PyTorch optimizes CI by aborting builds for 10+ deep ghstacks

PyTorch, the popular deep learning framework with over 100k GitHub stars, has merged a CI optimization that tackles resource waste from deep ghstack PR chains. Ghstack is a workflow for stacking multiple pull requests, but when a stack exceeds 10 PRs, running full CI on each push becomes inefficient because the top PRs often need retesting once the underlying PRs land. Developer malfet addressed this by parsing the ghstack breadcrumb in the PR description to detect stack depth and automatically aborting builds for chains deeper than 10 PRs.

The implementation reuses the existing filter_test_configs.py script, avoiding new infrastructure. The change was approved by reviewer huydhn and tested via python -m unittest test_filter_test_configs. This optimization reduces CI compute costs for large, stacked contributions, a common pattern in PyTorch development. While it doesn't cancel builds for stacks under 10, it targets the most wasteful cases. The move reflects PyTorch's ongoing focus on developer efficiency and resource management in open-source CI pipelines.