Performance: ConversationTranscriptExport.fetchAllMessages full-sorts every page just to pick the pagination anchor (redundant O(p log p) per page)

[erskingardner]

Severity

LOW (redundant work on the export path; scales with history depth)

Summary

Inside the export pagination loop, the code sorts the entire page only to read its single minimum element (to advance the before cursor). The full O(p log p) sort is wasted — a linear min scan suffices.

Location / Evidence

whitenoise-mac/Core/ConversationTranscriptExport.swift:112-120

guard page.hasMoreBefore else { break }
let orderedPage = sortChronologically(page.messages)
guard let oldest = orderedPage.first else {          // only .first is used
    throw ExportError.emptyPageWithMoreHistory
}
let nextBefore = oldest.timelineAt
let nextBeforeMessageId = oldest.messageIdHex

The only sort that is actually required is the final sortChronologically(Array(collectedById.values)) at :131 (dictionary values are unordered).

Failure scenario

Exporting a 10k-message group with pageLimit = 200 walks ~50 pages, sorting 200 elements each iteration (~50 × 200 log 200 comparisons) purely to find each page's minimum, instead of ~50 × 200 with a linear scan. The cost grows with history depth on the export path.

Why this is not a duplicate

• Performance: ConversationTranscriptExport.makeDocument re-sorts the already-sorted full message history (redundant O(n log n) on the uncapped export path) #516 is makeDocument re-sorting the final full history — a different location/call.
• Performance: transcript export materializes the entire group history plus ~5 full copies in memory simultaneously, with no cap on message count #500 is export memory; Performance: ConversationTranscriptExport pagination loop has no cooperative cancellation (runs the full multi-page FFI walk to completion even when cancelled) #316 is export cancellation; Bug: ConversationTranscriptExport silently truncates when a non-empty page fails to advance the cursor (incomplete fix of #139) #289/ConversationTranscriptExport silently truncates history if the FFI returns an empty page with hasMoreBefore == true #139 are cursor-advance truncation correctness.
  This per-page sort-to-find-min in the fetch loop is not covered by any of them.

Suggested fix

Replace the per-page sort with a single min scan using the same strict-weak comparator:

guard let oldest = page.messages.min(by: { lhs, rhs in
    lhs.timelineAt != rhs.timelineAt ? lhs.timelineAt < rhs.timelineAt
                                     : lhs.messageIdHex < rhs.messageIdHex
}) else { throw ExportError.emptyPageWithMoreHistory }

Keep the single sortChronologically(...) at :131 for output ordering.

Found during a full-codebase review.