https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113761 

Goal

Fix the incorrect behaviour identified in the upstream bug.

Investigate why the implementation stores empty vectors for 0 or 1 boundaries and then populates new vectors as needed. Is that to avoid long-lived heap allocations when the values are only needed transiently? See if the mailing list has an related discussion from around the time of the changes I mentioned in the upstream bug. Do we just want to fix the bug, or refactor the handling of n < 2? Can that even be done without breaking the ABI?

Acceptance criteria

Close the upstream bug.