I'm about to board a plane but I at least want to comment that we should all give each other the benefit of the doubt in a situation like this. Everyone on this thread is a long term contributer to QEMU that has more than earned the right not to be accussed of impropriety. Let's all take a deep breathe and tone down this discussion in this thread by assuming that noone is intentionally doing anything wrong
On Apr 5, 2012 9:17 AM, "Paolo Bonzini" <
address@hidden> wrote:
Il 05/04/2012 15:31, Andreas Färber ha scritto:
> Here's how I see it:
>
> * You add a realize callback to ObjectClass like I did, you add the
> Error** parameter that was requested as feedback to mine.
> * You add a static object_realize() method that clashes with my
> introducing it as a public wrapper function.
> * You introduce a function object_get_realized() like I did, only you
> defer your implementation to object_is_realized() which I didn't have
> and used a new bool realized instead of a state enum (since I left qdev
> unmodified).
> * You introduce a function object_set_realized() like I did, only you
> change the logic to also do unrealize.
> * You introduce additional stuff that I don't particularly care about.
Since we're nitpicking, I also do correct error propagation.
> So my point is, whether you've read some patch or not, I just can't
> understand why you couldn't wait a week for me to resend the updated
> version
Because a week is a long time 10 days before the feature freeze, and
(via object_is_realized and a few other small bits) the whole series
depends on the implementation of realized.
> While having unrealize and propagation is certainly nice, the most
> serious issue with yours I see is that it doesn't offer me a way to
> actually make use of it outside qdev, so that *I* am left with no
> benefit from your patch!
Can you explain? I definitely would need to fix this.
> Some practical thoughts on how to align both approaches would be helpful
> here. For starters, should I name my function object_realize_nofail()
> instead?
Yes, that would be an idea. I would hope that long-term there would be
only one object_realize call during in initial machine creation (i.e.
except for hot-plug), but it would be fine as a start.
> And could you prefer _one over _1 in your patch please?
Yes.
> If your problem is Signed-off-by specifically, feel free to invent some
> inofficial tag such as Inspired-by or Derived-from-commit-message-by or
> resort to a textual reference.
I can add the SoB, no problem.
Paolo