On Fri, 2023-02-24 at 12:45 +0100, Borislav Petkov wrote:
> On Thu, Feb 23, 2023 at 05:54:55PM +0000, Edgecombe, Rick P wrote:
> > The proposed Makefile solution seems a bit unusual. What about this
> > less complicated solution to just make this case work?
>
> I like simple. :)
Done!
>
> > So alternatively, why not just always encourage building the
> > headers
> > before running the selftests by warning if
> > ${abs_srctree}/usr/include/linux is not found?
>
> s/encourage/automate/
>
> Imagine this situation: maintainer says: "please run the selftests".
> User says, "uh oh, it fails building, I need to figure that out
> first."
>
> So we should not have users have to figure out stuff if we can code
> it
> to happen automatically for the default case.
>
> If they want something special, then they can do all the figuring out
> they want. :-)
I guess you also could run into the problem of using old headers. Like
say you checkout kernel A, build headers, then check out kernel B and
build selftests. You get surprised by using the old headers from A.
Today the selftests don't seem to depend on having a .config or
anything. But autobuilding does, so its kind of change to how tied in
the selftests are. But maybe it's how it should be for all of them.
Hmm.
