Closed michael-schwarz closed 2 months ago
Interestingly enough, this causes a regression where the increased precision leads to a lower number of contexts, which in turn causes us to lose 5 tasks on termination.
This is quite counter-intuitive: how do fewer contexts lead to us failing to prove termination?
Because the Context-sensitive callgraph becomes recursive: Earlier the partial(!) contexts happened to distinguish things by (happy) coincidence. This is no longer the case now.
In that case something from #1340 should be able to distinguish the sufficient number of non-recursive cases in a less roundabout way.
Yes, either something from there or one of the ideas from #1422.
Closes #1159