[sauercrowd]

it's a rejection of the thesis that it "does not work". It does, but it requires investing into a suppression file.

[hu3]

plus, LLMs can generate suppressor files from logs. It's much faster these days.

I've had success with this approach.

[gizmo686]

Valgrind can generate suppression files directly.

[skissane]

Yes, but in my past experience, one often wants to be edit those files to make them more generic… Valgrind struggles to distinguish which parts of the call stack are essential to a “known leak” versus which are coincidental.

I have never tried asking an LLM to do this-but it seems like the kind of problem with which an LLM might have some success, even if only partial success.

[hu3]

yes but LLMs have access and understanding of my code and can better discern what should be suppressed.

[amelius]

Yeah I tried that on a PySide6 application.

Trust me, it does not work.

[fragmede]

skill issue