On Thu, Nov 24, 2022 at 12:26:57PM +0100, Jean Abou Samra wrote: > Le 24/11/2022 à 12:12, Kevin Barry a écrit : > > GC is definitely not the kind of thing where "passes the > regtests" means there is no problem. As we have seen in the > past, the most infuriating and mysterious GC problems reproduce > 1 time in 20 doc builds ...
This is not ideal, but would some kind of manually run test that does things like that (build a bunch of pathological-case scores 100 times) be a bad idea? There would be a maintenance burden, but it might mean we catch problems before breaking users' scores. The test could be run (by a human) when there's a new Guile release, or before we do a stable release, or some such. Kevin