On Mon, Mar 3, 2025 at 10:53 PM Jani Nikula <jani.nikula@xxxxxxxxx> wrote: > > On Mon, 03 Mar 2025, Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote: > > On Mon, Mar 3, 2025 at 7:02 PM Jani Nikula <jani.nikula@xxxxxxxxx> wrote: > >> > >> On Mon, 03 Mar 2025, Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote: > >> And one of the underlying goals is to make for minimal headers with > >> minimal includes and minimal dependencies, preferring forward > >> declarations over includes, splitting functionality by header, etc. It's > >> just that doing that often leads to broken headers, unless you actually > >> build test them... and here we are. > > > > > > What I learned from my last attempt is that we cannot avoid > > false positives without adding a lot of exceptions. > > All of the drm core, xe and i915 headers build fine without > exceptions. There are no false positives. (*) > > > We can never be certain whether you are making DRM headers > > self-contained for valid reasons or for hypothetical, invalid ones. > > Please enlighten me. What are hypothetical, invalid reasons for making > headers self-contained? See this thread: https://lore.kernel.org/all/20190718130835.GA28520@xxxxxx/ When CONFIG_BLOCK=n, it does not make sense to ensure <linux/iomap.h> is self-contained. This is just one example. I am pretty sure I observed more false-positives in header compile tests. > > IMO headers should almost invariably be self-contained, instead of > putting the burden on their users to include other headers to make it > work. It's a PITA in a project the size of the kernel, or even just the > drm subsystem, to track these cases when you modify includes in either > users or the headers being included. > > The exception to this are headers that are not to be included directly > by users, but rather by other headers as an implementation detail. There > may be such cases in include/linux, but not under include/drm. This results in a false check for include/linux/. I don’t see much sense in doing this exceptionally for include/drm/ after we've learned that it doesn't work globally. > > BR, > Jani. > > > (*) Fine, there's one *intentional* special case in i915. > > -- > Jani Nikula, Intel -- Best Regards Masahiro Yamada
