Fixes
Nine fixes. You decide which run.
Trebuchet's Settings list every fix with a one-line explanation. Turn off "Use Trebuchet's fixes" and the game starts like Steam's Play button.
Fixes marked change how a few parts of the game's code run, in ways no standard CrossOver setup does. No game file and no byte of the game's code is changed, and results are proven identical in offline replays. They have not been tested online yet. If you want zero difference from a standard CrossOver setup in a match, switch them off. Gains measured on a Mac Studio M2 Max in offline replays unless marked as an estimate; FPS figures are Metal HUD readings. Your Mac may differ.
Always on
These two follow the main switch, "Use Trebuchet's fixes".
| Fix | What it does | Default |
|---|---|---|
Quick Traps |
60 units selected: 10–12 → 22 FPS (Metal HUD reading) The game runs one particular check thousands of times a second. Under Rosetta each one took a long detour through macOS. Quick Traps shortens that detour and gives the game exactly the answer it would have got the long way. Every other fix builds on this one. Measured: 53.5 µs → 0.62 µs per check in our test program; checks taking the detour in the game about 15,000 a second → 0. Paused, 60 units selected: about 10–12 FPS before, 22 after (read by eye from the Metal HUD, OpenGL renderer, 18 Nov 2026). |
On |
Renderer Guard |
keeps the chosen graphics path Makes sure the graphics path you pick is the one that loads. Without it, a missing library path made CrossOver fall back to its slower OpenGL renderer without telling you. |
On |
Tied to a game or macOS version
These switch themselves off after an update they don't recognise, and tell you.
| Fix | What it does | Default |
|---|---|---|
Swift Health Bars |
worst case, 37 units selected, health bars on: 22 → 150 FPS (Metal HUD reading) Every health bar asks a small question many times per frame, and the slow part of the answer came first. Rosetta is shown the same function with the cheap check first. Same answer for every input, far less work. The worst case it targets: 37 units selected, paused, 22 → 150 FPS on top of Quick Traps, read by eye from the Metal HUD (OpenGL renderer, classic health bars, 18 Nov 2026). |
On |
Hitch Guard |
freezes of 0.15–0.23 s every 30–40 s → under 5 ms Every 30 to 40 seconds Rosetta frees old translated code and used to freeze the game while it did. Hitch Guard makes it free that memory in small steps. The total work is the same; it no longer comes in one lump. Measured: the game's main thread waited 150–230 ms per clean-up without it, at most 4.3 ms with it, in two recorded team fights. Frame rate unchanged. |
On |
Lean Stack Check |
The game runs a safety check on most function returns. Rosetta is shown a shorter, identical form of it: same result, fewer instructions (19 → 16 per check). |
On |
More fixes
Each has its own switch in Settings.
| Fix | What it does | Default |
|---|---|---|
Metal Renderer |
about 25 % less CPU per frame than wined3d, no periodic 150 ms stall Uses DXMT, which draws the game through Apple's Metal. You can pick D3DMetal or CrossOver's OpenGL path instead, in the Graphics tab. Measured: about 25 % less processor time per frame than the OpenGL renderer (wined3d), and no periodic 150 ms stall. Frame rates were similar in single runs (1366 × 768, one replay). |
On |
Clear Menu Text |
readable menu buttons with DXMT Fixes the garbled button labels in the menus with DXMT, using our own DXMT build. DXMT copied small textures with the wrong row length; the fix is two lines. Off means CrossOver's own DXMT, with the broken labels. |
On |
Fast Sync |
A CrossOver setting (MSync) that lets the game wait on Wine without asking the Wine server each time. It helps when the game's main thread sets the frame rate. A gain is shown here once repeated measurements confirm it. |
Optional, off |
Fast Clock |
about 0.6 s faster loading (estimate) The game reads the clock millions of times while it loads, and hundreds of times per frame. Fast Clock answers those reads without a trip into Wine. The loading gain is an estimate, not yet measured in the game. Optional and off by default, because it is newer and less tested than the other fixes. |
Optional, off |
Later
Not in this release.
| Fix | What it does | Default |
|---|---|---|
Modern C Runtime |
Your bottle ships Microsoft's 2015 pre-release C runtime. Trebuchet will offer to swap in the newer Microsoft copy that is already in your bottle, with a checksummed backup and an Undo button. Nothing changes unless you say so. Measured in a test program: string compares in 0.43× and formatted printing in 0.52× the time. |
Next release |
Local Cursor |
Cursor and text-caret requests will be answered inside the game process instead of a round trip to the Wine server: three round trips per frame. |
Planned |
Why some numbers are missing
We only show a number we measured, with the scene and the Mac it came from. If a fix has no number, we either haven't measured it in the game yet or the measurement didn't show a clear gain. Estimates are labelled as estimates.