On Tue, Dec 7, 2021 at 7:17 PM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Tue, Dec 7, 2021 at 7:10 AM Arnd Bergmann <arnd@xxxxxxxxxx> wrote:
> >
> > From: Arnd Bergmann <arnd@xxxxxxxx>
> >
> > Working towards a cleaner header structure, start by moving the most
> > commonly embedded structures into a single header file that itself
> > has only a minimum set of indirect includes. At this moment, this
> > include structures for
>
> Ugh. I liked your 1/3 patch, but I absolutely detest this one.
>
> It makes no sense to me, and just makes that header file a completely
> random collection of some random structure types.
>
> And I absolutely hate how it splits out the definition of the struct
> from basic core infrastructure (initializers etc random inline
> functions) for said structures.
>
> So no. NAK on this one. I think it's a disaster.
Ok, thank you for taking a look!
Here is what I'd try next based on the feedback:
- reduce patch 1 somewhat to only include the types that have a strong reason
to get moved to linux/types.h, describing them individually but leaving them
it as a single combined patch
- keep patch 3, but leave 'struct super_block' in linux/fs.h for the moment
because of 'struct list_lru'.
- drop patch 2, but instead find a new home for each structure (along with
its initializers) if needed for defining the types in the new linux/fs_types.h
without too many indirect includes. I'll skip the structs that would be needed
later for mm.h/sched.h/device.h/.... for the moment.
For the added headers, do you have a preference for whether to try grouping
them logically or not? I could either split them out individually into many new
headers (xarray_types.h, idr_types.h, percpu_rwsem_types.h, rwsem_types.h,
...), or combine some of them when they define related types.
Arnd
