viable/strict/1788565816: Disable addr2line symbolization by default (#195462)
- PyTorch: 539 events in the last 90 days
- PyTorch: 533th Release in the last 90 days
- Previous: earlier the same day · viable/strict/1788560676: Build Windows MAGMA for CUDA 13.4, drop CUDA 12.x builds (#196014)
What happened
Motivation: I noticed people keep disabling this in production jobs because the addr2line symbolization is really slow. If this really is a problem, we should disable it in OSS too. Human direction: Flip TORCH_DISABLE_ADDR2LINE so it is true by default, and expose a private torch._C API so the behavior can be tested without compiling a C++ extension. AI-assisted implementation summary: Default C++ stack trace symbol…
Summary assembled by rule from the sources below