_html(
             html_path=html_path,
             page_size_config=page_size,
             session_ids=page_session_ids,
-            message_count=page_message_count,
+            message_count=page_sessions_message_count,
             first_timestamp=first_timestamp,
             last_timestamp=last_timestamp,
             total_input_tokens=total_input_tokens,
@@ -2046,16 +2057,9 @@ def convert_jsonl_to(
         assert cache_manager is not None  # Ensured by use_pagination condition
         # Use cached session data if available, otherwise build from messages
         if cached_data is not None:
-            warmup_session_ids = get_warmup_session_ids(messages)
-            current_session_ids: set[str] = set()
-            for message in messages:
-                session_id = getattr(message, "sessionId", "")
-                if (
-                    session_id
-                    and session_id not in warmup_session_ids
-                    and not is_agent_session(session_id)
-                ):
-                    current_session_ids.add(session_id)
+            current_session_ids = collect_trunk_session_ids(
+                messages, get_warmup_session_ids(messages)
+            )
             session_data = {
                 session_id: session_cache
                 for session_id, session_cache in cached_data.sessions.items()
@@ -2398,20 +2402,14 @@ def _generate_individual_session_files(
 
     ext = get_file_extension(format)
     suffix = _variant_suffix(detail, compact, format, no_timestamps, no_recaps)
-    # Pre-compute warmup sessions to exclude them
-    warmup_session_ids = get_warmup_session_ids(messages)
-
-    # Find all unique session IDs (excluding warmup and agent sessions)
-    session_ids: set[str] = set()
-    for message in messages:
-        if hasattr(message, "sessionId"):
-            session_id: str = getattr(message, "sessionId")
-            if (
-                session_id
-                and session_id not in warmup_session_ids
-                and not is_agent_session(session_id)
-            ):
-                session_ids.add(session_id)
+    # Find all unique session IDs, excluding warmup sessions and
+    # coalescing agent sessionIds to their trunk — same rule as
+    # compute_session_data() when it writes the sessions table.
+    # Dropping (rather than coalescing) agent ids left agent-sidechain-
+    # only sessions in the sessions table but never rendered, so
+    # get_stale_sessions() flagged them "not_cached" on every run —
+    # regenerating the project forever without ever writing the file.
+    session_ids = collect_trunk_session_ids(messages, get_warmup_session_ids(messages))
 
     # Get session data from cache for better titles
     session_data: dict[str, Any] = {}
@@ -3154,10 +3152,8 @@ def _rel_to_index(p: Path) -> str:
                     ),
                 ):
                     continue
-                if not hasattr(_msg, "sessionId"):
-                    continue
-                _sid = get_parent_session_id(getattr(_msg, "sessionId", ""))
-                if not _sid or _sid in warmup_for_teams:
+                _sid = coalesce_trunk_session_id(_msg, warmup_for_teams)
+                if not _sid:
                     continue
                 _tn = getattr(_msg, "teamName", None)
                 if _tn and _sid not in team_name_per_session:

claude_code_log/utils.py

@@ -692,6 +692,45 @@ def get_warmup_session_ids(messages: list[TranscriptEntry]) -> set[str]:
     return warmup_sessions
 
 
+def coalesce_trunk_session_id(
+    message: TranscriptEntry, warmup_session_ids: set[str]
+) -> str:
+    """Return ``message``'s sessionId coalesced to its trunk, or ``""``.
+
+    The shared normalization every session-collection site must agree
+    on: agent synthetic sessionIds (``{trunk}#agent-{id}``) coalesce to
+    their trunk via ``get_parent_session_id``, and warmup-only sessions
+    are excluded. Returns ``""`` for entries without a sessionId (e.g.
+    ``SummaryTranscriptEntry``) or whose trunk is a warmup session.
+    """
+    session_id = getattr(message, "sessionId", "") or ""
+    if session_id:
+        session_id = get_parent_session_id(session_id)
+    if session_id in warmup_session_ids:
+        return ""
+    return session_id
+
+
+def collect_trunk_session_ids(
+    messages: list[TranscriptEntry], warmup_session_ids: set[str]
+) -> set[str]:
+    """Collect the distinct trunk sessionIds present in ``messages``.
+
+    Coalescing (rather than dropping) agent sessionIds matters: a fork
+    session whose own messages were all deduplicated into the original
+    session can survive ONLY through its agent sidechains, and it must
+    still claim its page/file slot or its messages are grouped under a
+    session no output owns and vanish (regenerating forever — see
+    ``_generate_individual_session_files``).
+    """
+    session_ids: set[str] = set()
+    for message in messages:
+        session_id = coalesce_trunk_session_id(message, warmup_session_ids)
+        if session_id:
+            session_ids.add(session_id)
+    return session_ids
+
+
 # Artifact favicons are 1-2 emoji (ZWJ sequences can span ~a dozen code
 # points); anything longer is malformed input not worth echoing inline.
 # Shared by the HTML and Markdown renderers so the cap can't drift.