Running batch mode on macOS or iOS where the portability implementation is not complete results in early failures (e.g. when mvk throws errors due to SPIR-V to MSL conversion limitations, Vulkan feature limitations, etc). This makes it difficult to run through the complete list of samples without manual intervention (e.g. sample "xxx") or adding custom code to skip the known failing samples in batch mode which is not ideal. Instead, the error recovery framework should be more resilient and detect these issues, recover, and continue through the complete list of samples until the end.
I have prepared a pending PR with several changes that handles this problem in a general way, as well as solves several other issues observed on macOS.
Running batch mode on macOS or iOS where the portability implementation is not complete results in early failures (e.g. when mvk throws errors due to SPIR-V to MSL conversion limitations, Vulkan feature limitations, etc). This makes it difficult to run through the complete list of samples without manual intervention (e.g. sample "xxx") or adding custom code to skip the known failing samples in batch mode which is not ideal. Instead, the error recovery framework should be more resilient and detect these issues, recover, and continue through the complete list of samples until the end.
I have prepared a pending PR with several changes that handles this problem in a general way, as well as solves several other issues observed on macOS.