KonstantinSeurer
commented
6 months ago I went with adding a lock to enqueue_parent_pipeline since enqueue_create_.*_pipeline already lock their access, so it's trivial. The other approach got quite messy.
I also pushed a small cleanup commit which isn't really relevant for this PR, but I also don't see a need to open another PR for it.
As you suggested, the depth is now computed by enqueue_parent_pipelines. enqueue_parent_pipelines now writes the hashes of enqueued pipelines to a set of which the parent pipelines are enqueued until all dependencies are resolved. The loop over max_depth has been replaced by a vector which is sorted by compilation order.