trunk/6e700ca9014c0c7ba112f66d02a2e81b94cc1f20: Handle non-UTF-8 subprocess output in minifier isolate_fails (#196620)
- PyTorch: 780 events in the last 90 days
- PyTorch: 770th Release in the last 90 days
- Previous: earlier the same day · trunk/d80fa4da5bc1b591816f808aabb683f54b41bdef: [torchcomms hash update] update the pinned torchcomms hash (#196383)
What happened
#190696 added errors="replace" decoding to the minifier test harness after the ROCm flake in #190681 , but the production minifier has its own decode sites: isolate_fails reads the isolated subprocess's stdout/stderr logs with bare decode("utf-8") calls. When the subprocess emits a non-UTF-8 byte (0xb1 from HIP runtime output on ROCm), isolate_fails raises UnicodeDecodeError, minification dies before repro.py is wri…
Summary assembled by rule from the sources below