Re: [Qemu-devel] [PATCH 3/3] qapi: convert sendkey

[Anthony Liguori <aliguori@us.ibm.com>]

On 05/25/2012 08:06 AM, Luiz Capitulino wrote:
> On Fri, 25 May 2012 08:34:54 +0100
> "Daniel P. Berrange"<berrange@redhat.com>  wrote:
>
>> On Fri, May 25, 2012 at 02:20:33PM +0800, Amos Kong wrote:
>>> On 25/05/12 11:51, Eric Blake wrote:
>>>> On 05/24/2012 09:32 PM, Amos Kong wrote:
>>>>> Convert 'sendkey' to use. do_sendkey() depends on some variables
>>>>> in monitor.c, so reserve qmp_sendkey() to monitor.c
>>>>> Rename 'string' to 'keys', rename 'hold_time' to 'hold-time'
>>>>>
>>>>> Signed-off-by: Amos Kong<akong@redhat.com>
>>>>
>>>>> +##
>>>>> +# @sendkey:
>>>>> +#
>>>>> +# Send keys to VM.
>>>>> +#
>>>>> +# @keys: key sequence
>>>>> +# @hold-time: time to delay key up events
>>>>> +#
>>>>> +# Returns: Nothing on success
>>>>> +#          If key is unknown or redundant, QERR_INVALID_PARAMETER
>>>>> +#          If key is invalid, QERR_INVALID_PARAMETER_VALUE
>>>>> +#
>>>>> +# Notes: Send @var{keys} to the emulator. @var{keys} could be the name of the
>>>>> +#        key or the raw value in either decimal or hexadecimal  format. Use
>>>>> +#        @code{-} to press several keys simultaneously.
>>>>> +#
>>>>> +# Since: 0.14.0
>>>>> +##
>>>>> +{ 'command': 'sendkey', 'data': {'keys': 'str', '*hold-time': 'int'} }
>>>>
>>>> Rather than making 'keys' a free-form string where qemu then has to
>>>> parse '-' to separate keys, should we instead make it a JSON array?  For
>>>> example,
>>>
>>>
>>> Anthony, Luiz, Daniel,  what's your opinion?
>>
>> Using a JSON array for the key names does seem like the most
>> natural way to model this. A good rule of thumb is that the
>> implementation of a command should not need to further
>> parse the individual parameter values. Using a magic string
>> encoding instead of the JSON array requires such extra special
>> case parsing.
>
> That's true, and I agree this is better.
>
> Just would like to point out that we can't go too far on improving
> HMP-inherited commands, as our goal here is to convert most (or every single)
> HMP commands to QMP. If we go far on improving commands, we'll get stuck as we
> did some time ago.

I agree.  But obviously, changing the parameter syntax is straight forward 
enough.  I think we rat hole when we change the command semantics though.

>
> Btw, I remember someone saying that when libvirt wants to use a HMP command it
> first checks if the command exists in QMP before using passthrough. In that case,
> how will libvirt behave if we change the arguments?

I don't think libvirt can possibly be assuming that a QMP command exists that 
it's never seen before...  So hopefully this isn't a blind check.  If it is, 
it's a libvirt bug.

Regards,

Anthony Liguori

>