Re: [Qemu-devel] [PATCH v1 00/14]: Initial QObject conversion

[Gerd Hoffmann]

On 10/02/09 15:47, Luiz Capitulino wrote:
> On Fri, 02 Oct 2009 14:47:02 +0200
> Gerd Hoffmann<address@hidden>  wrote:
>
> >     Hi,
> >
> > >    Some people have suggested that we should have a better error handling
> > > in the Monitor, in the meaning that error information should be correctly
> > > propagated and handled in order to be used by the Monitor Protocol and
> > > the existing user protocol.
> >
> > A bunch of code paths can be called from both monitor and non-monitor
> > contexts (network configuration, device hotplug).  Right now there are
> > the qemu_error*() functions to make sure the error messages appear on
> > the correct place (monitor or stderr) without having to pass through a
> > Monitor pointer all the way down.
> >
> > How do you plan to handle this in the new world of monitor error reporting?
>
>   Good question.
>
>   The first thing to bear in mind is that MonitorError is not just about
> error reporting, but more importantly, it makes errors have a common
> structure so that they can be emitted by the Monitor Protocol.

So maybe they shouldn't be named MonitorError in the first place?
And make sure the design works for non-monitor contexts too?
Having the user_error callback in struct mon_cmd_t doesn't make sense 
from that point of view for example.

Why user_error is needed in the first place btw?  To maintain 
backward-compatible error message formating?

>   One way to solve the problem could be to move the MonitorError pointer
> to the Monitor struct.

I think that is a good idea nevertheless.

> This way, when there's a ERR_SINK_MONITOR
> qemu_error() would fill MonitorError instead of calling monitor_vprintf().
>
>   The bad news though, is that we would have to change all qemu_error()
> calls to have "structured" errors instead of pretty printing, that is,
> create macros, define its errors data and write functions to do
> pretty printing....

Or add a qemu_error_structed() variant for smooth switchover and have 
qemu_error() use a generic error code, so you don't have to switch over 
all at once.  There are not *that* many qemu_error users though, so just 
converting them all in one go might not be too hard.

cheers,
  Gerd