Closed
Description
I've been examining profiling results of the clap-rs test case (I am profiling the #51538 branch, more specifically). One clear hot spot is the do_dataflow
, which accounts for 18% of MIR borrow check time. It is a fairly naive "fixed point" iteration that propagate bits around. It can (probably) be improved by introducing a "dirty list" to avoid processing basic blocks whose contents have not changed.