Repository navigation
v2.6.1
TaskManager+ v2.6.1 — Crash fix for GPU-accelerated AI
A patch release fixing a crash introduced by the GPU generative backend. If you use the Enhanced AI tier with GPU acceleration, v2.6.0 could close the app without warning when opening the file inspector. Updating fixes it; no settings need changing.
The crash
Opening a file or folder in the inspector on the Enhanced tier asks the on-device writing model for a summary and name suggestions. With GPU acceleration active, that request could terminate the app instantly.
The cause was a mismatch in how the app described one of llama.cpp's configuration structures to the GPU library. The app's definition was 16 bytes shorter than the library's, so when the library filled it in, it wrote past the end of the space reserved for it and corrupted memory that belonged to something else. The next call then failed with an access violation.
Two details made this awkward to find:
- It only happened in release builds. Development builds lay out memory with enough slack that the overspill landed in unused space and nothing appeared to go wrong. Optimised builds pack memory tightly, so the same write hit live data. The shipped app crashed; the same code under a debugger did not.
- It looked like a timing problem. The crash arrived after a burst of activity, which pointed at concurrent requests. It wasn't — it was the very first request every time, and the burst was unrelated.
A second, quieter version of the same bug was fixed alongside it: the model-loading structure was missing a field in the middle, which shifted several later options by one position and meant the app was passing the wrong values for some of them.
Both structures are now checked against the library's real layout when the app is compiled, so this class of mismatch fails the build instead of shipping.
Logging
The app now writes its log to a file:
%LOCALAPPDATA%\com.taskmanagerplus.app\taskmanagerplus.log
Previously the log went only to a console, which a released build doesn't have — so when the app crashed, it left nothing behind to diagnose from. The log is flushed continuously so the final entry survives even an abrupt termination, and it starts a fresh file once it passes 5 MB. If you hit a problem worth reporting, this file is the thing to attach.
Build maintenance
The project's CI now runs on Node 24. GitHub is retiring Node 20 for Actions in September 2026, and the build was still pinned to it.
Full Changelog: v2.6.0...v2.6.1