Pip Cet wrote:
> Which is what I meant when I said the test
> was misleading: it's true for subtly different reasons on different
> hosts

Who would have thought that the tedium of writing floating-point tests could be 
overwhelmed by the tedium of documenting them and worse yet, writing ChangeLog 
entries for them? Anyway, I just now experienced this, and installed the 
attached into emacs-26. Though I was strongly tempted to remove the tests instead.