I'm closing and reopening. One of the appveyor tests did not seem to run properly.

If I am not mistaken, this tackles #6620 and #6604

No connection to #6620, which is about segfaults on 32-bit systems, not minuscule image failures.
I don't think it is connected to #6604 either, because that is also about 32-bit system-specific image failures--fairly large ones.

So on Windows there seem to be intermittent non-deterministic behavior episodes causing failures sometimes, but not always or on all of the runs. That's what the present trivial PR addresses--one such example. I expect many more will crop up over time.

In addition, it seems there are some real problems with 32-bit systems behaving differently than 64-bit systems, which should not happen. That's much harder to deal with, and I suspect impossible for us to track down. The problem is likely buried deep in some library, maybe with a helping hand from a CPU bug.