Re: [PATCH v1 2/3] io: Add zerocopy and errqueue

[Peter Xu]

On Wed, Sep 08, 2021 at 05:13:40PM -0300, Leonardo Bras Soares Passos wrote:
> On Tue, Sep 7, 2021 at 1:44 PM Peter Xu <peterx@redhat.com> wrote:
> >
> > On Thu, Sep 02, 2021 at 03:59:25AM -0300, Leonardo Bras Soares Passos wrote:
> > > I also suggested something like that, but I thought it could be good if 
> > > we could
> > > fall back to io_writev() if we didn't have the zerocopy feature (or
> > > the async feature).
> > >
> > > What do you think?
> >
> > That fallback looks safe and ok, I'm just not sure whether it'll be of great
> > help.  E.g. if we provide an QIO api that allows both sync write and 
> > zero-copy
> > write (then we do the fallback when necessary), it means the buffer 
> > implication
> > applies too to this api, so it's easier to me to just detect the zero copy
> > capability and use one alternative.  Thanks,
> >
> > --
> > Peter Xu
> >
> 
> I was thinking this way (async method with fallback) we would allow code using
> the QIO api to just try to use the io_async_writev() whenever the code
> seems fit, and
> let the QIO channel layer to decide when it can be used
> (implementation provides,
> features available), and just fallback to io_writev() when it can't be used.

Sure, I think it depends on whether the whole sync/async interface will be the
same, e.g., when async api needs some callback registered then the interface
won't be the same, and the caller will be aware of that anyways.  Otherwise it
looks fine indeed.  Thanks,

-- 
Peter Xu