[Feature flags]Seeking Guidance on PyTorch Feature Flag Design and Best Practices

I’d like to learn from PyTorch’s experience in designing feature flags to enable custom implementation. Could you help me resolve the following questions, or provide related resources for my reference: * 1. How does PyTorch design feature flags? What principles does it follow? * 2. How should feature flags be used in the source code to prevent interference among them? Are there specific rules, constraints, or even clever techniques? * 3. In testing, how can one ensure that all combinations of feature flags do not introduce bugs? Does it require testing every possible combination, or are there alternative assurances? Thx a lot