Function parseSessionFile

src/parser.ts:1406–1477

(
  filePath: string,
  project: string,
  seenMsgIds: Set<string>,
  dateRange?: DateRange,
)

}

async function parseSessionFile(
  filePath: string,
  project: string,
  seenMsgIds: Set<string>,
  dateRange?: DateRange,
): Promise<{ session: SessionSummary; canonicalCwd?: string } | null> {
  // Skip files whose mtime is older than the range start. A session file
  // can only contain entries up to its last-modified time; if that predates
  // the requested range, nothing in this file can match.
  if (dateRange) {
    try {
      const s = await stat(filePath)
      if (s.mtimeMs < dateRange.start.getTime()) return null
    } catch { /* fall through to normal read; missing stat shouldn't break parsing */ }
  }
  const entries: JournalEntry[] = []
  let hasLines = false

  // When a dateRange is given, skip user/assistant lines whose timestamp
  // is older than range.start - 24h without calling JSON.parse. Huge lines
  // that cannot be skipped are yielded as Buffers and compact-parsed without
  // converting the whole line into a V8 string.
  const earlySkipThreshold = dateRange
    ? new Date(dateRange.start.getTime() - 86_400_000).toISOString()
    : null
  const skipFn = earlySkipThreshold
    ? (head: string) => shouldSkipLine(head, earlySkipThreshold)
    : undefined

  for await (const line of readSessionLines(filePath, skipFn, { largeLineAsBuffer: true })) {
    hasLines = true
    const entry = parseJsonlLine(line)
    if (entry) entries.push(compactEntry(entry))
  }

  if (!hasLines) return null

  if (entries.length === 0) return null

  const sessionId = basename(filePath, '.jsonl')
  const dedupedEntries = dedupeStreamingMessageIds(entries)
  let turns = groupIntoTurns(dedupedEntries, seenMsgIds)
  if (dateRange) {
    // Bucket a turn by the timestamp of its first assistant call (when the cost was
    // actually incurred). Filtering entries directly produced orphan assistant calls
    // when a user message sat in one day and the response landed in another -- those
    // got pushed as turns with empty timestamps, which some code paths counted and
    // others dropped, producing inconsistent Today totals.
    turns = turns.filter(turn => {
      if (turn.assistantCalls.length === 0) return false
      const firstCallTs = turn.assistantCalls[0]!.timestamp
      if (!firstCallTs) return false
      const ts = new Date(firstCallTs)
      return ts >= dateRange.start && ts <= dateRange.end
    })
    if (turns.length === 0) return null
  }
  const classified = turns.map(classifyTurn)