Re: [Qemu-devel] [4367] Align file accesses with cache=off (Kevin Wolf, Laurent Vivier)

["Blue Swirl" <blauwirbel@gmail.com>]

On 5/7/08, Kevin Wolf <kwolf@suse.de> wrote:
> Jamie Lokier schrieb:
>
>
> > Kevin Wolf wrote:
> >
> > > Btw, it is quite interesting to see that a serious discussion of a patch
> happens only if it is already committed. This could have been discussed a
> week ago when we agreed to go in the apparently wrong direction. And the
> patch has been on the list much longer than this one week.
> > >
> >
> > IIRC, the same issues (recursive callback vs. queued callback,
> > problems with direct and non-direct I/O in flight at the same time)
> > were raised before it was committed, but it was committed anyway.
> >
>
>  No, nobody mentioned the recursion problem. And you were talking about
> problem with two different file descriptors for one file, not about the
> fcntl solution. Ok, might also be that the hints were just not explicit
> enough for me. ;-)
>
>  But even if so, this is more of a general feeling about how patches are
> handled and not only related to this patch.

Good point. For most patches sent to the list I don't have strong
opinions, for example I don't know the necessary technical background
or the patch could affect areas that I'm not interested in. Maybe if I
comment on the patches it gives a wrong signal about the future
prospects for the patch getting committed. I still think it is a waste
of resources if the effort made for the patches gets lost or don't get
much review like what is happening here very often.

Usually I run a set of tests before committing. From the amount of
breakage fixes and reverts I get the impression that some of the other
developers test less. Though testing will not catch all deeper
problems like in this case.

Maybe more developers with strong focus on x86/x86_64 host and target
would help.