Two costs read straight out of the player's own `WOW_UI_HANDLERS=1` session log
(2026-10-18 20:33), where a `/reload` blocks the main thread for 7.84 s.
**The saved-variables flush, 305 ms.** The reload's first act is to write every
loaded addon's saved variables, one file each, and `write_atomic_bytes` ends each
with `sync_all()` — a disk round trip per file, 60 of them, inside the one blocked
frame. Measured on this project's own btrfs home: **327 ms flushed, 6 ms not**,
which is the +27→332 ms the log spends between `ReloadUI — ending the UI session`
and the last `wrote …/saved/*.lua`. `write_atomic_bytes_bulk` keeps the temp file
and the rename (so a file is whole or absent) and gives up only a power loss in
the seconds after; the reference writes these with plain buffered CRT writes and a
`.bak` rotation whose result it discards, so this is still the stronger of the two.
**A whole VM built and thrown away, 310 ms.** `end_ui_session` installs a fresh
boot VM, which is right for the roots that stop there — the character screen runs
on it — but `run_pending_reload` then calls `load_ingame_ui_on_world_entry`, which
mints its own in `mint_entry_vm`. So every reload ran `UiScript::new`,
`GlobalStrings.lua`, the emote table and the font registry twice and let the first
one go without running a line: the log prints `GlobalStrings loaded` at +350 ms and
again at +662 ms. `end_session` now takes what to leave behind, and the reload
leaves the dying VM until the mint replaces it — safe because the two calls are one
exclusive system with nothing in between, the old VM is still dropped before any
file loads, and `fold_dying_vm_cvars` drains what it folds, so the mint's second
fold of the same VM finds nothing. Only when the rebuild is certain to mint: with
no UI wanted (a capture) the load returns early and the world would be left with no
VM at all.
`chunkbench` is the instrument the next item needs, and the one that priced this
one's neighbours: it runs a file list three ways — a VM per file, one VM with the
stock collector, one with it stopped — so parse, execute and the GC's share are
separate numbers. Over his enabled addon set (1,471 files, 70 MiB) the parse is
703 ms against 74 ms to compile the same chunks from bytecode, which is what a
bytecode cache would be worth; the collector is not the problem (829 ms against
731 ms stopped), and pfQuest's 23 database files are 417 ms and 110 MiB of heap.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SgstFWtn1RxEJ5kVrpgYwn
Part of #41: the comments of `benilla-ui` (the Lua UI engine, its FrameXML loader and its widgets) are rewritten to the rule in `docs/METHOD.md`, which is what the code does and the reference fact behind it, in a line. 43,140 comment lines become 15,161.
- **The mechanical strip.** The citations that were only a private record number go first, and each edit is shown to be a pure removal.
- **Wrong facts corrected.** The rewrite corrects the comments that stated wrong facts. Among them:
- `0x64a480` is the case-sensitive compare (a byte `repz cmpsb`), and `0x64a4c0` the case-insensitive one.
- A frame's id sits at `+0xb0`, not `+0xb4`. `EnableKeyboard`/`IsKeyboardEnabled` are `0x776ec0`/`0x776f90`.
- A region's unpinned edge has no fallback to its owner's rect: the owner supplies only the scale.
- `$parent` expands to the nearest named ancestor, or "Top" when none is named (`0x76c5b0`), not simply to the parent.
- A StatusBar fills from its internal normalised fraction, not from `GetValue()`.
- The addresses of `StartDuel` and `StartDuelUnit` were swapped.
- Many "untraced" notes now give the traced reference fact: the drag threshold, the per-grip sizing edges, the empty-`autoFocus` skip, and `SetTexCoordModifiesRect`'s default.
- Stock FrameXML line citations are checked against the stock 1.12 files and corrected. `ref-` file names become the stock file names.
- **Deviation labels.** "Deviation:" now marks only a choice with a reason. A surface that exists only for the retired Era addon target says so plainly: it is not 1.12.
- **Verification.** Every file strips to identical code, and `cargo fmt --check` is clean. Nine second readers compared every file old against new, restored the facts the rewrite had dropped, and checked each new citation against the code, the stock files, vmangos and the reference.
Behaviour differences the rewrite surfaced are filed as their own issues (#131–#142), and more are being validated. The comments here state what the code does. String literals are untouched; they are #92.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- a nameplate's name and level stop popping a pixel every other step as the plate slides, and the whole plate no longer jumps 19 px to the top of the screen on a rounding tie; the chat bubble takes the same fix (2172)
- creatures and other players stop sinking through the ground while the world is still streaming in, where a Stormwind patroller had been 31 yd under its street: a body our world cannot yet answer for holds the server's pose (2174)
- select and the string metatable are gone, as in 1.12's Lua 5.0: pfUI's heal prediction parses its messages with the right format, BuffCheck2's own select wins its fallback, and a method call on a string raises the reference's own error (2171)
- the camera's water collision comes back out for now: with the pivot a hair under the plane, a surface swimmer's camera snapped in and out against the waterline, so the boom ignores water again (2173)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- the mail window's send tab takes typing again: the subject, body and money boxes were taking no input while the recipient box swallowed every key, and a sent letter clears its subject and item icon instead of putting them straight back (2145)
- seven events fire the way the reference does: UNIT_DYNAMIC_FLAGS fires at all, swapping two identical stacks between bank slots fires PLAYERBANKSLOTS_CHANGED twice, the pet sheet's training points repaint, and ITEM_LOCK_CHANGED loses its invented arguments (2140)
- nine widget methods 1.12 addons probe for are in: ShaguTweaks stops recolouring textures the reference leaves alone, pfUI and _Nameplates stop raising on a tooltip's GetAnchorType, and 63 AceEvent addons get RegisterAllEvents (2142)
- six post-1.12 methods leave the widget surface (SetSize, SetShown, HasFocus, SetFormattedText and two tooltip verbs), so an addon reaching for a later client's API finds what a real 1.12 client gives it (2142, 2146)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- benilla loads real 1.12 addons from Interface/AddOns — TOC, XML and Lua through one loader, with per-addon saved variables, Bindings.xml and the lifecycle events addons expect
- both AddOns screens are in: the character-select list with per-addon enable state, and the ESC menu's in-game list
- the API surface is measured, not guessed — a corpus harness runs hundreds of real 1.12 addons and ranks exactly which globals, templates and verbs they demand
- the Lua dialect is the client's own, read off the binary — 5.0 semantics with the 1.12 additions, and UI source files are bytes, not UTF-8
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
gromandClaude Opus 5