TOGBankClassic Changelog

[v0.9.14] (2026-03-24) - Inventory Sort Improvements

Status: Production Ready

Improvements

• Sort by Rarity and Sort by Level added to inventory window — The sort button now cycles through four modes: A-Z → By Type → By Rarity → By Level. Rarity sort orders highest rarity first (epic before rare before uncommon), with A-Z as a tiebreaker. Level sort orders by required level descending, also with A-Z as a tiebreaker. Location: Item.lua Sort; UI/Inventory.lua.

---

[v0.9.13] (2026-03-23) - Request Index Flood Fix & UI Polish

Status: Production Ready

Bug Fixes

• Request index flood eliminated for large request logs — Three compounding issues caused guilds with large request lists (500+ requests, 85+ index chunks) to flood guild chat when multiple peers responded to the same index query simultaneously:

  • Duplicate drain on re-query — New queries arriving while an 85-chunk drain was already in progress were coalescing and re-queuing a full second copy of the index on top of the active drain, repeating indefinitely. flushIndexQueue now drops the re-send when a guild-broadcast drain is already in flight. Location: RequestLog.lua flushIndexQueue.
  • Multi-peer simultaneous send — All peers responded at the same fixed delay, causing N×85 chunks from N peers. Index responses now use random jitter (20–40s instead of fixed 20s), and a first-responder rule cancels a peer's pending send if it sees another peer's togbank-ri already draining — one responder covers the whole guild. Location: RequestLog.lua EnqueueIndexResponse, ReceiveRequestsIndex.
  • First-responder starvation — Each incoming chunk from an active drain was restarting the suppression timer, permanently blocking other peers. A sliding suppression window (indexResponseSuppressedUntil) is now extended per chunk and expires ~3s after the last chunk arrives, at which point all peers compete fairly via jitter. Location: RequestLog.lua.

• "Broadcasted hash for <alt>" respects Mute Sync Progress Messages — The message printed during /togbank share was always shown regardless of settings. Now gated behind IsSyncProgressMuted(). Location: Guild.lua Share.

• "Syncing requests with guild…" respects Mute Sync Progress Messages — Same fix as above for the message printed during /togbank sync. Location: Chat.lua PerformSync.

---

[v0.9.12] (2026-03-20) - Stale Banker Indicators, Version Check & Bug Fixes

Status: Production Ready

Bug Fixes

• Quality border colors fixed for all gear — Three compounding issues caused weapons and armor to always show a white quality border regardless of actual rarity. (1) GetItems Branch 1 used GetItemInfoInstant (no rarity field) — fixed in v0.9.10 by switching to GetItemInfo. (2) Item:Sort was defaulting nil rarity to 1 (common/white) via rarity = rarity or 1 before DrawItem ran — this masked nil rarity with a truthy value, making the sync/async fallbacks in DrawItem unreachable. The sort comparators already handle nil safely, so the assignment was both unnecessary and harmful; it has been removed. (3) Remote-synced gear not yet in the client cache still had nil rarity even after fix (1) — DrawItem now has a sync fallback (GetItemInfo(item.Link) at draw time) and an async fallback (ContinueOnItemLoad → GetItemInfo(item.Link) → SetVertexColor) for items that load into cache after first render. All lookups use the full item link; ID-based lookup is intentionally excluded because the base item ID returns the wrong rarity for suffixed gear. Locations: Item.lua Sort (~line 449); UI.lua DrawItem (~line 145).

• Mail item tooltip fixed — Items sourced from the mailbox were stored with a double item:item:… prefix, causing tooltips to fail and show a blank name. Location: UI.lua mail item handling.

• Empty index responses after /togbank wipe eliminated — After a wipe the local requests hash is 00000000. We were still responding to peers' index queries with an empty index. Peers now skip responding when their own hash is zero (nothing to offer). Location: Chat.lua index query respond condition.

• False [WARN] Invalid request version 0 after /togbank wipe removed — Version 0 is valid for a freshly wiped or initialised client; the out-of-range check now exempts it. Location: RequestLog.lua GetRequestsVersion.

• P2P "peer acknowledged" message respects Mute Sync Progress Messages — The "P2P: Peer X acknowledged Y – will send delta" line was always printed to chat regardless of settings. It is now gated behind the Mute Sync Progress Messages checkbox. Location: Chat.lua P2P acknowledgement handler.

Improvements

• Stale banker tab indicators — Banker tabs in the inventory window now turn red when a peer has broadcast a newer hash for that banker (i.e. the alt is HLR-pending per the same definition as /togbank hashdebug). Hovering a red tab shows a tooltip explaining that other guild members have newer data and current availability may not be accurate. The staleness check (Guild:IsAltSyncPending) covers both inventory and mail hash mismatches, missing content, and is the single source of truth used by both the tab color and the tooltip — consistent with /togbank hashdebug output. Location: Guild.lua IsAltSyncPending; UI/Inventory.lua DrawContent.

• /togbank versioncheck — Broadcasts a version check to all online guild members using VersionCheck-1.0 (VC:FireBatch), waits 21 seconds for responses, then prints a sorted list of who is running which version. Because it piggybacks on VersionCheck-1.0's own protocol (VC10_REQ/VC10_RSP), it reaches members running any version of the addon that had the library — including those running versions too old to receive newer custom protocols. Replaces /togbank versions, which only saw members who had sent a message since login. Location: Chat.lua versioncheck command handler.

• Version displayed in inventory window title — The main inventory window title now shows the addon version (e.g. "TOGBankClassic v0.9.12") via GetAddOnMetadata. Location: UI/Inventory.lua DrawWindow.

Performance

• NormalizeRequestList dirty flag — The function now skips its O(N) full-table rebuild when request data hasn't changed since the last run. Previously it was called twice per index send (once via EnsureRequestsInitialized, once directly) even in steady state. It now runs only after actual data changes: first load, peer data merges, and migrations. Location: RequestLog.lua NormalizeRequestList.

Internal

• Debug system overhaul — All DebugComm calls converted to Debug(category, tag, …). P2P, COMMS, SYNC, and REQUESTS categories cleaned up. New REQUESTS/PROTO2 tag covers togbank-ri / togbank-rd2 compact protocol traffic.
• /togbank versions removed — Superseded by /togbank versioncheck, which reaches all guild members regardless of when they last sent a message.
• Removed QueryRequestsSnapshot shim — No callers remain; the modern QueryRequestsIndex replaced it entirely.
• PERF-002: removed data.requests from togbank-dv2 broadcasts — The request version/hash field was included in every periodic inventory broadcast but never consumed by receivers. Removed to avoid confusion and marginal bandwidth waste.

---

[v0.9.10] (2026-03-19) - 60% Bandwidth Reduction for Request Sync

Status: Production Ready

Performance

• Request sync uses ~60% less bandwidth — The request index and per-record wire format has been rewritten from verbose key-value dicts to compact positional arrays. Two new prefixes carry this traffic:

  • togbank-ri — requests index as a flat positional array ({version, liveCount, id, updatedAt, ..., tombId, tombTs, ...}), eliminating per-field string keys across hundreds of IDs.
  • togbank-rd2 — individual request records as positional arrays, avoiding repeated field-name overhead when syncing large request logs.

  Guilds with 500+ requests will see the most noticeable improvement during initial sync and after being offline.

Internal

• Request IDs changed to 14-char random hex — The previous actor:random composite format has been replaced with 14 random hex characters. Existing requests retain their old IDs.
• Removed statusUpdatedAt field — The per-request status-change timestamp field has been dropped from the wire format and storage schema.
• Removed dead protocol slots — togbank-dr and togbank-dc (DELTA-006 delta chain replay) were never triggered by current clients and have been removed.

---

[v0.9.8] (2026-03-19) - Request Expiry Fixes & Dropdown Improvements

Status: Production Ready

Bug Fixes

• Expiry clock anchored to wrong timestamp — PruneRequests used updatedAt as the 30-day expiry anchor, but updatedAt is bumped on every sync so done requests never actually aged out. The anchor is now statusUpdatedAt, which only changes when the request is fulfilled, cancelled, or completed. The 30-day clock now starts when the request was actually finished.
• Expired requests re-imported from peers — Stale done requests received from peers running older clients were being merged back into the local database, undoing the prune. They are now tombstoned on arrival (backdated timestamp so the tombstone itself also expires within 30 days).
• Pruning skipped on login — PruneIfNeeded was not called when the addon loaded from SavedVariables, so expired requests lingered until the first periodic share timer fired (~3 minutes). Pruning now runs immediately during Guild:Init.

Improvements

• Requester/Banker dropdowns show full history — Previously only requesters and bankers with at least one open request appeared in the filter dropdowns, making it impossible to filter by someone whose requests were all fulfilled or cancelled. The dropdowns now show everyone, split into two labelled sections: "-- Open requests --" (active names, sorted by open count) and "-- History --" (completed-only names, sorted by total count).

Internal

• Removed legacy full-snapshot request sync — The type="requests" snapshot protocol (replaced by the index/by-id protocol seven weeks ago) has been removed. A thin shim remains in QueryRequestsSnapshot to avoid crashes on mixed-version guilds during the transition.
• /togbank reqscan — New diagnostic command showing total/done/expired request counts, status breakdown, and statusUpdatedAt age distribution (0-7d, 7-14d, 14-21d, 21-30d, >30d).
• COMMS log clarity — togbank-rd log lines now include the subtype (idx / by-id) immediately before the byte-count line, making it easier to correlate log output with protocol activity.

---

[v0.9.7] (2026-03-18) - Request Retirement Fix & Status Bar Cleanup

Status: Production Ready

Bug Fixes

• Expired requests never pruned — Fulfilled and cancelled requests from 30+ days ago were accumulating indefinitely. PruneIfNeeded was only called after mutations, never on a timer, so requests were never cleaned up on clients that hadn't recently submitted or fulfilled a request. It now runs automatically every ~3 minutes via the periodic share timer.

Improvements

• Status bar network labels renamed — The network counters in the inventory status bar are now labelled Tx: (outgoing sends), Rx: (P2P data fetches), and Bcast: (sync broadcast queue), replacing the cryptic send:, P2P:, and q: labels.
• Status bar layout — When the window is too narrow to show all three sections, the right section drops first (was: center). Left + center are shown until even narrower, then left only.
• Status bar refactored into StatusBar.lua — All status bar logic (formatters, inventory summary, network parts, ticker lifecycle, hover callbacks) is now in Modules/UI/StatusBar.lua. Inventory.lua retains only two lines of status bar surface.

---

[v0.9.6] (2026-03-16) - Request Sync Throttle Overhaul & Network Status Bar

Status: Production Ready

New Features

• Request sync throttle overhaul — The requests-index pipeline was reworked to eliminate multi-minute CTL backlogs on guilds with 1000+ requests:
  • Deduplicating response drain — Replaced SendRequestsById with a queriedRequests map. Duplicate requests for the same ID from the same peer are dropped; requests from two different peers are automatically upgraded to a guild broadcast. Responses drain at one batch per second, gated on CTL queue depth.
  • Coalesced index responses — Multiple guild members querying the requests index within a 20-second window now trigger a single response instead of N. If different senders query, one guild broadcast replaces N individual whispers; sends are also deferred while the CTL queue is busy.
  • Chunked index sending — Requests-index payloads are now split into chunks of 20 IDs sent 1 second apart (previously: one ~400-packet burst). Receivers can begin fetching missing requests after the very first chunk arrives. Old clients (v0.9.5 and below) remain compatible.
• Network status bar (opt-in) — The inventory window status bar now optionally shows live sync activity. Enable in Options -> General -> "Show Network Status in Status Bar":
  • Left: send / queue / fetch counters
  • Centre: "Sending [type] to [recipient]" (next queued CTL message)
  • Right: "Backlog: N packets[, N recipients][, N requests]"
  • Sections hide automatically when the window is too narrow to show all three without overlap.
• /togbank netq — New expert command showing a full CTL queue breakdown by message type and recipient count.
• Request count in status bar — The Requests window now shows the total request count alongside the filtered count (e.g. "3 / 47").

Bug Fixes

• Self-query loop on login — If the logged-in character was the only eligible banker, QueryAltPullBased would whisper itself, triggering a useless sync loop. Self is now excluded from the banker search.

Diagnostics

• Hash values in request-index log lines are now shown in hex, consistent with the rest of the codebase.
• Outgoing COMMS log messages now include the recipient.
• Requests-by-id queries and responses (both directions) are now logged under REQUESTS/SEND and REQUESTS/RECEIVE.
• Requests-index log lines now show both the querier's hash and the local hash side-by-side, making SYNC-011 hash-match decisions easier to trace.

---

[v0.9.5] (2026-03-16) - Request Sync Diagnostics & UI Polish

Status: Production Ready

✨ New Features

• Network queue status bar — The main window status bar now shows live network activity: pending sends (send:1/3), outbound sync queue depth (q:2), P2P data fetches in flight (fetch:1), and request index sync state. During batch ID syncs the progress is shown as r:2/7 (current batch / total batches).
• Descriptive request sync status — The request index query indicator in the status bar now shows the target player name (e.g. Querying requests index from Skywise) instead of the cryptic r:idx label.
• /togbank versions restored — Addon version tracking is now populated from two sources: banker hash-list-broadcasts and non-banker requests-index queries, so /togbank versions shows all online guild members regardless of role.

🐛 Bug Fixes

• Request sync stalling on large guilds — Three timing fixes for guilds with 1000+ requests:
  • INDEX_INFLIGHT_TIMEOUT increased from 30 s to 180 s — the old value expired before a 20-batch sync (20 × 5 s = 100 s) could complete, causing the sync to silently abort mid-flight.
  • REQUESTS_BY_ID_BATCH_DELAY increased from 2 s to 5 s — gives the responding peer more time to reply before the next query batch arrives.
  • requests-index inFlight fallback timer increased from 5 s to 10 s — reduces false "guild in sync" clears when the peer's send queue is momentarily congested.
• Mail age showing "20000 days ago" — lastScan = 0 is truthy in Lua, so time() - 0 produced a timestamp relative to the Unix epoch. GetMailDataAge now treats lastScan = 0 as absent data.
• Requests date column misaligned — Dates in open/pending requests appeared shifted left of fulfilled/cancelled dates because centre-alignment repositions shorter strings. All rows now receive a same-width invisible prefix so the date text centres identically across all states.

🔍 Diagnostics

• Outgoing requests-index responses now logged — Previously only incoming queries were logged; the response send was silent, making it impossible to confirm whether we had responded.
• Hash values shown in requests-index log — The query log line now appends (their:NNNN ours:NNNN) to help diagnose SYNC-011 hash-match decisions.

---

[v0.9.4] (2026-03-15) - Request Sync Overhaul & UI Fixes

Status: Production Ready

✨ New Features

• Request status colours — Fulfilled requests are now tinted green with a checkmark icon; cancelled requests are tinted red with an X icon, making it easy to scan request history at a glance.
• Item sort toggle — Inventory view now has an A–Z / By Type sort toggle button.

🐛 Bug Fixes

• Request sync stalling on login — Fixed a critical issue where syncing after being offline could stall indefinitely. Three root causes addressed:
  • /togbank sync was silently blocked by a 60-second cooldown; it now always fires immediately when invoked manually.
  • Querying a peer for 1500+ missing requests sent one massive message, overwhelming WoW's chat throttle. Queries are now batched (50 IDs each) and staggered 2 seconds apart.
  • The responding peer now also sends replies in staggered batches, preventing their outbound throttle queue from being monopolised for several minutes.
• Options window crash — Fixed crash on open (AceConfig: expected a table, got 'function').
• Debug message formatting — Fixed %d appearing literally in bank-related debug messages instead of actual numbers.
• /togbank hello crash — Fixed crash when running from source due to unsubstituted @project-version@ placeholder.
• Slot counts showing 0/0 — Fixed slot count display for non-banker characters (UI-001).
• Gold icon broken — Fixed broken gold icon in the inventory UI (UI-002).
• Zone transition errors — Fixed ChatThrottleLib errors during zone transitions (PERF-021).
• Delta sync crash — Fixed delta syncs missing slot metadata and a nil crash on Info access (SLOTS-001, SYNC-015).

🗑️ Removed

• /togbank requestlog command — superseded by the Requests UI tab. The command had no handler and crashed on use.

---

Unreleased - Hash Sync Fixes

Status: In Development Priority: CRITICAL

🐛 Bug Fixes

[PERF-021] Eliminated ChatThrottleLib Errors During Zone Transitions (CRITICAL)

• FIXED: Added 2.5s zone-in cooldown period to defer expensive operations and give ChatThrottleLib breathing room
• PROBLEM: Despite PERF-019 (guard overlapping roster refreshes) and PERF-020 (batch hash broadcasts), users still got "script ran too long" errors when zoning. ChatThrottleLib itself (Despool function) + Bagnon + other addons exceeded cumulative execution budget during zone transitions
• ROOT CAUSE: When zoning with large message backlog (180+ queued messages from ongoing delta sends taking 26+ seconds), ChatThrottleLib's Despool() must process queue during zone transition (execution-budget-constrained window). If addon operations compete for budget (GUILD_ROSTER_UPDATE, periodic timer broadcasts, hash processing, Bagnon UI updates), cumulative execution exceeds limit
• IMPACT:
  • Intermittent "script ran too long" errors in ChatThrottleLib.lua:415 (Despool) when zoning mid-send
  • Bagnon execution errors (Script from "Bagnon" has exceeded its execution time limit)
  • Stuttering/freezing during zone transitions from budget exhaustion
  • Errors persisted even after PERF-019/020 because ChatThrottleLib needed isolation to drain queue
• BEHAVIOR:
  • User zones while 65-chunk send in progress (26+ seconds total) → 30-40 chunks still queued in CTL → 180+ pipe entries
  • PLAYER_ENTERING_WORLD fires → GUILD_ROSTER_UPDATE (even with PERF-019 guard) → other addons process events
  • Periodic OnShareTimer fires during zone window (3-minute cycle can coincide) → broadcasts compete with CTL
  • Guild members broadcast hashes during login waves (even with PERF-020 batching) → responses add CTL traffic
  • CTL Despool + our operations + Bagnon/other addons = cumulative budget exceeded
• SOLUTION:
  1. Add zoningCooldown flag set to true on PLAYER_ENTERING_WORLD (ANY type: login, reload, OR zone change)
  2. Clear flag after 2.5 seconds via C_Timer.After (gives CTL breathing room)
  3. Guard OnShareTimer: if zoningCooldown active, defer and reschedule without processing
  4. Guard SyncDeltaVersion: block BULK priority (timer-based) during cooldown, allow NORMAL priority (login broadcasts)
  5. Guard GUILD_ROSTER_UPDATE login broadcasts: if cooldown active when roster init completes, reschedule QueryRequestsIndex + SyncDeltaVersion for 2.6s
  6. Guard share-request handler: return early if zoningCooldown active
  7. Guard /togbank share command: defer 2.6s and warn user if zoningCooldown active
• RESULT:
  • ChatThrottleLib gets 2.5s breathing room after ANY world entry to drain queue without competition
  • Periodic broadcasts deferred until cooldown expires (timer reschedules normally)
  • Login broadcasts still execute but deferred by 2.6s if cooldown hasn't expired yet
  • Share requests deferred (both incoming and manual) to avoid adding work during cooldown
  • No functional impact: P2P collect windows are 60s, 2.5s delay = 4% increase (negligible)
  • Execution errors eliminated during zone transitions
  • CTL Despool operates in isolation during critical zone-in window
• LOCATION:
  • Events.lua PLAYER_ENTERING_WORLD (~284-310): zoningCooldown init + 2.5s timer
  • Events.lua OnShareTimer (~169-177): defer check + early return + reschedule
  • Events.lua SyncDeltaVersion (~252-260): priority filter (BULK blocked, NORMAL allowed)
  • Events.lua GUILD_ROSTER_UPDATE (~350-370): deferred login broadcast reschedule (2.6s)
  • Chat.lua share-request handler (~1850-1858): defer check + early return
  • Chat.lua share command (~2086-2104): defer with warning + 2.6s reschedule

[PERF-020] Eliminated Stuttering from Synchronous Hash Broadcast Processing (CRITICAL)

• FIXED: Hash-list broadcasts now batched with 0.15s timer to prevent main thread blocking during sync storms
• PROBLEM: Hash-list broadcasts from guild members processed immediately and synchronously when received
• ROOT CAUSE: When 4+ broadcasts arrived within seconds (common during login waves, zone changes, raids), each triggered 36 hash comparisons (timestamp lookups + HasAltContent checks + hash building). 4 broadcasts = 144 hash comparisons blocking main thread for 100-300ms = visible stuttering during gameplay
• IMPACT: Stuttering compounded when player simultaneously sending data (ChatThrottleLib chunk callbacks processing) or handling other sync operations. Game visibly froze during hash broadcast storms
• BEHAVIOR:
  • Player zones/logs in → guild members broadcast hashes
  • 4 broadcasts arrive within 2 seconds
  • Each processed immediately: 36 hash comparisons synchronously
  • Total: 144 comparisons + ongoing ChatThrottleLib callbacks = main thread blocked
  • Result: 100-300ms freeze, visible stuttering in game
• SOLUTION:
  1. Add batching queue and timer to Chat module (hashBroadcastQueue, hashBroadcastTimer, 0.15s delay)
  2. First broadcast starts 0.15s timer, additional broadcasts within window get queued
  3. When timer fires, process all queued broadcasts in one deferred operation
  4. Automatic sender deduplication (if same sender broadcasts twice, process most recent)
  5. ProcessQueuedHashBroadcasts() handles batch processing with performance timing
• RESULT:
  • Spreads work across multiple frames (~6 frames at 60 FPS with 0.15s delay)
  • Prevents overlapping hash comparison operations (no more synchronous storms)
  • Adds only 0.25% latency to 60s P2P collect window (150ms / 60,000ms = negligible)
  • Eliminates stuttering during sync storms (4+ broadcasts batched into single deferred operation)
  • Reduces network congestion (responses staggered by 0-150ms instead of all at once)
• LOCATION:
  • Chat.lua Init() (~23-26): Added hashBroadcastQueue, hashBroadcastTimer, HASH_BROADCAST_BATCH_DELAY variables
  • Chat.lua ProcessQueuedHashBroadcasts() (~569-653): New batch processor with deduplication, timing, queue management
  • Chat.lua OnCommReceived() hash-list-broadcast handler (~1820-1847): Replaced synchronous processing with queueing logic

[PERF-019] Prevented Overlapping Roster Refresh Operations During Zone Changes (CRITICAL)

• FIXED: GUILD_ROSTER_UPDATE handler now guards against concurrent execution to prevent cascading expensive operations
• PROBLEM: When player zoned while sending data (65-chunk ChatThrottleLib queue processing) or during roster refresh retries, WoW fires GUILD_ROSTER_UPDATE automatically during zone transition
• ROOT CAUSE: No guard flag preventing concurrent roster refresh operations. If needsFullRosterRefresh=true from previous retry attempt (roster API returned 0 members), GUILD_ROSTER_UPDATE would start another 0.5s deferred operation even if previous one still in progress
• IMPACT: Explained intermittent "execution limit exceeded" errors (1 in 5 login/reload/zone) - only occurred when multiple expensive operations overlapped during critical init/zone window
• BEHAVIOR:
  • Zone while sending → GUILD_ROSTER_UPDATE fires → starts InvalidateBanksCache + RefreshOnlineCache (loops all members) + RebuildBankerRoster (scans all notes)
  • If roster API slow, retry flag = true → next GUILD_ROSTER_UPDATE starts ANOTHER 0.5s timer
  • Multiple overlapping timers + ChatThrottleLib chunk callbacks (30 failures × callback overhead) = cumulative execution limit exceeded
• SOLUTION:
  1. Add refreshInProgress guard flag alongside existing needsFullRosterRefresh flag
  2. Check needsFullRosterRefresh AND NOT refreshInProgress before starting refresh cycle
  3. Set refreshInProgress = true when starting 0.5s deferred work
  4. Clear refreshInProgress = false at end of deferred function (after retry check, broadcasts)
  5. Log skip message when GUILD_ROSTER_UPDATE arrives while refresh in progress
• RESULT:
  • Maximum one roster refresh operation in-flight at any time
  • Prevents cascading retries when GUILD_ROSTER_UPDATE fires during zone transitions
  • Eliminates overlap with ChatThrottleLib chunk processing during sends
  • Combined with PERF-008/013/014/015/016/017/018: Should eliminate execution limit errors completely
• LOCATION:
  • Events.lua GUILD_ROSTER_UPDATE() (~305-360): Added refreshInProgress guard check at start, set flag before deferral, clear flag in deferred completion

[PERF-018] Deferred Debug Options UI Creation Until Needed (MEDIUM)

• FIXED: Debug options UI no longer built at addon load - created lazily on first access
• PROBLEM: BuildDebugArgs() created ~55 AceConfig entries (15 categories, 28 sub-tags, 12 headers/buttons) during Options:Init() at addon load
• ROOT CAUSE: Debug tab args field called BuildDebugArgs() directly in options table definition, executing at addon start regardless of whether user ever opens Debug tab
• IMPACT: Contributed to intermittent "execution limit exceeded" errors (1 in 5 login/reload/zone changes) when combined with other init operations (GUILD_ROSTER_UPDATE retries, event registration, AceComm prefixes)
• BEHAVIOR: Every login/reload created all debug category/tag options UI (~55 elements), even though most users never open Debug tab
• SOLUTION:
  1. Change debug tab args from direct BuildDebugArgs() call to lazy function
  2. Function checks TOGBankClassic_Options.debugArgsBuilt flag
  3. First access builds UI and caches result; subsequent accesses return cached table
• RESULT:
  • Default (never open Debug tab): Zero options UI creation overhead at addon load
  • First time opening Debug tab: Builds ~55 entries and caches
  • Subsequent opens: Instant (reuses cache)
  • Combined with PERF-008/013/014/015/016/017: Reduced intermittent execution limit errors
• LOCATION:
  • Options.lua Init() debug tab definition (~372-384): Changed args from BuildDebugArgs() call to lazy function with caching

[PERF-017] Cleaned Up Unnecessary SavedVariables Persistence (LOW)

• FIXED: Removed unnecessary SavedVariables from being persisted to disk
• PROBLEM 1: TOGBankClassicIcon was declared in SavedVariables but always nil (never used)
• PROBLEM 2: TOGBankClassic_PerfMetrics persisted ~400 lines per session even when TOGBankClassic_PerfEnabled = false
• PROBLEM 3: TOGBankClassic_MailDebugLog was developer debugging code left in production, persisting at every logout
• ROOT CAUSE: SavedVariables list included unused/debug variables; perf metrics didn't clear when disabled
• IMPACT: Unnecessary data written to SavedVariables on every logout, increasing file size and parse time
• SOLUTION:
  1. Removed TOGBankClassicIcon from .toc SavedVariables declaration
  2. Set TOGBankClassic_PerfMetrics = nil when performance tracking is disabled (similar to PERF-014/016 pattern)
  3. Removed TOGBankClassic_MailDebugLog entirely - deleted from .toc and removed entire debug block from PLAYER_LOGOUT
• RESULT:
  • TOGBankClassicIcon: No longer persisted (was always nil anyway)
  • PerfMetrics: Only persists when explicitly enabled; cleared to nil when disabled
  • MailDebugLog: Completely removed (was dev debugging, not production feature)
  • File size savings: ~400-600 lines removed from SavedVariables for typical user
• LOCATION:
  • TOGBankClassic.toc (~7): Removed TOGBankClassicIcon and TOGBankClassic_MailDebugLog from SavedVariables
  • Performance.lua Initialize() (~42): Added TOGBankClassic_PerfMetrics = nil when disabled
  • Events.lua PLAYER_LOGOUT (~278-345): Removed entire MailDebugLog collection block

[PERF-016] Fixed Performance Tracking Initialization Overhead (LOW)

• FIXED: Performance tracking now skips all initialization when disabled (default)
• PROBLEM: Performance:Initialize() always created sessions, ran GC, initialized data structures even when TOGBankClassic_PerfEnabled = false (default)
• ROOT CAUSE: Initialize() didn't check enabled flag before expensive operations
• IMPACT: Session creation, table inserts, GC loops for a disabled debugging feature on every addon load
• BEHAVIOR: Initialize() checked enabled flag AFTER creating tables and running GC
• SOLUTION:
  1. Move TOGBankClassic_PerfEnabled nil check to top of Initialize()
  2. Early return if TOGBankClassic_PerfEnabled is false
  3. Only create tables, run GC, and initialize session when explicitly enabled
• RESULT:
  • Default (disabled): Zero overhead - no tables created, no GC, no session initialization
  • When enabled: Normal behavior - creates sessions, tracks metrics, runs GC
  • Load time savings: Skip all performance tracking initialization for 99% of users
• LOCATION:
  • Performance.lua Initialize() (~33-48): Reordered to check flag first, early return if disabled

[PERF-015] Fixed UI Frame Creation During Addon Load (MEDIUM)

• FIXED: Inventory, Donations, Mail, Search, and Requests windows now defer frame creation until first Open()
• PROBLEM: All 5 UI modules called DrawWindow() during Init() at addon load, creating AceGUI frames (windows, buttons, scrollframes, etc.) that may never be opened
• ROOT CAUSE: Init() unconditionally called DrawWindow() instead of deferring to first use
• IMPACT: Frame creation overhead at load for 5 windows that most users won't open every session
• BEHAVIOR: Init() → DrawWindow() → AceGUI:Create("Frame") + create all child widgets → frames held in memory all session
• LAZY PATTERN: All 5 modules already had if not self.Window then self:DrawWindow() end check in Open() function
• SOLUTION:
  1. Remove DrawWindow() call from Init() in Inventory, Donations, Mail, Search, Requests
  2. Rely on existing lazy initialization check in Open() function: if not self.Window then self:DrawWindow() end
  3. Frame creation only happens when user actually opens the window
• RESULT:
  • On load: Init() does nothing (or just initializes state variables), zero frame creation
  • First Open(): Checks if not self.Window, calls DrawWindow(), caches frame
  • Subsequent opens: Reuses cached frame
  • Load time savings: Skip creating 5 AceGUI windows + all child widgets unless actually used
  • Memory savings: Don't hold unused frame objects in memory
• LOCATION:
  • UI/Inventory.lua Init() (~3-5): Removed DrawWindow() call, added PERF-015 comment
  • UI/Donations.lua Init() (~3-5): Removed DrawWindow() call, added PERF-015 comment
  • UI/Mail.lua Init() (~3-5): Removed DrawWindow() call, added PERF-015 comment
  • UI/Search.lua Init() (~3-5): Removed DrawWindow() call, added PERF-015 comment
  • UI/Requests.lua Init() (~365-371): Removed DrawWindow() call, added PERF-015 comment

[PERF-014] Fixed Persistent Debug Log Loading Unconditionally (HIGH)

• FIXED: Persistent debug log (SavedVariables) now only loads/saves when explicitly enabled by user (OFF by default)
• PROBLEM: Output:Init() always loaded TOGBankClassicDB_DebugLog from SavedVariables (up to 50,000 entries) even though persistent logging was disabled by default
• ROOT CAUSE: Load/save logic didn't check TOGBankClassic_DebugLogEnabled flag before accessing SavedVariables
• IMPACT: Every player paid the cost of parsing 1-5 MB of debug log entries on every reload, plus garbage collection loop checking 50k timestamps, even though 99% never enable persistent logging
• BEHAVIOR: Init() unconditionally loaded log → GarbageCollectPersistentLog() looped through all entries → held 50k entries in memory all session
• CLARIFICATION: Regular debug logging (showing messages in chat/debug frame) works independently from persistent logging (saving to SavedVariables). Debug categories control what's shown; persistent logging checkbox controls whether messages are also saved to disk.
• SOLUTION:
  1. Check TOGBankClassic_DebugLogEnabled in Init() - only load if true
  2. Check TOGBankClassic_DebugLogEnabled in SavePersistentLog() - only save if true
  3. Check TOGBankClassic_DebugLogEnabled in AddToPersistentLog() - only add if true
  4. Updated option description to clarify persistent logging is separate from regular debug message display
• RESULT:
  • Default (disabled): Debug messages still show in chat but NOT saved to SavedVariables. Zero parsing overhead, zero GC, empty log array
  • When enabled: Debug messages shown in chat AND saved to SavedVariables for later review via /togbank debuglog
  • Load time savings: Skip parsing 1-5 MB of text and GC loop on every reload for 99% of users
  • Memory savings: Don't hold 50k log entries unless explicitly enabled
• LOCATION:
  • Output.lua Init() (~72-89): Added TOGBankClassic_DebugLogEnabled check before loading
  • Output.lua SavePersistentLog() (~456-470): Added early return if disabled
  • Output.lua AddToPersistentLog() (~418-433): Added early return if disabled
  • Options.lua debugLogEnabled (~302-324): Updated to clarify persistent logging vs regular debug messages

[PERF-013] Fixed ChatThrottleLib Timeout from Zone Change Message Spam (CRITICAL)

• FIXED: PLAYER_ENTERING_WORLD now only triggers roster refresh + broadcasts on login/reload, NOT on zone changes
• PROBLEM: "Script ran too long" in ChatThrottleLib.lua:389 during zone changes
• ROOT CAUSE: PLAYER_ENTERING_WORLD fires on EVERY zone change, triggering 2 GUILD broadcasts per player (togbank-r + togbank-hl)
• IMPACT: When 40+ players enter MC/BWL simultaneously, 80+ messages queue instantly and overwhelm ChatThrottleLib's message queue
• BEHAVIOR: Zone changes → PLAYER_ENTERING_WORLD → needsFullRosterRefresh=true → GUILD_ROSTER_UPDATE → QueryRequestsIndex + SyncDeltaVersion → 2 GUILD broadcasts
• CUMULATIVE EFFECT: Every player in guild broadcasts on every zone, causing "bMyTraffic = true" loops in ChatThrottleLib Despool to exceed execution time
• SOLUTION:
  1. Check isInitialLogin and isReloadingUi parameters from PLAYER_ENTERING_WORLD event
  2. Only set needsFullRosterRefresh = true on login/reload (not zone changes)
  3. Zone changes skip roster refresh entirely (no broadcasts)
  4. OnShareTimer still broadcasts every 10 minutes (periodic sync unaffected)
• RESULT:
  • Login/Reload: Normal behavior (2 GUILD broadcasts per player)
  • Zone Change: Zero broadcasts (no ChatThrottleLib queue spam)
  • Raid Entry: No more timeout errors when 40 players zone together
  • Data Freshness: OnShareTimer maintains 10-minute sync cycle
• LOCATION:
  • Events.lua PLAYER_ENTERING_WORLD (~348-367): Added isInitialLogin/isReloadingUi check
  • Events.lua GUILD_ROSTER_UPDATE (~362-413): Only fires deferred block when needsFullRosterRefresh=true

[PERF-008] Fixed Bagnon Execution Timeout from BAG_UPDATE Spam (CRITICAL)

• FIXED: ItemHighlight now registers BAG_UPDATE events ONLY when highlighting is actively enabled by a banker (on-demand registration)
• PROBLEM: Bagnon exceeded execution time limit during zone changes, affecting even non-banker characters
• ROOT CAUSE 1: BAG_UPDATE events were processed by ALL players during zone changes (50+ events in 0.2 seconds)
• ROOT CAUSE 2: Each addon with BAG_UPDATE handlers adds to cumulative execution budget, even if handler does nothing
• ROOT CAUSE 3: Highlighting doesn't need ANY events until a banker explicitly enables it, but events were registered at addon load
• IMPACT: "Script from Bagnon has exceeded its execution time limit" errors on zone change for ALL guild members
• BEHAVIOR: Zone changes fire 50+ rapid BAG_UPDATE events; cumulative processing across all addons exceeded Bagnon's execution time limit
• SOLUTION:
  1. CRITICAL FIX: Don't register ANY events at Initialize() - highlighting doesn't need them yet
  2. When banker clicks "Enable Highlighting" in Requests tab, check banker status and register events
  3. When banker clicks "Disable Highlighting", unregister ALL events
  4. Result: Zero overhead for everyone until feature is actively used
  5. Events include throttling (500ms) and search string caching when registered
  6. Prevents rapid-fire Bagnon UI rebuilds during zone transitions
• RESULT:
  • Non-bankers: NEVER register BAG_UPDATE events (zero overhead forever)
  • Bankers with highlighting disabled: Zero overhead (same as non-bankers)
  • Bankers with highlighting enabled: Events registered on-demand with throttling + caching
  • Guild-wide: Eliminated Bagnon execution timeout errors completely
• WHY THIS IS THE CORRECT APPROACH: Highlighting only needs to work when actively fulfilling orders. There's no reason to have ANY event handlers registered during normal gameplay, at addon load, or during zone changes. On-demand registration = zero overhead until actually needed.
• COMPARISON TO PREVIOUS APPROACHES:
  • v1: Registered events for all, checked self.enabled → all players processed 50+ events
  • v2: Early exit in Initialize if not banker → broke highlighting (guild data not loaded)
  • v3: Lazy check on first event, then unregister → non-bankers still processed first batch
  • v4: Wait for GUILD_ROSTER_UPDATE, then register → still registered events before needed
  • v5 (FINAL): On-demand registration only when highlighting enabled → zero overhead until used
• LOCATION:
  • ItemHighlight.lua registerBagEvents (~23-59): Register BAG_UPDATE events when highlighting enabled
  • ItemHighlight.lua unregisterBagEvents (~62-74): Unregister events when highlighting disabled
  • ItemHighlight.lua Initialize (~77-84): Minimal initialization, no event registration
  • ItemHighlight.lua SetEnabled (~87-136): Check banker status, register/unregister events based on enabled state

[HASH-001] Fixed Hash Broadcast Not Triggering P2P Requests (CRITICAL)

• FIXED: hash-list-broadcast handler now triggers P2P requests for changed data
• PROBLEM: /togbank share only updated latestBankerHashes cache without triggering any data requests
• IMPACT: Complete sync failure - receivers saw "Updated banker hashes" but never requested changed data
• BEHAVIOR: Users had to manually run /togbank sync or wait 3 minutes for auto-sync timer
• ROOT CAUSE: hash-list-broadcast (togbank-hl) only cached hashes, while hash-list-reply (togbank-hlr) also compared and broadcast P2P
• INCONSISTENCY: Same data (hash updates) handled differently depending on source (broadcast vs reply)
• SOLUTION:
  • hash-list-broadcast handler updates latestBankerHashes cache (immediate)
  • Converts message to hash-list-reply format
  • Recursively calls OnCommReceived with "togbank-hlr" prefix to reuse existing logic
  • hash-list-reply handler now updates cache incrementally (not replacement) to support partial broadcasts
  • Avoids code duplication - single code path for all hash processing
• RESULT: /togbank share now triggers immediate P2P broadcasts, data syncs automatically
• LOCATION: Modules/Chat.lua (~1773-1800, ~1816-1828): Forward broadcast to reply handler, incremental cache update
• NOW: Hash broadcasts work identically whether from /togbank share (broadcast) or /togbank sync (reply)

[COMM-003] Fixed Offline Player Detection from Whisper Errors

• FIXED: CHAT_MSG_SYSTEM now detects "No player named X is currently playing" errors
• PROBLEM: Addon repeatedly attempted whispers to offline players causing error spam
• ROOT CAUSE: CHAT_MSG_SYSTEM handler only detected "has gone offline" but not whisper failure errors
• IMPACT: Hundreds of error messages when trying to communicate with offline players
• SOLUTION:
  • Added pattern matching for "No player named X is currently playing" in CHAT_MSG_SYSTEM
  • When detected, immediately marks player as offline in both onlineMembers and recentlySeen caches
  • Updated UpdateOnlineMember() to clear recentlySeen cache when marking offline
  • Added debug logging for all online/offline state changes
• RESULT: Player marked offline immediately when whisper fails, preventing repeat attempts
• LOCATION:
  • Events.lua CHAT_MSG_SYSTEM (~334-375): Added error pattern detection
  • Guild.lua UpdateOnlineMember (~1425-1443): Clear recentlySeen on offline

[COMM-003b] Fixed Whisper Error Pattern Not Matching Single-Quoted Names

• FIXED: CHAT_MSG_SYSTEM now detects both single-quoted and unquoted variants of whisper failure messages
• PROBLEM: Pattern only matched No player named Axkva is currently playing. but Classic Era can also send No player named 'Axkva' is currently playing. (with single quotes around name)
• ROOT CAUSE: COMM-003 documentation incorrectly stated "Classic Era does NOT use quotes" but testing showed single quotes are sometimes used around player names
• IMPACT: Whisper failures with single-quoted names not detected, causing repeated whisper attempts and error spam
• SOLUTION:
  • Added dual pattern matching: tries single-quoted pattern '(.+)' first, falls back to unquoted (.+) if no match
  • Updated documentation to reflect both formats are possible
• RESULT: All whisper failure formats now detected, player marked offline immediately
• LOCATION:
  • Events.lua CHAT_MSG_SYSTEM (~353-361): Dual pattern matching
  • DELTA_BUGS.md (~2521-2527): Updated pattern documentation

[COMM-003c] Fixed Whisper Error Messages Still Appearing in Chat

• FIXED: Added chat message filter to suppress "No player named X is currently playing" errors from appearing in chat
• PROBLEM: While COMM-003/COMM-003b fixed offline detection, the error messages still appeared in chat window
• ROOT CAUSE: Event handler detected and processed errors but didn't suppress chat display
• IMPACT: Chat spam with "No player named 'X' is currently playing" even though addon correctly marked players offline
• SOLUTION:
  • Added ChatFrame_AddMessageEventFilter for CHAT_MSG_SYSTEM
  • Filter returns true to suppress any message matching "No player named .+ is currently playing."
  • Works for both single-quoted and unquoted name variants
• RESULT: Error messages silently handled - player marked offline without chat spam
• LOCATION:
  • Events.lua RegisterEvents (~59-66): Added chat message filter

[PERF-009] Fixed ChatFrame_AddMessageEventFilter Performance Issue

• FIXED: Optimized chat filter to use fast plain-text check before pattern matching
• PROBLEM: Stuttering during gameplay immediately after adding chat filter in COMM-003c
• ROOT CAUSE: Pattern match ran on EVERY CHAT_MSG_SYSTEM event (guild achievements, player online/offline, etc)
• IMPACT: 20-30 expensive pattern matches per minute causing ~10-30ms cumulative frame time waste
• SOLUTION:
  • Added fast plain-text prefix check: find("No player named ", 1, true) (~0.01ms)
  • Pattern match only runs if prefix found (rare - only actual whisper errors)
  • 99%+ of events skip expensive pattern matching
• RESULT: 50x performance improvement, stuttering eliminated while maintaining error suppression
• LOCATION:
  • Events.lua Initialize (~59-68): Added plain-text prefix check before pattern match

[PERF-010] Fixed Login Freeze from Synchronous Data Migrations

• FIXED: Deferred Database:Load() migrations and hash cache initialization to eliminate 3-5 second freeze on login/reload
• PROBLEM: Game completely froze for 3-5 seconds when logging in or reloading UI with large SavedVariables (70+ alts)
• ROOT CAUSE: Database:Load() synchronously looped through ALL alts performing migrations on EVERY login - most expensive was RecalculateAggregatedItems() for each banker alt (~30-50ms per alt)
• IMPACT: Cannot move, cast spells, or interact during freeze - appeared as if game crashed
• WHY DEFERRABLE: Data already loaded from SavedVariables, migrations are cleanup/optimization operations that don't need to be immediate
• SOLUTION:
  • Wrapped entire Database:Load() migration block in C_Timer.After(0.5)
  • Also deferred latestBankerHashes initialization in Guild:Init (lighter but still blocking)
  • Migrations run in background after UI becomes responsive
  • Still complete before first UI interaction or sync
• RESULT: Instant login, no freeze, migrations ready before first use
• LOCATION:
  • Database.lua Load (~175-243): Deferred migration block
  • Guild.lua Init (~295-313): Deferred latestBankerHashes initialization
• LOCATION:
  • Events.lua Initialize (~59-68): Added plain-text prefix check before pattern match

[COMM-003d] Fixed recentlySeen Cache Undermining Guild Roster Cache (CRITICAL)

• FIXED: Removed recentlySeen cache, IsPlayerOnline now uses only guild roster cache
• PROBLEM: Addon still tried to whisper players for 5 minutes after they logged off
• ROOT CAUSE: IsPlayerOnline checked both onlineMembers (accurate) and recentlySeen (5-minute stale cache)
• FLOW: Player sends message → added to recentlySeen → logs off → onlineMembers cleared correctly → but recentlySeen keeps them "online" for 5 minutes → whispers sent → errors
• IMPACT: Despite accurate guild roster updates (COMM-003), whisper errors still occurred due to secondary stale cache
• WHY IT EXISTED: Originally added for "cross-realm/cross-guild" players, but this is a guild-only addon
• SOLUTION:
  • Removed recentlySeen check from IsPlayerOnline - guild roster cache is single source of truth
  • Removed MarkPlayerSeen call in Chat.lua
  • MarkPlayerSeen() kept as no-op for backwards compatibility
• RESULT: IsPlayerOnline returns false immediately when player logs off, no stale 5-minute window
• LOCATION:
  • Guild.lua IsPlayerOnline (~1542-1547): Removed recentlySeen logic, only check onlineMembers
  • Guild.lua MarkPlayerSeen (~1533-1537): Made no-op
  • Chat.lua OnCommReceived (~617-619): Removed MarkPlayerSeen call

[DELTA-020] Fixed Delta Computation Using Wrong Baseline (CRITICAL)

• FIXED: ComputeDelta now uses requester's actual item structures from state summary instead of responder's snapshot
• PROBLEM: When responder broadcast multiple times (hash 461905621 → 317352773), GetSnapshot returned responder's NEW snapshot (317352773) instead of requester's OLD baseline (461905621)
• IMPACT: Item count duplication/corruption - delta computed as (317352773 - 317352773) = empty/minimal instead of (317352773 - 461905621) = proper changes
• BEHAVIOR: Requester with hash 461905621 applied incorrect delta to their old data, causing items to double instead of updating
• ROOT CAUSE:
  • Snapshot system stores ONE snapshot per alt (keyed only by altName, not hash)
  • When responder broadcasts twice, old snapshot overwritten (461905621 deleted, only 317352773 remains)
  • State summary previously sent aggregated items {[itemID] = count} (useless for delta computation)
  • ComputeDelta used GetSnapshot(altName) which returned responder's OWN latest state, not requester's actual baseline
• DESIGN ISSUE: Delta needs requester's actual item structure (bank/bags/mail) to compute delta = current - requester's baseline
• SOLUTION:
  1. Modified ComputeStateSummary to send minimal item structures {ID, Count} (no Links) for separate bank/bags/mail arrays
  2. Modified ComputeDelta to accept optional requesterBaseline parameter with minimal structures
  3. ComputeDelta now uses requester's sent baseline as previous instead of GetSnapshot when available
  4. RespondToStateSummary extracts bank/bags/mail from state summary and passes through SendAltData → ComputeDelta chain
  5. SendAltData signature updated to accept and forward requesterBaseline parameter
  6. All SendAltData call sites updated to pass baseline (or nil for legacy paths)
• BANDWIDTH SAVINGS: Minimal structures ~1-2KB vs full with Links ~20-50KB (~85% reduction)
• RESULT: Delta computation uses correct baseline - requester's actual data (what they have) vs responder's current (what to send)
• LOCATIONS:
  • Guild.lua ComputeStateSummary (~1485-1542): Send bank/bags/mail arrays with {ID, Count} only
  • Guild.lua SendStateSummary (~1593-1606): Updated logging to count bank+bags+mail items
  • Guild.lua RespondToStateSummary (~1640-1644): Extract requesterBaseline from state summary
  • Guild.lua SendAltData (~2231, ~2301): Added requesterBaseline parameter
  • Guild.lua ComputeDelta wrapper (~2862): Pass through requesterBaseline
  • DeltaComms.lua ComputeDelta (~565-620): Accept requesterBaseline, use expandMinimalItems() helper, compute from requester's actual data
• NOW: Proper delta application - item counts update correctly without duplication, works regardless of how many broadcasts requester missed
• RELATED: Completes DELTA-019 fix - hash stays at local value until correct delta (computed from actual baseline) received and applied

[DELTA-019] Fixed Premature Hash Update Before Data Received (CRITICAL)

• FIXED: Removed HLR first pass branch that updated hash when localHash == 0
• PROBLEM: Hash updated from banker's broadcast before delta data arrived and was applied
• IMPACT: Data/hash desynchronization - old data stored with new hash value
• BEHAVIOR: User has hash 529743613 with data, banker broadcasts 461905621, hash immediately updates to 461905621 but old data remains
• CONSEQUENCE: Future sync attempts see matching hashes and skip update, leaving permanent stale data until /wipe command
• ROOT CAUSE: HLR first pass had three branches:
  1. if not localAlt - Create new stub with banker's hash (CORRECT - for brand new alts)
  2. elseif localHash == 0 - Update existing alt's hash (BUG - fires during pending sync)
  3. elseif mismatch - Log mismatch without updating (CORRECT)
• TRIGGER SCENARIO: Between HLR broadcasts, localHash becomes 0 (ApplyDelta creates stub, or other process clears it), next HLR hits branch 2 and updates hash prematurely
• USER SCENARIO: "i have hash 5xxxx with data, banker sends hash 4xxxx, I should use hash 4xxxx to trigger delta sync comparing hash 5xx with 4xx to determine the delta, and then get new data"
• ACTUAL BUG BEHAVIOR: Hash 5xxxx → 4xxxx update happens immediately in HLR first pass, before delta data request sent/received
• SOLUTION: Removed branch 2 entirely from HLR first pass (Chat.lua lines 1861-1873)
  • Only create NEW stubs for brand new alts (branch 1)
  • Only update hash in ApplyDelta after successful data application (DeltaComms.lua:971)
  • Hash mismatch detection (branch 3) triggers requests without updating hash
• RESULT: Hash stays at local value until delta received and applied, maintaining data/hash consistency
• LOCATION: Chat.lua HLR handler first pass (~1847-1878): Removed elseif localHash == 0 branch
• RELATED: Works with DELTA-018 fix - latestBankerHashes cache tracks "what banker says", local inventoryHash tracks "what data we have", only ApplyDelta updates local hash

[DELTA-018] Fixed Hash Broadcast Circular Comparison (CRITICAL)

• FIXED: Hash sync protocol now maintains separate in-memory cache from local storage
• PROBLEM: hash-list-broadcast immediately updated local alt.inventoryHash, then comparison read from same local storage
• IMPACT: Complete sync failure - /togbank share broadcasts updated hash without triggering sync requests
• BEHAVIOR: Receivers had stale data but /togbank hashdebug showed "matched" (compared local hash against itself)
• ROOT CAUSE: Circular comparison - BuildBankerHashList() read from alt.inventoryHash, ReportHashListCoverage compared alt.inventoryHash vs BuildBankerHashList() output (both same source)
• DESIGN FLAW: No separation between "banker's authoritative hash" (what they broadcast) vs "hash of data we actually have" (what's in SavedVariables)
• SOLUTION:
  • Initialize latestBankerHashes in-memory cache on addon load from all local alt.inventoryHash values
  • hash-list-broadcast handler: Only update cache, never modify local storage
  • hash-list-reply handler: Only update cache on mismatch, never modify local storage
  • ReportHashListCoverage: Use latestBankerHashes exclusively (no BuildBankerHashList, no merge)
  • Local alt.inventoryHash: Only updated when actual delta data received and applied
• CACHE STRUCTURE: {hash, updatedAt, version, mailHash, mailUpdatedAt} per alt
• COMPARISON LOGIC: cache.hash ("what banker says") vs localAlt.inventoryHash ("what we have")
• RESULT: Proper mismatch detection - cache updated by broadcasts, local unchanged until delta received, comparison detects staleness
• LOCATION:
  • Guild.lua Init (~276-290): Initialize latestBankerHashes from SavedVariables
  • Guild.lua ReportHashListCoverage (~693-710): Use cache directly for comparison
  • Chat.lua hash-list-broadcast (~1773-1795): Update cache only
  • Chat.lua hash-list-reply (~1843-1847): Update cache only on mismatch
• VERIFICATION: Tested with manual hash revert - broadcast updated cache, local stayed stale, hashdebug showed pending, sync requested delta

[DELTA-016] Fixed Delta Protocol Sending Aggregated Items (CRITICAL)

• FIXED: ComputeDelta now sends separate bank/bags/mail inventories instead of aggregated items
• PROBLEM: Used alt.items (UI display field) which was often empty on sender side despite non-zero hash
• IMPACT: Complete data sync failure - deltas contained only money updates, no item data
• BEHAVIOR: Debug showed "hasChanges.items=false, itemCount=0" with non-zero inventoryHash (contradiction)
• ROOT CAUSE: alt.items computed during Bank:Scan() for UI aggregation, not guaranteed during delta computation
• PROTOCOL DESIGN: Should send bank/bags/mail separately so receiver populates individual inventories
• SOLUTION:
  • ComputeDelta: Source from currentAlt.bank.items, bags.items, mail.items separately
  • ApplyDelta: Apply to current.bank.items, bags.items, mail.items individually
  • Recalculate aggregated current.items after delta application (UI display only)
  • DeltaHasChanges: Check bank/bags/mail separately
  • ValidateDeltaStructure: Validate mail delta
  • SanitizeDelta: Sanitize mail delta
  • StripDeltaLinks: Strip mail links
• RESULT: Deltas now contain actual item data (ID + Count) in separate bank/bags/mail structures
• LOCATION: DeltaComms.lua ComputeDelta (~627-648), ApplyDelta (~912-969), DeltaHasChanges, validation/sanitization
• NOW: Full inventory synchronization working - items populate correctly on receiver side

[DELTA-017] Fixed Empty Baseline Missing Bank/Bags/Mail Structures (CRITICAL)

• FIXED: ComputeDelta empty baseline now includes bank/bags/mail structures
• PROBLEM: Empty baseline fallback only had { items = {}, money = 0, mailHash = 0 } without bank/bags/mail
• IMPACT: First-time sync sent empty deltas despite sender having inventory data
• BEHAVIOR: ComputeDelta compared empty baseline to sender's current but both appeared empty
• ROOT CAUSE: When accessing previous.bank.items, defaulted to {} but didn't distinguish between incomplete baseline vs empty inventory
• SOLUTION:
  • Changed empty baseline to include complete structures: { items = {}, money = 0, mailHash = 0, bank = { items = {} }, bags = { items = {} }, mail = { items = {} } }
  • Fixed in 3 locations: mail-only change without snapshot, hash mismatch without snapshot, requester has no data
• RESULT: First-time sync and hash mismatch scenarios now send actual items (not empty deltas)
• LOCATION: DeltaComms.lua ComputeDelta empty baseline initialization (~594, ~606, ~613)
• NOW: All sync scenarios populate receiver correctly with sender's inventory data

[MAIL-010] Fixed Mail-Only Change Sync Abort (CRITICAL)

• FIXED: ComputeDelta now uses empty baseline fallback instead of returning nil
• PROBLEM: When mail changed but inventory matched, and no snapshot existed, returned nil (line 567)
• IMPACT: Complete sync failure - requesters with matching inventory but outdated mail never received updates
• BEHAVIOR: Guild.lua aborted sync at line 2054-2055 with "Failed to compute delta" error
• ROOT CAUSE: Inconsistent error handling - inventory mismatch used empty baseline, mail-only change returned nil
• SOLUTION: Changed to previous = { items = {}, money = 0, mailHash = 0 } (same as inventory mismatch case)
• RESULT: Mail-only changes always sync successfully via delta (contains all items as additions against empty baseline)
• LOCATION: DeltaComms.lua (~557-567): Mail-only change handler now matches inventory mismatch fallback behavior
• NOTE: Still pure delta protocol - empty baseline causes delta to include all items, but transmitted as delta message

[P2P-010] Fixed P2P Broadcast Never Sent (CRITICAL)

• FIXED: togbank-rr handler now actually sends P2P broadcast to guild
• PROBLEM: Handler built P2P request but never serialized or sent it
• IMPACT: P2P only worked when no banker online initially; failed when banker responded with hash (common case)
• BEHAVIOR: Request was built, log said "Broadcasting", but SendCommMessage was missing
• RESULT: 5-second timeout always triggered, forcing 100% fallback to banker despite peers having data
• LOCATION: Chat.lua (~1053-1054): Added missing SerializeWithChecksum and SendCommMessage calls
• NOW: Full P2P flow works - peers receive broadcasts and respond with matching hashes

[P2P-011] Fixed pendingSendCount Leak

• FIXED: Added 30-second timeout to auto-decrement counter when requester never sends state summary
• PROBLEM: Peer ACKs request and increments counter, but if requester goes offline before sending state summary, counter never decrements
• IMPACT: After 3 stuck sends, peer permanently blocks all P2P responses with "send queue full" until /reload
• BEHAVIOR: Now auto-decrements counter after 30 seconds if SendAltData never called
• RESULT: Peers self-recover from stuck sends, preventing permanent P2P queue blocking
• LOCATIONS:
  • Guild.lua (~22): Added pendingSendTimeouts tracking table
  • Chat.lua (~829-838): Added 30-second safety timeout after incrementing counter
  • Guild.lua (~1997-2000): Cancel timeout when SendAltData actually called
• NOW: Robust P2P send queue management with automatic recovery from edge cases

[P2P-012] Added Peer-Side Fallback Timeout

• FIXED: Added 15-second timeout on requester side after peer ACK
• PROBLEM: If peer ACKs but never sends data (disconnect/crash), requester waits indefinitely
• IMPACT: User must manually retry with /togbank sync
• BEHAVIOR: Now falls back to banker after 15 seconds if peer never delivers
• RESULT: Automatic recovery from peer failures without manual intervention
• LOCATION: Chat.lua (~1091-1101): Secondary timeout after clearing pending P2P request
• NOW: Full fallback chain works - peer timeout → banker fallback → data arrives

[P2P-013] Fixed expectedHashUpdatedAt Memory Leak

• FIXED: Added cleanup for expectedHashUpdatedAt after successful hash validation
• PROBLEM: Timestamps stored but never cleared, accumulating indefinitely
• IMPACT: Minor memory leak (just timestamps), no functional impact
• RESULT: Clean memory management for hash tracking
• LOCATION: Guild.lua (~2250-2252): Clear expectedHashUpdatedAt after validation

[PERF-007] Fixed GUILD_ROSTER_UPDATE Stuttering

• FIXED: Changed OR to AND logic in initialization condition to stop repeated full roster refreshes
• PROBLEM: Used fullRosterInitAttempts < 2 OR (roster incomplete) which kept triggering after initialization
• IMPACT: Every online/offline event triggered full guild roster scan (1000+ members), causing 5-10ms+ stuttering
• BEHAVIOR: After 2 initialization attempts, fullRosterInitAttempts >= 2 BUT condition stayed true due to OR
• ROOT CAUSE: Second condition totalMembers <= onlineMembers often true (WoW API reports equal values)
• SOLUTION: Changed to AND logic - only refresh if BOTH conditions true: (not initialized yet) AND (roster incomplete)
• RESULT: Full refresh only during addon load, online/offline uses lightweight CHAT_MSG_SYSTEM handler (<1ms)
• LOCATION: Events.lua (~305-313): Fixed needsFullRosterRefresh flag logic
• OPERATIONS AVOIDED: RefreshOnlineCache, RebuildBankerRoster, GetGuildRosterInfo loops, RefreshRequestsUI
• NOW: Smooth gameplay without stuttering, full scan only on joins/leaves (not online/offline)
• DOCUMENTATION: See docs/DELTA_BUGS.md for comprehensive analysis (PERF-005, PERF-007)

[DELTA-015] Fixed Delta Duplication Bug (Complete)

• FIXED: Added snapshot validation for inventory changes to prevent item duplication
• PROBLEM: When inventory changed but no snapshot existed, delta computed against empty baseline
• IMPACT: Requester would receive delta additions on top of existing stale data, causing duplicates
• BEHAVIOR: Now checks for snapshot before computing delta for both mail-only AND inventory changes
• RESULT: Forces full data (hash=0) when no snapshot available, preventing duplication
• LOCATIONS:
  • Guild.lua (~1568-1594): Mail-only change validation (previously fixed)
  • Guild.lua (~1596-1624): Inventory change validation (newly fixed)

[SYNC-009] Fixed Non-Banker Hash Sync

• FIXED: HLR handler now checks hash equality BEFORE skipping alts
• PROBLEM: Previously skipped any alt with hasContent=true without comparing hashes
• IMPACT: Non-banker updates never propagated to peers with stale data
• BEHAVIOR: Now only skips if BOTH hasContent AND hashes match
• RESULT: Non-banker-to-non-banker sync working correctly

[MAIL-009] Fixed mailHash Storage When Hashes Differ

• FIXED: HLR and HL-broadcast handlers now update mailHash when it differs from banker
• PROBLEM: Only stored mailHash when localHash=0, not when hashes differed
• IMPACT: Mail-only changes never cached banker's new mailHash
• LOCATIONS: Fixed in both HLR handler (togbank-hlr) and hash broadcast handler (togbank-hl)
• RESULT: Banker's authoritative mailHash properly cached in all scenarios

Technical Details:

// OLD: Only update when localHash == 0
elseif localHash == 0 then
    localAlt.inventoryHash = summary.hash
    if summary.mailHash then
        localAlt.mailHash = summary.mailHash  // Only here!
    end
end

// NEW: Also update when hashes differ
elseif localHash ~= summary.hash or (localAlt.mailHash or 0) ~= (summary.mailHash or 0) then
    localAlt.inventoryHash = summary.hash
    if summary.mailHash then
        localAlt.mailHash = summary.mailHash  // Now cached properly!
    end
end

Files Changed:

• Modules/Chat.lua (~1740-1753): HLR handler - added elseif block for hash diff
• Modules/Chat.lua (~1654-1679): HL broadcast handler - updated stub creation and hash updates
• docs/DELTA_BUGS.md: Documented SYNC-009 and MAIL-009 with full analysis

---

Unreleased - Hash Broadcast Improvements

Status: In Development Priority: MEDIUM

🔄 Hash Broadcasting Overhaul

Changes to /togbank share

• CHANGED: Now broadcasts hash for ONLY the current banker character (single alt)
• CHANGED: Uses togbank-hl channel for hash announcement (P2P discovery)
• CHANGED: No longer pushes full data - clients pull data via sync cycle
• IMPACT: Reduces spam when banker shares (1 hash vs 35+ data packets)

New Command: /togbank hashupdate

• NEW: Banker-only command to broadcast ALL bank alt hashes (the "nuke")
• USE CASE: Force guild-wide hash refresh after bulk inventory changes
• BEHAVIOR: Broadcasts hash-list for all bank alts on togbank-hl
• OUTPUT: "Broadcasted hash-list for N bank alts"

Hash Broadcast Enhancements

• ENHANCED: BuildBankerHashList() now includes mailHash and mailUpdatedAt
• ENHANCED: Hash-list-broadcast handler stores both inventory and mail hashes
• ENHANCED: Handler tracks what changed: "Updated AltName: inv: 123->456, mail: 789->999"
• FIXED: Non-bankers can now detect mail-only changes from hash broadcasts
• BEHAVIOR: No automatic requests triggered - users must run /togbank sync or wait for automatic sync cycle

Technical Details

• Hash broadcasts contain: inventoryHash, inventoryUpdatedAt, version, mailHash, mailUpdatedAt
• Clients update local hash stubs when received
• Actual data requests happen during next sync cycle (manual, UI open, or 3-minute timer)

Files Changed:

• Modules/Guild.lua:
  • Updated BuildBankerHashList() to include mail hashes
  • Modified Share() to broadcast single-alt hash on togbank-hl
  • Added HashUpdate() function for all-alts hash broadcast
• Modules/Chat.lua:
  • Updated hash-list-broadcast handler to store both inventory and mail hashes
  • Added /togbank hashupdate command registration
  • Enhanced hash change detection and logging

---

[v0.8.9] (2026-02-07) - P2P Hash Backfill Complete

Status: Production Ready Priority: HIGH

🎯 P2P Hash Backfill Implementation

Core Features Implemented

• NEW: Banker HLR (hash list reply) stores authoritative hashes for all roster alts
• NEW: Version broadcasts store peer hashes when local hash is missing
• NEW: 3-minute rebroadcast timer requests hash list and rebroadcasts P2P for missing alts
• NEW: Roster sync fallback for officer-note-only guilds (config option, default OFF)
• FIXED: /wipe now rebuilds banker roster immediately (progress shows 0/35 instead of 0/2)
• FIXED: Data payloads always use GUILD channel (removed WHISPER routing for togbank-d3/d4)
• FIXED: SendWhisper now correctly treats AceComm nil return as success
• FIXED: Nil table access errors in Guild.lua (table initialization before access)
• FIXED: Inventory disappearing during async item loads - added loading indicator [UI-005]
• FIXED: Severe tooltip performance issues causing PC stuttering
• ENHANCED: P2P broadcasts happen for both "pending" (hash mismatch) and "missingContent" (hash match, no data)

📝 Technical Details

Hash Storage Priority System:

1. Primary: Banker HLR (togbank-hlr) - Most authoritative

   • Banker has scanned all alts, provides definitive hash list
   • Stores hash+updatedAt for all roster alts immediately
   • Triggered on /sync, UI open, and every 3 minutes via timer
   • Implementation: Chat.lua lines 1476-1510

2. Secondary: Peer Version Broadcasts (togbank-dv2) - Supplemental

   • Fills gaps if HLR not received yet
   • Only stores hash if we don't have one (hash=0 or nil)
   • Implementation: Chat.lua lines 394-434

Result: After HLR, all 35 alts have authoritative hashes (pending=0)

Files Changed:

• Modules/Chat.lua:
  • Added hash storage in HLR handler (two-pass: store hashes, then compare)
  • Added hash storage in version broadcast handler (only if missing)
  • Added debug logging for pending/missingContent broadcasts
• Modules/Guild.lua:
  • Added RebuildBankerRoster() call to Reset() for immediate banker list after /wipe
  • Removed forceFull bypass logic (preserves P2P design)
  • Changed SendAltData() to always use GUILD channel for togbank-d3/d4
• Modules/Events.lua:
  • Added RequestHashListFromBanker() call to OnShareTimer() (3-minute rebroadcast)
• Modules/DeltaComms.lua:
  • Updated FastFillMissingAlts() to use P2P broadcasts when banker offline

How It Works:

1. On /sync or UI open: Request hash list from banker (togbank-hl)
2. Banker replies: Send all alt hashes via togbank-hlr
3. Store hashes: Create/update local alt stubs with banker's authoritative hash+updatedAt
4. Compare: Categorize as "pending" (hash mismatch) or "missingContent" (hash match, no data)
5. Broadcast: Send P2P requests to GUILD for missing alts with expectedHash
6. Peers respond: Players with matching hash send data via GUILD
7. Fallback: After 5s timeout, query banker directly via whisper
8. Rebroadcast: Every 3 minutes, repeat steps 1-7 for still-missing alts

Design Principles:

• Banker is authoritative source for hash list
• P2P broadcasts reduce banker load
• GUILD channel for data, WHISPER for handshakes
• Newest-wins conflict resolution using inventoryUpdatedAt timestamps

---

[v0.8.4] (2026-02-02) - Mail Hash Synchronization Fix

Status: Critical Bug Fix Priority: HIGH

🐛 Critical Bug Fix

[MAIL-012] Mail Hash Never Set - Fixed Mail Synchronization

• FIXED: mailHash field was referenced but never assigned, breaking mail synchronization
• ROOT CAUSE: Mail scan created mail data but never computed hash for change detection
• IMPACT: Mail items never synchronized between clients via /togbank share or /togbank sync
• FIX:
  • Bank.lua now computes mailHash after mail scan using ComputeInventoryHash()
  • DeltaComms.lua tracks mailHash changes in delta computation
  • DeltaComms.lua applies mailHash changes when receiving deltas
  • DeltaComms.lua recognizes mailHash as a valid change type
• RESULT: Mail data now properly syncs between guild members
• LOGGING: Added [MAIL-012] debug markers for mail hash operations

📝 Technical Details

Files Changed:

• Modules/Bank.lua - Added mailHash computation after mail scan (lines 289-310)
• Modules/DeltaComms.lua - Track mailHash in ComputeDelta() (lines 533-545)
• Modules/DeltaComms.lua - Apply mailHash in ApplyDelta() (lines 806-810)
• Modules/DeltaComms.lua - Recognize mailHash in DeltaHasChanges() (lines 590-593)
• docs/DELTA_BUGS.md - Comprehensive bug documentation with root cause analysis

How It Works:

1. When mail is scanned, mailHash is computed from alt.mail.items
2. mailHash changes trigger version updates in deltas
3. Receivers see mailHash and know mail data is "new format" (not legacy)
4. Clients can detect mail changes and request updates
5. Backward compatible: clients without mailHash still treated as "old format"

---

[v0.8.0] (2026-01-21) - Pull-Based Delta Protocol

Branch: feature/pull-based-delta Status: Testing Phase Latest Update: 2026-01-21 (Evening)

🚀 Major Features

Pull-Based Protocol with Inventory Hashing

• NEW: Hash-based inventory comparison replaces version timestamps
• NEW: Automatic pull sync when inventory hashes differ
• NEW: Dual broadcast system (togbank-v + togbank-dv) for compatibility
• NEW: /togbank share command broadcasts version data with hashes
• NEW: Selective querying - only request data when hashes mismatch
• NEW: Data migration system computes hashes for existing alts on load
• NEW: Fast-fill feature - automatically requests missing banker alts when UI opens or /togbank sync is used
• NEW: Smart message prioritization - NORMAL priority for reliable delivery
• NEW: Communication debug filtering - Optional "(comm)" prefixed debug messages with separate toggle

Inventory Hashing System

• ComputeInventoryHash() generates numeric hash from bank + bags + money
• Hash computed automatically on bank scan (BANKFRAME_CLOSED event)
• Stored alongside version timestamp in alt.inventoryHash field
• More reliable than version timestamps for detecting real changes
• Minimal overhead (single number vs full data comparison)

Protocol Simplification

• REMOVED: Guild support threshold requirement (was 5% minimum)
• NEW: Delta protocol always enabled if PROTOCOL.SUPPORTS_DELTA = true
• NEW: Works immediately without waiting for guild adoption
• SIMPLIFIED: ShouldUseDelta() now only checks feature flags

Broadcast Enhancements

• Version Broadcast (togbank-v): Legacy format with version timestamps only
• Delta Version Broadcast (togbank-dv): New format with version + hash
• Format: data.alts[name] = {version = X, hash = Y}
• Share Command: Now sends BOTH broadcasts for maximum compatibility

Hash Comparison Logic

• Compares inventory hashes to detect changes
• We have no data: Query for everything
• Hashes differ: Query for update
• Hashes match: Skip query (no changes)
• No hash available: Fall back to version comparison
• Handles nil checks gracefully after database wipes

🐛 Bug Fixes

[SEARCH-003] Search Returning 0 Results

• Fixed: Search now correctly processes all aggregated items
• Root Cause: BuildSearchData was using ipairs() on hash table returned by Aggregate()
• Solution: Changed to pairs() for proper hash table iteration, fixed item counting
• Impact: Search functionality now works correctly for all item queries
• Location: Search.lua lines 405-410

Mail Data Persistence

• Removed: Unused IsMailDataStale() function and 1-hour staleness threshold
• Change: Mail data now persists indefinitely like bank/bags data
• Impact: Mail inventory remains visible regardless of age, with timestamp displayed for information
• Location: MailInventory.lua

[UI-002] Item Links Not Appearing After Integration

• Fixed: Items now display immediately after async item link reconstruction
• Root Cause: ReconstructItemLinks() was using async Item:ContinueOnItemLoad() callbacks without triggering UI refresh
• Solution: Added UI refresh calls after successful link reconstruction (both immediate and async)
• Impact: Items now appear in UI as soon as their links become available from WoW API
• Location: Guild.lua lines 970-995

[PROTO-001] Delta Validation

• Fixed: Delta validation now accepts link-less deltas without baseVersion
• Made baseVersion field optional in ValidateDeltaStructure()
• Maintains backwards compatibility with old protocol deltas
• Location: Core.lua line 118-122

[UI-001] Inventory UI Crash

• Fixed: UI handles missing bank.slots and bags.slots data
• Added defensive nil checks in Inventory.lua lines 177-187
• Data migration initializes missing slots fields with {count = 0, total = 0}
• Prevents crashes when opening UI with incomplete alt data

[DATA-001] Missing Inventory Hashes

• Fixed: Existing alt data migrated to include inventory hashes
• Migration runs once on addon load via Database:InitializeDatabase()
• Computes hashes from saved bank/bags/money data
• Successfully migrated 60+ existing alts in testing

Hash Comparison Edge Cases

• Fixed: Handles nil hash values after database wipe
• Fixed: Pull decision logic checks for missing data
• Fixed: Broadcasts send even when no alt data exists locally
• Debug output shows "has bank data for X (we have none), querying"

📝 Documentation Updates

• Updated DELTA_IMPLEMENTATION_TODO.md with current architecture
• Documented inventory hashing system and pull protocol flow
• Removed outdated guild support threshold documentation
• Added hash comparison algorithm documentation
• Updated bug tracker (DELTA_BUGS.md) with resolved issues

🔧 Technical Changes

• Guild.lua: Added dual broadcast to Share() function
• Guild.lua: FastFillMissingAlts() auto-requests missing banker alts (lines 458-498)
• Guild.lua: ReconstructItemLinks() now refreshes UI after async link loading (lines 970-1008)
• Events.lua: SyncDeltaVersion() uses NORMAL priority for reliable delivery (was BULK)
• Events.lua: Changed all query messages from BULK to NORMAL priority
• Chat.lua: Enhanced hash comparison logic with nil handling
• Chat.lua: Added UI auto-refresh when data adopted (togbank-d and togbank-d3 handlers)
• Chat.lua: /togbank sync command now also triggers fast-fill for missing alts
• Database.lua: Added hash migration for existing alts
• Database.lua: Added slots migration to prevent UI crashes
• Core.lua: Made baseVersion optional in delta validation
• UI/Inventory.lua: Calls FastFillMissingAlts() on Open() in delta mode (line 49)
• Output.lua: Added DebugComm() function for filterable communication debug logging
• Options.lua: Added commDebug toggle in config UI below log level

🎯 Performance Improvements

• Message priority optimization: Changed queries and delta broadcasts from BULK to NORMAL
• Improved responsiveness of pull-based protocol handshake
• Faster UI updates with async item link reconstruction
• Reduced query spam with fast-fill on-demand loading
• Communication debug filtering: Separate toggle for comm debug messages with "(comm)" prefix

⚠️ Breaking Changes

None - Full backwards compatibility maintained with v0.7.0 clients

---

v0.7.0 (2025-01-17)

Latest Update: 2026-01-20 - Fixed error tracking issues

🐛 Bug Fixes (2026-01-20)

Error Tracking System

• Fixed: Error tracking now works even when Guild.Info is not initialized

  • Implemented temporary in-memory storage for errors occurring before guild initialization
  • Automatic migration to database when guild data loads
  • No error data loss during addon startup phase
  • Query functions check both temporary and database storage

• Fixed: RecordDeltaError() logs debug messages when using temporary storage

  • Shows: "Using temporary error storage for <alt> (<type>): Guild.Info not initialized"
  • Helps identify initialization timing issues

• Fixed: Test function testDeltaErrorTracking() parameter mismatch

  • Was calling RecordDeltaError() with 2 parameters instead of required 3
  • Now correctly passes: altName, errorType, errorMessage
  • Ensures proper error categorization in test suite

Impact: Error tracking is now fully functional throughout addon lifecycle, including early delta failures before guild initialization completes.

---

🚀 Major Features

Delta Sync Protocol

• NEW: Intelligent delta synchronization protocol reduces bandwidth by 90-99%
• NEW: Automatic protocol version negotiation between v0.6.8 and v0.7.0+ clients
• NEW: Smart snapshot management with automatic corruption recovery
• NEW: Comprehensive error handling with automatic full sync fallback
• NEW: Backward compatible with v0.6.8 clients (seamless mixed-guild support)

Bandwidth Optimization

• Only transmits changed items instead of entire inventories
• Automatic size comparison (uses delta only if <30% of full sync size)
• Estimated bandwidth savings: 90-99% for typical inventory updates
• Guild-wide adoption threshold (50%) ensures efficient operation

Performance Metrics

• Real-time bandwidth tracking (delta vs full sync)
• Performance monitoring (computation and application times)
• Success rate tracking with automatic failure detection
• Estimated bandwidth savings calculations

✨ New Commands

• /togbank deltastats - Display comprehensive delta sync statistics including:

  • Bandwidth usage breakdown (delta vs full syncs)
  • Estimated total bandwidth saved
  • Operation counts and success rates
  • Average performance metrics (computation/application times)

• /togbank deltaerrors - Show recent delta sync errors for debugging:

  • Last 10 errors with timestamps and error types
  • Failure counts per alt
  • Highlights alts with repeated failures (3+)
  • Persists across /reload

• /togbank deltahistory - Show stored delta chain history:

  • Delta storage per alt with version transitions
  • Change types and ages for each delta
  • Verifies chain replay infrastructure is working

• /togbank protocol - Show protocol version distribution:

  • Online member protocol versions (v1 vs v2)
  • All-time protocol adoption statistics
  • Delta sync enablement status with threshold indicator
  • Recently seen members with their protocol versions

• /togbank clearsnapshots - Clear all delta snapshots (forces next sync to be full)

• /togbank forcefull - Toggle forcing full sync mode (temporarily disables delta sync)

• /togbank resetmetrics - Reset all delta sync statistics to zero

🔧 Technical Improvements

Database Layer

• Added deltaSnapshots table for efficient snapshot storage (1-hour expiration)
• Added guildProtocolVersions table for peer protocol tracking
• Added deltaMetrics table for bandwidth and performance tracking
• Implemented snapshot validation and automatic corruption recovery
• Added 23 new database functions for delta operations

Guild Module

• Rewrote SendAltData() with intelligent protocol selection
• Implemented ComputeDelta() for efficient change detection
• Implemented ApplyDelta() with robust error handling
• Added error tracking system with repeated failure detection
• Enhanced debug output with detailed size and timing information

Communication Layer

• Registered new togbank-d2 prefix for delta protocol
• Enhanced version broadcast to include protocol capabilities
• Added delta structure validation before application
• Implemented automatic QueryAlt on all failure paths

Error Handling & Recovery

• 6 error detection points covering all failure scenarios
• Automatic full sync fallback on any delta failure
• Per-alt failure tracking with user notification (after 3 consecutive failures)
• Automatic failure count reset on successful sync
• Snapshot corruption detection and automatic purge

📊 Monitoring & Visibility

Enhanced Debug Output

• Delta selection logging with size comparisons and savings calculations
• Performance timing for all delta operations
• Color-coded status indicators (✓/✗) for quick visual parsing
• Detailed error messages with context for troubleshooting

Statistics Display

• Bandwidth metrics with color-coded percentages
• Success rate with threshold-based coloring (green ≥95%, yellow ≥80%, red <80%)
• Performance averages for computation and application
• Protocol adoption visualization

🧪 Testing & Quality

• Created comprehensive test module with 30+ unit tests
• Test coverage for delta computation, size estimation, protocol negotiation
• Error handling tests for all failure scenarios
• Integration tests for full delta roundtrip
• Backwards compatibility tests for v0.6.8 mixed guilds

📝 Documentation

• Added comprehensive README.txt with all commands and features
• Updated installation instructions with CurseForge App (recommended) method
• Added troubleshooting section specific to delta sync issues
• Created detailed DELTA_IMPLEMENTATION_TODO.md documenting all phases
• Added FEATURE_IMPROVEMENTS.md with technical architecture

🔄 Protocol Specifications

Version 2 Features

• Protocol version: 2
• Supports delta updates: Yes
• Delta size threshold: 30% of full sync
• Snapshot max age: 1 hour
• Guild adoption threshold: 50%

Backwards Compatibility

• v0.7.0+ ↔ v0.7.0+: Delta sync via togbank-d2 (when threshold met)
• v0.7.0+ ↔ v0.6.8: Full sync via togbank-d (automatic fallback)
• v0.6.8 ↔ v0.6.8: Full sync via togbank-d (unchanged)
• No breaking changes - seamless upgrade path

🐛 Bug Fixes

• Fixed potential race conditions in snapshot management
• Improved error messages for version mismatch scenarios
• Enhanced validation to prevent corrupted delta application
• Added nil checks throughout delta codepaths

⚙️ Configuration

New constants in Modules/Constants.lua:

PROTOCOL = {
    VERSION = 2,
    SUPPORTS_DELTA = true,
    MIN_DELTA_SIZE_RATIO = 0.3,     -- 30% threshold
    DELTA_SNAPSHOT_MAX_AGE = 3600,  -- 1 hour
    DELTA_SUPPORT_THRESHOLD = 0.5,  -- 50% adoption
}

FEATURES = {
    DELTA_ENABLED = true,           -- Master enable/disable
    FORCE_FULL_SYNC = false,        -- Force full sync for testing
}

📈 Performance Impact

• Bandwidth Reduction: 90-99% for typical inventory updates
• Computation Overhead: ~2-3ms average per delta computation
• Application Overhead: ~1-2ms average per delta application
• Memory Impact: Minimal (~50KB per snapshot, auto-expiring)

🔮 Known Limitations

• Delta sync requires 50%+ of online guild to use v0.7.0+ for enablement
• Snapshots expire after 1 hour (forces full sync on first update after expiration)
• Large inventory changes (>30% of items) automatically fall back to full sync
• Options panel GUI for delta configuration deferred to future update

---

v2.3.0 (2025-10-27)

Full Changelog Previous Releases

• Merge pull request #3 from GrumpyPlayer/merge/fix-search-normalization-into-handle-malformed-data Merge/fix search normalization into handle malformed data
• fix issues related to clean-up and init, and also allow GM to author /bank roster updates
• Merge PR #1 (fix-search-normalization) into integration branch
• Perform cleanup of malformed data on addon initialization and handle malformed data better
• Fix nil version comparison in ReceiveAltData
  • Added nil check for existing alt data version before comparison
  • Prevents 'attempt to compare number with nil' error
  • Fixes crash when receiving data for alts without version field
• Fix Search.lua crash and implement name normalization with security improvements
  • Fixed nil table index crash in Search.lua duplicate detection
  • Added NormalizePlayerName helper for consistent 'Name-Realm' format
  • Implemented sender authentication to prevent communication spoofing
  • Added debug tools: /bank debug toggle and /bank debugdump command
  • Auto-enable bank reporting for detected bank characters
  • Relaxed delegation policy for multi-bank account support
  • Normalized player keys across all modules for data consistency