Tom Tobin
In temporary lieu of posing questions explicitly to the SFLC, I dug up a copy of _Intellectual Property and Open Source_ by Foobar (and published by O'Reilly), and found this (from an entire chapter — Chapter 12 — about the GPL):
"Nevertheless, there is a persistent issue that won’t go away—whether linking programs together creates a derivative work. If linking creates a derivative work, the GPL applies to the linked program; ^^^^^^^^^^^^^^^^^^^^^ otherwise, the GPL doesn’t apply."
According to this, application Z linking GPL library X and BSD library Y *may* be required to be GPL-redistributable. The is no such requirement on the source code of Y.
If it turns out that Hakyll *is* okay to be BSD3 licensed so long as neither any binary nor the GPL'd work's source is distributed under non-GPL terms, well ... I'll say that the meaning of "BSD licensed" will have become much less reliable, since it means you actually have to trace the genealogy of the libraries you use *all* the way back in order to understand the situation for certain.
How so? To me it's the exact converse: if the author of Hakyll may *not* distribute his work under the BSD license, just because it is intended to be linked with some GPL code, this complicates issues tremendously. I don't think the FAQs you cited is all that confusing: copyright covers the expression of an idea - i.e. actual source code. If my source code doesn't contain bits of somebody elses source code, I can license it as I wish. All my programs are "intended" to be linked with Linux's libc, and "intended" to call into to Linux kernel. That this should imply a particular licensing seems very counterintuitive, would be impossible to police, and would have a tremendous effect on the software ecosystem. -k -- If I haven't seen further, it is by standing in the footprints of giants