Skip to content

Review 2026-04-18 — eMuleAI vs Main Hardening Pass

Scope

Focused comparison of:

  • analysis\emuleai
  • current workspaces\v0.72a\app\eMule-main

The pass intentionally focused on:

  • networking
  • performance
  • file-handling
  • security-adjacent hardening
  • concrete bug fixes, races, and logical flaws

Theme and translation work was intentionally ignored.

The branch goal remained the default workspace rule: stay as close as possible to stock eMule behavior and only promote narrow, high-value fixes.

Already Landed On Main

The comparison first revalidated what current main already absorbed and therefore should not be reopened as backlog work:

  • REF-007 — MIME magic-table refresh plus WebM / Matroska split
  • FEAT-020 — peer geolocation landed with DB-IP rather than GeoLite2
  • FEAT-022 — startup config override via -c
  • FEAT-026 — shared startup cache / known-file lookup index / sharedcache.dat
  • FEAT-027 — startup sequencing and trace-backed profiling
  • FEAT-028 — Shared Files list virtualization and hardening
  • BUG-029 — long-path tail hardening across config/media/GeoLocation/shell edges
  • BUG-025 / BUG-026 / BUG-027 already cover eMuleAI's matching:
  • hash-open diagnostic corruption
  • Close All Search Results crash
  • destructive IP-filter promotion failure

New Confirmed Backlog Promotions

BUG-030 — obfuscated server login capability advertisement

eMuleAI contains a narrow guard that suppresses REQUESTCRYPT / REQUIRECRYPT in the server login packet when the main server socket is already obfuscated. Current main still advertises those flags unconditionally from preferences, and the eMuleAI rationale matches a plausible current-tree networking symptom: fully obfuscation-capable server lists can require multiple attempts before login succeeds.

BUG-031 — transient lock/share retry for shared-file hashing

Current main still performs only a single open attempt when hashing a newly discovered shared file. eMuleAI adds a bounded retry on ERROR_SHARING_VIOLATION and ERROR_LOCK_VIOLATION, which is a narrow, stock-preserving improvement for files still being copied or moved into shared folders.

BUG-032 — known2.met lock timeout during AICH hashset save

Current main still lets CAICHRecoveryHashSet::SaveHashSet() give up after five seconds waiting for the global known2.met mutex. eMuleAI removes that timeout. This looks like a real false-failure race rather than a feature difference.

Revalidated But Not Promoted

Several eMuleAI release-note items were rechecked but were not promoted into new backlog items in this pass:

  • duplicate shared files rehashed on startup / duplicate-files history interactions
  • uploads disappearing while shared files reload
  • shared-files status-counter crash
  • long UI freezes before hashing large shared folders
  • upload speedometer visual quirks

Reason:

  • current main already has large landed share/startup/list-control changes on that same surface (FEAT-026 / FEAT-027 / FEAT-028 / BUG-029)
  • the remaining eMuleAI fixes in that area are intertwined with broader worker-thread, virtual-list, duplicate-history, or watcher behavior
  • there was not enough direct current-tree evidence to promote a new narrow stock-friendly bug item yet

Those areas should be revalidated again only if a live repro appears on current main.

Explicit Non-Promotions

The following eMuleAI areas were intentionally not promoted because they are too feature- drifty for the current branch goal:

  • uTP transport
  • NAT traversal / buddy extensions
  • threaded shared-files reload and filesystem watcher
  • source saver / source cache product features
  • dark mode and theme work
  • translation-system changes

Outcome

This pass refreshed the active backlog to match current main and promoted only the remaining narrow hardening deltas that still look worth bringing in:

It also closed or corrected stale backlog state for already-landed work on main.