On Fri, May 02, 2025 at 08:10:30PM -0600, Simon Glass wrote:
> Hi Tom,
>
> On Fri, 2 May 2025 at 09:04, Tom Rini <tr...@konsulko.com> wrote:
> >
> > On Fri, May 02, 2025 at 08:52:37AM -0600, Simon Glass wrote:
> > > Hi Tom,
> > >
> > > On Fri, 2 May 2025 at 08:06, Tom Rini <tr...@konsulko.com> wrote:
> > > >
> > > > On Fri, May 02, 2025 at 07:11:56AM -0600, Simon Glass wrote:
> > > > > Hi Tom,
> > > > >
> > > > > On Thu, 1 May 2025 at 09:33, Tom Rini <tr...@konsulko.com> wrote:
> > > > > >
> > > > > > On Thu, May 01, 2025 at 09:04:45AM -0600, Simon Glass wrote:
> > > > > > > Hi Tom,
> > > > > > >
> > > > > > > On Thu, 1 May 2025 at 08:06, Tom Rini <tr...@konsulko.com> wrote:
> > > > > > > >
> > > > > > > > On Wed, Apr 30, 2025 at 07:04:27PM -0600, Simon Glass wrote:
> > > > > > > > > This file reads from the environment but does not include the
> > > > > > > > > correct
> > > > > > > > > header. Update it.
> > > > > > > > >
> > > > > > > > > Drop the unnecessary config.h while we are here.
> > > > > > > >
> > > > > > > > Are you sure about that? It was explicitly added in commit
> > > > > > > > ab61cc7d98f6 ("board: congatec: Remove <common.h> and add needed
> > > > > > > > includes") probably because of CFG_SYS_... usage.
> > > > > > >
> > > > > > > Well I assumed it wouldn't build if it had it, but perhaps that is
> > > > > > > incorrect? I understood that CFG_ things had to be values and you
> > > > > > > couldn't #idef them?
> > > > > >
> > > > > > The cases where in that series I added config.h were for good
> > > > > > reason,
> > > > > > either failure to build or tricky size change reasons. All were
> > > > > > intentional. An audit-and-remove of config.h would be it's own
> > > > > > series as
> > > > > > yes, there's perhaps not the need now as whatever complex chain was
> > > > > > unwound.
> > > > >
> > > > > Oh dear, OK. I'll respin the series.
> > > > >
> > > > > Hmm but I see you have sent a series that does some similar tweak[1].
> > > > > What was the goal there?
> > > >
> > > > I fixed your problem. Please don't bother re-spinning this series, it's
> > > > not clear what problems it's solving correctly.
> > >
> > > OK thanks. The problem was trying to include net.h in include/efi.h:
> > >
> > > https://patchwork.ozlabs.org/project/uboot/cover/20250501010456.3930701-1-...@chromium.org/
> >
> > Adding headers to headers needs extra care (so that we don't end up with
> > massive implicit include chains).
> >
> > > So long as that works, we're fine. Having said that, even if it were a
> > > problem, we could always refactor things to put the Ethernet decls in
> > > a separate header.
> >
> > Yes, if there's some problem down the line then some refactoring is
> > likely the thing to look at, not adding more headers to C files (the
> > right path is to unwind the header chain and fix C files, your series
> > doesn't remove env.h from any header).
>
> I hope we haven't got confused here.
>
> My series removed env.h from being *transitively* included. I had to
> add it to C files due to that.Yes, which may or may not be helpful. A transitive include in a C file is not terrible, otherwise many files will have 20+ includes. A transitive include in header file, which in turn pulls in another which pulls in another, is a problem. And the header should be unwound and C files then fixed. > Your series doesn't help with including net.h in efi.h - you can try > that and see that it still fails. Perhaps I took it out too quickly then, I'd have sworn I build sandbox with that change added. > But as I said, we can move the decls into a separate file. Yes, that might be the right answer, depending on the problem. I'm not sure adding more headers to headers is good in general. -- Tom
signature.asc
Description: PGP signature
