I’ve often thought it would be interesting to study what developers do while their code is waiting to compile. The complexities of the software tooling behind machine learning research have recently introduced a considerable amount of waiting around. This does not always involve waiting for compiling (although it often does). Rather, it relates to waiting for unwieldy development environments to be “solved” – perhaps by a package manager like conda. There is also, of course, the massive amount of time that is consumed during training. When an experiment is properly scheduled, there should be minimal wastage of the researcher’s time but all of us know that (at least during development) there is a degree of waiting around. There are whole libraries (e.g. tqdm) whose sole purpose is to mark the passage of wall time, while computation is taking place.
What do developers do when they are waiting?