Re: [Qemu-devel] [RFC 0/8]: QError v2

[Luiz Capitulino <lcapitulino@redhat.com>]

On Thu, 12 Nov 2009 08:44:03 -0600
Anthony Liguori <aliguori@linux.vnet.ibm.com> wrote:

> Luiz Capitulino wrote:
> >
> >> #define QERR_DEVICE_ALREADY_OPEN "{'class': 'DeviceAlreadyOpen', 'data' 
> >> : {'bus_num': %d, 'addr': %d}"
> >>
> >> qemu_error_new(QERR_DEVICE_ALREADY_OPEN, bus_num, addr);
> >>     
> >
> >  What about DeviceAlreadyOpen errors with a different argument list?
> >   
> 
> Why would you have this?  That would seem like a problem to me.  I think 
> the errors need to be very well structured (just like everything else in 
> QMP).

 This can happen with errors that carry specific info which are different
among subsystems, eg. USB device info vs. PCI device info.

 We could have 'USBDeviceAlreadyOpen', but then I think the class
attribute will lose generality.

> >> That gives us a nice simple interface with full error checking on the 
> >> parameters.
> >>     
> >
> >  I've said this is not so simple because people writing those macros
> > would find out that the 'class' or 'data' _keys_ are missing or incorrect
> > only at run-time, when the error is triggered.
> >   
> 
> Sure but introducing new types of errors is not the common case.  Using 
> existing errors is the common case.

 Right, although there's a long road until we stabilize.

> >> For human readable strings, I'd suggest making a table somewhere else 
> >> that looked like:
> >>
> >> QErrorStringTable qerror_descriptions[] = {
> >> { QERR_DEVICE_ALREADY_OPEN, "This device at %(bus_num)d.%(addr)d is 
> >> already open." },
> >> ...
> >> };
> >>     
> >
> >  How do you suggest we lookup the table? Doing a strcmp() on
> > QERR_DEVICE_ALREADY_OPEN?
> >   
> 
> We can either change the index on the table to be just the class code or 
> find something more clever.

 I'm working on this and trying to find something.
 
> >> There are a number of advantages to an approach like this.  The table 
> >> can be reused by both in the server and by a client.
> >>     
> >
> >  My suggestions on both problems makes me willing go back to my initial
> > series, which had a table indexed by an error number.
> >   
> 
> I don't understand why.

 I've found other problems with it, let's pretend I didn't mention it. :)