First of all, thanks for taking a look!!
> > prelude::*,
>
> Should this be expanded too if we want to discourage wildcard imports ?
The prelude is a bit special; it provides a subset of items that most
users will want to include into the namespace. You're right that it
goes against the rule of discouraging wildcard imports, and in fact
that's why it almost entirely consists of extension traits (e.g.
DeviceMethods, SysBusDeviceMethods, ObjectCast, ObjectDeref,
ObjectClassMethods). You shouldn't have to name the traits at all if
the API works, and having them in the prelude lowers the cognitive
load.
There are a few more symbols that are imported in the prelude: for
example the BqlCell and BqlRefCell, the ObjectType and IsA traits, or
the qom_isa macro; those will be used almost everywhere.
The exact shape of the prelude might change in the future, since
qemu_api might get split into separate crates to enable using Rust in
tools (which do not have qdev). For example there could be separate
block, qdev and qom preludes.
In any case, it's an exception and for this reason the amount of
non-extension-traits symbols that it includes should be limited.
> > qdev::{DeviceClass, DeviceImpl, DeviceState, Property,
> > ResettablePhasesImpl},
> > qom::{ClassInitImpl, ObjectImpl, ParentField},
> > + sysbus::SysBusDevice,
>
> This addition seems distinct from the bindings::* removal. Should it
> have been a separate patch ?
SysBusDevice is re-exported from sysbus, similar to DeviceState or
VMStateDescription that you see in the context. This is done because
in general using "bindings" is an anti-pattern; you should use APIs
that have been wrapped properly as is the case for SysBusDevice.
The tests use some "bindings::foo" symbols either because they're
setting up the test environment (module_call_init) or because they're
intentionally bypassing the high-level API (object_new/unref). The
only case where something isn't yet covered by safe wrappers is
qdev_prop_bool.
In fact, a lot of the structs that the tests need were imported twice,
once from bindings and once from e.g. qdev or vmstate. SysBusDevice
was the odd one out that was imported only from bindings, so it has
added.
Paolo
> > vmstate::VMStateDescription,
> > zeroable::Zeroable,
> > };
>
> With regards,
> Daniel
