Re: [Qemu-devel] [PATCH] qom: add helpers REGISTER_STATIC_TYPE[S]()

[Eduardo Habkost <ehabkost@redhat.com>]

On Tue, Oct 03, 2017 at 05:01:39PM +0200, Igor Mammedov wrote:
> On Tue, 3 Oct 2017 11:25:01 -0300
> Eduardo Habkost <ehabkost@redhat.com> wrote:
> 
> > On Tue, Oct 03, 2017 at 02:14:46PM +0200, Igor Mammedov wrote:
> > > it will help to remove code duplication in places
> > > that currently open code registration of static
> > > type[s] and remove necessity to declare function
> > > for type_init() to call, when only static types
> > > need to be registered.
> > > 
> > > Signed-off-by: Igor Mammedov <imammedo@redhat.com>
> > > ---
> > > I'm going to use it for CPU types in followup patches
> > > 
> > > CC: ehabkost@redhat.com
> > > ---
> > >  include/qemu/module.h | 10 ++++++++++
> > >  include/qom/object.h  | 10 ++++++++++
> > >  qom/object.c          |  9 +++++++++
> > >  3 files changed, 29 insertions(+)
> > > 
> > > diff --git a/include/qemu/module.h b/include/qemu/module.h
> > > index 56dd218..60dd632 100644
> > > --- a/include/qemu/module.h
> > > +++ b/include/qemu/module.h
> > > @@ -52,6 +52,16 @@ typedef enum {
> > >  #define type_init(function) module_init(function, MODULE_INIT_QOM)
> > >  #define trace_init(function) module_init(function, MODULE_INIT_TRACE)
> > >  
> > > +#define REGISTER_STATIC_TYPES(t, nr)                                        \  
> > 
> > I'm unsure about the name.  We already have a declarative way to
> > register machine types (DEFINE_MACHINE), maybe use "DEFINE_*" for
> > consistency?
> > 
> > Also, this function won't work if t isn't a static variable, will
> > it?  The "STATIC" part looks redundant.
> > 
> > In other words, how about "DEFINE_TYPE" and "DEFINE_TYPES"?
> 'define' word remind me more of declaration but here we actually
> register types that's why 'register' looks better to me as it says
> exactly what's going here.

Making it look like a declaration and not an action like
"register" is on purpose, as it's not a function call anymore:
it's a declaration that will make the type get registered
automatically.

> 
> > 
> > 
> > > +static void do_qemu_init_ ## t(void)                                        \
> > > +{                                                                           \
> > > +    type_register_static_array(t, nr);                                      \  
> > 
> > qemu/module.h doesn't include qom/object.h.  I think this should
> > go to either qom/object.h or a new header file.
> I could move it to qom/object.h
> 
> > 
> > > +}                                                                           \
> > > +type_init(do_qemu_init_ ## t)
> > > +
> > > +#define REGISTER_STATIC_TYPE(t)                                             \
> > > +    REGISTER_STATIC_TYPES(t, 1)
> > > +
> > >  #define block_module_load_one(lib) module_load_one("block-", lib)
> > >  
> > >  void register_module_init(void (*fn)(void), module_init_type type);
> > > diff --git a/include/qom/object.h b/include/qom/object.h
> > > index f3e5cff..17fcadd 100644
> > > --- a/include/qom/object.h
> > > +++ b/include/qom/object.h
> > > @@ -789,6 +789,16 @@ Type type_register_static(const TypeInfo *info);
> > >  Type type_register(const TypeInfo *info);
> > >  
> > >  /**
> > > + * type_register_static_array:
> > > + * @infos: The array of the new type #TypeInfo structures.
> > > + * @nr_infos: number of entries in @infos
> > > + *
> > > + * @infos and all of the strings it points to should exist for the life time
> > > + * that the type is registered.
> > > + */
> > > +void type_register_static_array(const TypeInfo *infos, int nr_infos);  
> > 
> > This new function is probably less controversial than the new
> > module.h-based code.  Maybe worth moving to a separate patch so
> > it can be merged earlier?
> > 
> > (If I were you, I would just use type_register_static_array()
> > inside the existing *_cpu_register_types() functions by now, so
> > this won't block the whole series, and worry about
> > REGISTER_STATIC_TYPE*/DEFINE_TYPE* later.)
> it would be nice to have both in one series so that
> we won't have to touch/rewrite more than once code where
> static cpu types are registered.
> 
> If you prefer I can do what you suggest and keep
> *_cpu_register_types() functions
> but I can't promise to return back to cleaning them up
> or to REGISTER_STATIC_TYPE*/DEFINE_TYPE, so probably
> someone else will have to pickup that work.

No problem to me, if this is left for later.  The macro saves
only 3 lines of code.

I think REGISTER_STATIC_TYPE*/DEFINE_TYPE* is a good idea, but
considering that QOM types are not the only kind of entity we
register at module_init(), we need to pick a style that can be
consistently used by other subsystems (e.g. for QemuOpts,
BlockDrivers, and TraceEvents).  This means this could take a
little longer to be discussed, and I guess you wouldn't like the
discussion to block the cleanups in this series.

> 
> > > +
> > > +/**
> > >   * object_class_dynamic_cast_assert:
> > >   * @klass: The #ObjectClass to attempt to cast.
> > >   * @typename: The QOM typename of the class to cast to.
> > > diff --git a/qom/object.c b/qom/object.c
> > > index 3e18537..40b1729 100644
> > > --- a/qom/object.c
> > > +++ b/qom/object.c
> > > @@ -151,6 +151,15 @@ TypeImpl *type_register_static(const TypeInfo *info)
> > >      return type_register(info);
> > >  }
> > >  
> > > +void type_register_static_array(const TypeInfo *infos, int nr_infos)
> > > +{
> > > +    int i;
> > > +
> > > +    for (i = 0; i < nr_infos; i++) {
> > > +        assert(type_register_static(&infos[i]));  
> > 
> > Why the assert()?  type_register_static() is already guaranteed
> > to never return NULL.
> according to doc comment type_register_static might return 0 on failure
> so qemu should die in that case.

We have absolutely no callers of type_register*() that check its
return value, and the existing implementation guarantees it will
never return NULL.  Instead of making
type_register_static_array() more commplex, we could just update
the documentation.

> 
> > 
> > This will break if compiled with -DNDEBUG: assert() won't even
> > evaluate the expression if NDEBUG is defined.
> > 
> > > +    }
> > > +}
> > > +
> > >  static TypeImpl *type_get_by_name(const char *name)
> > >  {
> > >      if (name == NULL) {
> > > -- 
> > > 2.7.4
> > >   
> > 
> 

-- 
Eduardo