On Mon, 6 Apr 2026 at 14:01, Chao Li <[email protected]> wrote: > > On Apr 4, 2026, at 11:30, David Rowley <[email protected]> wrote: > > Not quite perfect as a set made first as a single word set that later > > becomes a multi-word set will be double counted. The number of > > operations on the sets is likely more important anyway, not the number > > of sets being created. The point is, multi-word sets are rare for most > > workloads. > > > > What tests did you run after adding the logs to collect the data? This is a > method I might borrow in the future for similar investigations.
"make check" then grep regression.diffs for the NOTICE message and pipe to "wc -l"
