On Thu, Jun 13, 2024 at 04:10:12PM -0400, Steven Rostedt wrote:
> On Thu, 13 Jun 2024 13:04:20 -0700
> Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> > On Thu, 13 Jun 2024 15:34:02 -0400 Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> >
> > > On Thu, 13 Jun 2024 22:22:18 +0300
> > > Alexey Dobriyan <adobriyan@xxxxxxxxx> wrote:
> > >
> > > > g++ doesn't like forward enum declarations:
> > > >
> > > > error: use of enum ‘E’ without previous declaration
> > > > 64 | enum E;
> > >
> > > But we don't care about g++. Do we?
> >
> > It appears that g++ is a useful enum declaration detector.
> >
> > I'm curious to know how even the above warning was generated. Does g++
> > work at all on Linux?
With out-of-tree patch, yes.
What happens is that "enum E;" works in C but doesn't work in C++.
The fix (in C++) is to either delete, or change to "enum E:int;".
The same applies to
const struct S s;
const struct S s = {};
First declaration is compile error in C++, sometimes it can be deleted.
This patch is some "unused" parts merged together because it doesn't
make sense to split this much -- every chunk is independent of each
other.
> > > I would make that a separate patch.
> >
> > What are you referring to here?
>
> The enum change should be separate from the struct changes.
>
> >
> > > >
> > > > Delete those which aren't used.
> > > >
> > > > Delete some unused/unnecessary forward struct declarations for a change.
> > >
> > > This is a clean up, but should have a better change log. Just something
> > > simple like:
> > >
> > > Delete unnecessary forward struct declarations.
> >
> > Alexey specializes in cute changelogs.
>
> eh
Steven is right. That's what my literature teacher said in high school.
> > I do have a concern about the patch: has it been tested with all
> > possible Kconfigs? No. There may be some configs in which the forward
> > declaration is required.
> >
> > And... I'm a bit surprised that forward declarations are allowed in C.
> > A billion years ago I used a C compiler which would use 16 bits for
> > an enum if the enumted values would fit in 16 bits. And it would use 32
> > bits otherwise. So the enumerated values were *required* for the
> > compiler to be able to figure out the sizeof. But it was a billion
> > years ago.
>
> Well, I only looked at the one change in ftrace.h which has a
> "struct seq_file;" that is not used anywhere else in the file, so that
> one definitely can go.
It was tested on arm64 allmodconfig too.
OK if this is concern, I could dust off my compile test farm.
