Arma Reforger Tools 1.8.0.13, Linux via Proton (Steam default, then Proton Experimental 11.0-100),
RTX 5090 through vkd3d-proton.
Moving the World Editor camera down towards ground level on a terrain whose surface layers have just
been filled ends the session immediately:
ENGINE (F): Crashed
Access violation. Illegal read by 0x1421a550a at 0xffffffffffffffff
It happened three times on 2026-09-12, once per attempt, each uploaded by the crash reporter:
Crash GUID 68902920-aa82-4078-8390-3c7054af8d09 (99 s after a clean reload)
Crash GUID 85dc5113-2b72-490f-b737-6353d0ed75f2 (just after a satellite map import finished)
Crash GUID 39153e72-08cb-495c-a90d-4a48fd476a02 (first session on Proton Experimental)
The same fault address in all three, across two Proton versions, which is why I do not think this is
Proton. It is a read of a null pointer (0xffffffffffffffff), and it happens while the near-field
surface layers stream in for the first time.
What it is NOT, in case the logs look similar:
- not the satellite import — the first crash happened without one;
- not the same bug as the consistency-check crash at 0x1421a0829 reading 0x20000000c, which I have
filed separately. Two terrain crashes on this build read alike on the log's first line; the fault
address tells them apart.
Nothing was lost in any of the three: Terrain.terr, all 289 .ttile files with their GCTD chunks, and
a 1,902-instance layer were intact after each relaunch.
Steps to reproduce:
1. Build or open a terrain and fill its surface layer.
2. In the World Editor, fly the camera down to near ground level over the filled area.
3. The session ends with the access violation above.
Workaround I am using: never approach the terrain in the viewport; judge surface work from the GCTD
chunks in the .ttile files, and prefer -wbsilent command-line runs that never draw.
Note: the session logs from 2026-09-12 have since rotated off disk, so I cannot attach them. The three
crash GUIDs above should have the uploads.