On Mon, Sep 11, 2017 at 07:56:19PM +0200, David Hildenbrand wrote: > > >>>> > >>>> #endif > >>> > >>> Why not use typedefs.h? > >> > >> See Markus's reply. But, maybe it's even better to use S390CPU* and > >> include target/s390x/cpu-qom.h, which by design provides as little > >> definitions as needed. > > > > I don't see an argument against moving typedef CPUS390XState to > > typedefs.h in Markus' reply. I see one argument for it (reducing > > the need for non-cyclic includes). > > > > cpu-qom.h includes cpu.h, so I don't know why using S390CPU* > > would solve any problem. I don't disagree about changing the > > function to use S390CPU* eventually, but it would still require > > us make a choice between: a) including the header where the > > typedef name is declared (cpu.h or cpu-qom.h); or b) moving the > > typedef name declaration to typedefs.h. > > It includes qom/cpu.h, not cpu.h. That's why using cpu-qom.h for such > typedefs works (see v4). >
Oops, I was looking at an older tree (before commit 347b1a5c). You're right, sorry for the noise. -- Eduardo