IMHO this is a case of the test case being broken, not the code itself, and
thus this is the correct course of action (the alternative being to make the
test case more complex to account for the different alignment of 128bit
integers on various platforms, but I am not even sure why upstream tests that
in the first place to be honest).
happy to be corrected though, in particular if porters disagree :)
For what it's worth, I also started looking into this issue and
independently
came to the same conclusions as Fabian.
Furthermore, the failure was only a psuedo-regression, the failure was
already happening in 1.39.0-1, it was just being ingored.
