I also just want to reiterate my understanding here to make sure I’m not missing something: if we follow through with this patch and, minimally, consume it using a forked zinc, we can expect a ~4x speedup in local incremental compile times (when deep in the graph, i.e. on very leafy targets)