Andrey Mokhov

Many thanks @christinerose! Agreed about "one-shot benchmarks".

> It would be nice to have a way for dune to detect that it is in such kind of loop and report it with the cause. At least for...

I don't think so. Note that deferring promotions is not going to fix the looping.

> The changes look OK to me. I added @snowleopard as a reviewer since it touches the engine @rgrinberg Could you import this change so we can benchmark it and...

cc @rleshchinskiy who recently implemented something similar internally (I believe he simply called `rm_rf` in both cases).

> Probably we need a `(version N.NN)` line in file `dune-project`. @xavierleroy Yes, that's the simplest way forward. You will need to remember to update this line when making a...

@adithyaselv Thanks! Yes, after reading various discussions, including the ones you link, I guess we can live with slightly less elegant code for the sake of having covariance, which seems...

Implementation-wise, switching to a depth-first search ordering seems easy: replace a queue of jobs with a stack of jobs. However, it's unclear to me that this will actually speed-up this...

For a prototype, it's probably easiest to replace this queue with a stack: https://github.com/ocaml/dune/blob/b074bbef23ba2016650c9cfff0a17948b8da64a3/vendor/fiber/src/throttle.ml#L5-L9

@Janno Cool :) How about trying random ordering instead? I'd expect it to do best on average. (Resource management is tricky and doing it optimally is hard. But randomness can...