Agreed on all points. We happen to be in a situation where we have very long development cycles, and also rarely need to go back and rebuild older versions. S

Agreed on all points. We happen to be in a situation where we have very long development cycles, and also rarely need to go back and rebuild older versions. So, in our case we could get away with this. And yes, the speed improvements came from a combination of iterating over Git commits in a single Python process vs running Git as an external command, and avoiding touching disk for everything except reading and writing blobs and commits.