#34857 Cypress 15.21.1 binary crashes with SIGILL (invalid opcode) at a fixed offset on WSL2/WSLg

Current behavior

The Cypress binary (Electron 37.6.0) repeatedly crashes with SIGILL (signal 4, invalid opcode) when running on WSL2 with WSLg. It happened 10 times within ~24 hours (2026-09-14 13:15 – 2026-09-15 12:23).

The crash is deterministic in location: every time the trap is at the same offset in the Cypress executable (0x4a83146), across different PIDs and ASLR base addresses. This points to an intentional abort (CHECK() / IMMEDIATE_CRASH()) in Chromium/Electron, not to a CPU incompatibility (the CPU supports AVX2).

Kernel log (dmesg) from two of the crashes:

[ 7215.487754] traps: Cypress[2637] trap invalid opcode ip:58fc54b9a146 sp:7ffc4d869da0 error:0 in Cypress[4a83146,58fc52861000+92d3000]
[ 7215.487988] Cypress: Cypress: potentially unexpected fatal signal 4.
[ 7215.488355] CPU: 10 UID: 0 PID: 2637 Comm: Cypress Not tainted 6.18.33.2-microsoft-standard-WSL2 #1 PREEMPT(none)
[ 7215.492079] WSL (2645 - CaptureCrash): Capturing crash for pid: 30861, executable: !root!.cache!Cypress!15.21.1!Cypress!Cypress, signal: 4, port: 50005

[ 8937.271931] traps: Cypress[21391] trap invalid opcode ip:5ac44a188146 sp:7ffc9605cf00 error:0 in Cypress[4a83146,5ac447e4f000+92d3000]
[ 8937.272400] Cypress: Cypress: potentially unexpected fatal signal 4.
[ 8937.272426] CPU: 2 UID: 0 PID: 21391 Comm: Cypress Not tainted 6.18.33.2-microsoft-standard-WSL2 #1 PREEMPT(none)
[ 8937.274907] WSL (21399 - CaptureCrash): Capturing crash for pid: 41251, executable: !root!.cache!Cypress!15.21.1!Cypress!Cypress, signal: 4, port: 50005

Side effect: because WSL collects crash dumps by default, each crash produced a core dump in %LOCALAPPDATA%\Temp\wsl-crashes on the Windows host — up to 194 GB per dump (577 GB total for 10 dumps, on a VM with only 7 GB RAM), which filled the host's system drive to 99 %. The dump size itself is WSL behavior (see microsoft/WSL#41314), but it makes this crash especially harmful for WSL users.

Setup details:

Cypress runs as root (binary cache in /root/.cache/Cypress)
WSLg is active: DISPLAY=:0, WAYLAND_DISPLAY=wayland-0
Testing type: e2e, started via cypress open --config-file cypress.config.ts

Desired behavior

Cypress should not crash. If Electron hits a fatal internal check (e.g. sandbox, GPU or Wayland/X11 display problems under WSLg), Cypress should print a readable error message explaining the cause instead of dying with SIGILL.

Test code to reproduce

Environment to reproduce:

Windows 11 with WSL 2.7.14.0 and WSLg 1.0.73.2
Ubuntu 24.04.3 LTS distro, logged in as root
Install [email protected] in an e2e project and run npx cypress open (or npx cypress run)
Cypress crashes with SIGILL after some time; check dmesg | grep Cypress

Cypress Version

15.21.1 (Electron 37.6.0, bundled Node 22.19.0)

Debug Logs

Other

Environment details

WSL version: 2.7.14.0
Kernel: 6.18.33.2-2
WSLg: 1.0.73.2
MSRDC: 1.2.7214
Direct3D: 1.611.1-81528511
CPU: 13th Gen Intel Core i7-1355U (12 threads, AVX2), 7 GB RAM assigned to WSL

Possibly related

#7045 – "The Test Runner unexpectedly exited via a exit event with signal SIGILL" (open, CI/Docker, often as root)
#28559 – "SIGILL (invalid opcode) crash after Network.loadingFailed loop of death" (closed as stale, same symptom on Cypress 13)
#23962 – SIGILL in Docker (closed as stale; maintainers asked to open a new issue for similar reports)
microsoft/WSL#41552 – WSLg Weston SIGSEGV crash loop with the same WSLg version (1.0.73.2); a crashing compositor could trigger a fatal check in Electron
microsoft/WSL#41314 – WSL crash collector writes huge Electron dumps to %TEMP%\wsl-crashes