https://bugzilla.redhat.com/show_bug.cgi?id=2006590
--- Comment #4 from Ben Beasley code@musicinmybrain.net --- Thanks!
-----
Honestly, it looks a bit incomplete in the guidelines: to put headers into arch-specific packages only to satisfy arch-specific tests, when the "library should have tests which should be run on all architectures" condition can be archived a better way.
Agreed!
Also, there is no "install process may modify the installed headers depending on the build architecture" situation.
Also agreed. While it’s not always easy for the packager to tell if this is happening, there are plenty of other cases where noarch packages *are* allowed or encouraged but builder-arch-dependent differences can slip in anyway—perhaps chief among them, -doc subpackages. Besides, the tooling exists to reject arched builds when a noarch subpackage is not consistent across build architectures.
I think if the guidelines were changed to specify that only the *base* package of a header-only library must be arched, and the -devel package *may* be noarch, it would still satisfy all of the stated justifications for the current mandate.