@ mention search in thread composer hangs when it searches thread history

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #31065

Powered by Codex Action

Turns out, RepoPrompt created threads inside / (can also be elsewhere), with first message starting as <codex reminder> You are operating in text only mode... which effectively named the thread AS the first message (aka up to tens of thousands of tokens), instead of a short title, therefore messing up search in-thread or in-app.

So take-home could be to have a thread title character limit, no matter "who" the originator is

Steps I took:

# Find Codex state DBs
find "$HOME/.codex" -name 'state_*.sqlite' -print

# Count unarchived cwd=/ threads per DB
for db in "$HOME/.codex/state_5.sqlite" "$HOME/.codex/sqlite/state_5.sqlite"; do
  [ -f "$db" ] || continue
  echo "== $db =="
  sqlite3 "$db" "select archived, count(*) from threads where cwd='/' group by archived order by archived;"
done

# Count unarchived <codex reminder> rows, any cwd
for db in "$HOME/.codex/state_5.sqlite" "$HOME/.codex/sqlite/state_5.sqlite"; do
  [ -f "$db" ] || continue
  echo "== $db =="
  sqlite3 "$db" "select cwd, count(*) from threads where archived=0 and (title like '<codex reminder>%' or preview like '<codex reminder>%' or first_user_message like '<codex reminder>%') group by cwd;"
done

# Sample unarchived cwd=/ rows
for db in "$HOME/.codex/state_5.sqlite" "$HOME/.codex/sqlite/state_5.sqlite"; do
  [ -f "$db" ] || continue
  echo "== $db =="
  sqlite3 -header -column "$db" "select id, rollout_path, source, tokens_used, length(title) title_len, substr(title,1,100) title from threads where cwd='/' and archived=0 order by updated_at desc limit 20;"
done

To check whether cwd=/ rows were RepoPrompt-originated:

python3 - <<'PY'
import sqlite3, json
from pathlib import Path

for db in [Path.home()/".codex/state_5.sqlite", Path.home()/".codex/sqlite/state_5.sqlite"]:
    if not db.exists(): continue
    con = sqlite3.connect(db)
    rows = con.execute("select id, rollout_path from threads where cwd='/' and archived=0").fetchall()
    print("\n==", db, "==")
    for thread_id, path in rows[:50]:
        origin = "?"
        try:
            for line in open(path):
                o = json.loads(line)
                if o.get("type") == "session_meta":
                    origin = (o.get("payload") or {}).get("originator")
                    break
        except FileNotFoundError:
            origin = "missing-file"
        print(thread_id, origin, path)
PY

To archive RepoPrompt-originated cwd=/ rows, close Codex first, then:

python3 - <<'PY'
import sqlite3, json, shutil, time
from pathlib import Path

now = int(time.time())
for db in [Path.home()/".codex/state_5.sqlite", Path.home()/".codex/sqlite/state_5.sqlite"]:
    if not db.exists(): continue
    backup = Path(f"/tmp/{db.name}.{db.parent.name}.before-rp-slash-archive.{now}.sqlite")
    shutil.copy2(db, backup)

    con = sqlite3.connect(db)
    rows = con.execute("select id, rollout_path from threads where cwd='/' and archived=0").fetchall()
    ids = []
    for thread_id, path in rows:
        try:
            origin = next(
                ((json.loads(line).get("payload") or {}).get("originator")
                 for line in open(path)
                 if json.loads(line).get("type") == "session_meta"),
                None,
            )
        except FileNotFoundError:
            origin = None
        if origin == "repoprompt":
            ids.append(thread_id)

    con.executemany(
        "update threads set archived=1, archived_at=? where id=?",
        [(now, thread_id) for thread_id in ids],
    )
    con.commit()
    print(db, "archived", len(ids), "backup", backup)
PY