On Tue, Oct 20, 2026 at 01:04:03PM +0100, Gary Guo wrote: > On Thu Oct 15, 2026 at 1:39 PM BST, Daniel Almeida wrote: > > `IoCapable<T>` provides both `io_read` and `io_write`, so a backend that > > supports accesses of type `T` supports them in both directions, and every > > (non-copy) `Io` method that reads or writes a value requires both. There > > is no way to express a backend, or a view, that can only be read or only > > be written. > > > > Thus, replace `IoCapable<T>` with an `IoReadable<T>` trait holding > > `io_read` and an `IoWritable<T>` trait holding `io_write`, and implement > > both for all existing backends. The non-copy `Io` methods that read a > > value now require `IoReadable` and those that write one require > > `IoWritable`. `update` and `try_update` read and then write, so they > > require both. > > > > The copy methods use `IoCopyable` and are not changed here. > > > > No functional change intended. This prepares for read-only and write-only > > views, added in a following patch. > > I think by Rust convention these should be named `IoRead`/`IoWrite`, but we've > aleady named this `IoCapable` and I further made the mistake of adding > `IoCopyable`... > > So I guess I'm not in a good position to ask for name change, but we should > consider it. Apart from the name:
IMO the -able part is quite important. As a reader, seeing `IoRead`/`IoWrite` doesn't tell me enough about what they actually mean on their own. > Reviewed-by: Gary Guo <[email protected]> > > Best, > Gary > > > > > Signed-off-by: Daniel Almeida <[email protected]> > > --- > > rust/kernel/dma.rs | 12 ++++-- > > rust/kernel/io.rs | 101 > > +++++++++++++++++++++++++++++--------------------- > > rust/kernel/pci/io.rs | 10 +++-- > > 3 files changed, 75 insertions(+), 48 deletions(-) >
