I’m sorry that you had issues with the Xcode 11.4 image. I’m here to tell you that no changes were made on Xcode or macOS itself. Although their website says that macOS 10.15.2 is required for this version of Xcode, apparently this requirement is not enforced as you could see.
Thanks for raising this issue, we’ll look into it further and I will keep you posted.
Does that really matter? If 10.14 is officially unsupported, it’s not guaranteed to work there.
At the very least, as you could see from the OP’s message, testing applications targeting 10.15 does not work.
Any app which is many might require Combine or another framework used in 10.15 therefore the CI will not work
IMHO something is fishy with this Xcode 11.4. When I switched to using it, everything seemed to build ok; but results (this is a c++ cli-app calculating tons of numbers) are slightly off. To be fair, this could well be an error in my code, but my own builds as well as those on travis were absolutely consistent until I switched travis to 11.4. My own builds on 11.4 also have no problems, it is just travis 11.4; when I switched it back it went back to normal.
What really irritates me: I have absolutely no idea what kind of error in the environment could cause the behaviour that I am seeing. It’s calculating millions of numbers and only a handful are off. My best bet is that somehow floating point numbers are being handled differently in edge cases, but I would think (and hope) that things like that depend on compiler flags (like -ffast-math) which I have not changed…
Tracing time, I guess… then diff the output
There is nothing to debug if I can’t reproduce the error
I have found the problem though; using the -flto flag changes the results. That means I’d have to debug llvm and that is way beyond my knowledge.