Hmmm... The image is loaded and detected as an RT26 here.
So, this is because of another feature that tox gives us: HASHSEED randomization. Later versions of python randomize the hash seed on startup. Without that, dictionaries and other such structures appear to have stable ordering, even though they’re not supposed to. That can hide bugs (as it is here) where things on a system all appear to happen in the same order, but on someone else’s system they happen in a different order. So, tox helps poke those bugs by randomizing the hash seed for the tests as well. If I run tests on my system here with tox, I get the same failure that the build system just found about 20% of the time. That means that if the detection logic finds the RT21 driver first when loading an RT26 image, it’ll think the RT21 is right. If the RT26 sorts first, then it’ll be fine.
Maybe you can spot something in the RT26 driver and image to make it a little tighter? If not, I’ll take a look.
—Dan