cfallin commented on issue #5072:
Hmm -- the test failure (which is hit in a fuzzing smoketest for GC but is unrelated to GC) comes from observing an "invalid" type in the elaborated CLIF; these are placeholders that are filled in on multi-arity producers when their projection nodes are processed. If one output of a multi-output inst is unused then it would never be filled in. I had figured that it wouldn't matter what type was present when a particular output was not actually used; but this appears not to be the case, so the node-shrink strategy will need a little more thought.
cfallin commented on issue #5072:
... and the latest commit fixes it, but r? @fitzgen again if you'd prefer I fix it in a different way...
Last updated: Nov 22 2024 at 17:03 UTC