Description
#12153 introduce a performance regression for the stdlib test. I tracked it down to attempts to do match type reductions that seem very slow.
I saw that overall there were 566 attempts at match type reductions, presumably when compiling the Tuple.scala (and maybe also Tuples.scala) files. Without the change to translucentSuperType there were none. But the change is necessary and 566 does not seem an excessive number. Nevertheless these reduction attempts (which all fail in the end) seem responsible for the significant slowdown of stdlib compilation as a whole.
We should get to the bottom of this. As a first step, we should try to get a flamegraph or other profile just for compiling Tuple.scala and Tuples.scala. Once we know where the slowdown happens we should see what we can do to optimize.