Open AtticusKuhn opened 3 months ago
I have one theory, which is maybe we should use
lake exe cache get!
instead of
lake exe cache get
By the way, "Compile mlirnatural
Executable 🧐" takes up 58% of the total runtime of the github workflow, so this could possibly speed up the GitHub workflow significantly.
Right, the same is true for our builds. In some way the cache does not work as expected. I often see on my laptop more recompiling as needed. It might be worth debugging this. Part of the reason for this might be supportInterpreter = true
in the lakefile.toml.
Take a look at the GitHub workflow logs, for example, https://github.com/opencompl/lean-mlir/actions/runs/10177448324/job/28148970867
Here is an excerpt of the output:
As you can see, it seems that lake is building "Batteries". Later on, it also builds "Aesop" and "Mathlib". I believe that this slows down the GitHub workflow significantly. I also believe that this is a bug, as Lean should not be building these files, but instead getting them from the cache.