hermes-admin/hermes-agent
1
0
Fork 0
Code Issues Pull Requests Actions 4 Packages Projects Releases Wiki Activity

fix(state): keep stored token_count when a digest-matched replay rewrites its row

Browse Source 
Replay dicts from _rows_to_conversation never decode token_count, so the
serialized row always carried NULL. On the unaddressed (message_uid + CAS
snapshot) path that meant every legitimate rewrite nulled the original row's
token_count, and a dict whose content was unchanged still differed in
token_count and triggered a needless _rewrite_row. A missing key means
"unknown", so keep the stored value before comparing owned columns.

Also reuse message_metadata.message_uid_or_none at the lookup call site instead
of a second hand-rolled uid guard, and reflow an over-wide comment.
This commit is contained in:
kshitijk4poor
2026-10-01 19:33:24 +05:30
committed by kshitij
parent d037aba450
commit 440b97eae0
1 changed files with 10 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
agent/transcript_repair.py
View File

@@ -10,7 +10,7 @@ from typing import Any, Callable, Dict, List, Mapping
from agent.context_compressor import _DB_PERSISTED_MARKER
from agent.message_metadata import (
CANONICAL_ROW, DB_ROW_SNAPSHOT, MESSAGE_UID, copy_identity_fields)
CANONICAL_ROW, DB_ROW_SNAPSHOT, MESSAGE_UID, copy_identity_fields, message_uid_or_none)
from hermes_state_common import _id_chunks, _placeholders
from hermes_state_identity import _fill_missing_tool_call_uids, _restore_row_identity
from hermes_state_messages import _MESSAGE_WRITE_COLUMNS
@@ -89,7 +89,7 @@ def is_content_blank(content: Any) -> bool:
def _active_logical_message_row(
conn: sqlite3.Connection, session_id: str, role: str, message_uid: Any,
conn: sqlite3.Connection, session_id: str, role: str, message_uid: str | None,
) -> Mapping[str, Any] | None:
"""Newest active physical row for one durable logical message.
@@ -98,7 +98,7 @@ def _active_logical_message_row(
when the live dict also carries a stored-row snapshot, so a fresh message that merely resembles an
older one can never be adopted here.
"""
if not isinstance(message_uid, str) or not message_uid:
if message_uid is None:
return None
return conn.execute(
"SELECT * FROM messages WHERE session_id = ? AND active = 1 AND role = ? AND message_uid = ? "
@@ -135,7 +135,7 @@ def resolve_and_repair_transcript_batch(
elif isinstance(expected, str):
# Logical identity + the CAS version prove this dict came from durable replay. A new message
# has neither proof, even when role/content/timestamp happen to equal an older message exactly.
target_row = _active_logical_message_row(conn, session_id, role, msg.get(MESSAGE_UID))
target_row = _active_logical_message_row(conn, session_id, role, message_uid_or_none(msg))
if target_row is None:
inserted_rows.append(msg)
continue
@@ -155,6 +155,9 @@ def resolve_and_repair_transcript_batch(
adopt = transcript_row_snapshot(target_row) != expected
if not adopt:
serialized = serialize_message_fn(msg, float(target_row["timestamp"]))
if "token_count" not in msg:
# Replay dicts never carry token_count; a missing key means "unknown", not NULL.
serialized = {**serialized, "token_count": target_row["token_count"]}
if any(target_row[column] != serialized[column] for column in _OWNED_COLUMNS):
_rewrite_row(conn, session_id, target_row, serialized)
wrote = True
@@ -172,9 +175,9 @@ def resolve_and_repair_transcript_batch(
).rowcount > 0
else:
# Legacy dict (no digest: a row-addressed resume, a clone, or a repair_alternation=False
# projection) over a non-blank assistant row: another writer already filled it. Adopt its content only, never the whole row: the live tool_calls /
# reasoning* / codex_* fields may be sanitizer-fixed while the durable JSON still holds the raw
# escaped surrogate, and live-only fields must survive.
# projection) over a non-blank assistant row: another writer already filled it. Adopt its content
# only, never the whole row: the live tool_calls / reasoning* / codex_* fields may be sanitizer-fixed
# while the durable JSON still holds the raw escaped surrogate, and live-only fields must survive.
if role == "assistant":
canonical = {"content": decode_content_fn(target_row["content"]), _CONTENT_ONLY: True}