Re: [Qemu-devel] [PATCH 1/1] QAPI schema: turn example commands/returns into proper JSON

[Markus Armbruster <armbru@redhat.com>]

Cleber Rosa <crosa@redhat.com> writes:

> On 08/08/2017 05:13 PM, Eric Blake wrote:
>> On 08/08/2017 03:53 PM, Cleber Rosa wrote:
>>> Most QMP commands and returns in the QAPI schema documentation
>>> are valid "JSON-based wire format".  A few examples are either
>>> malformed, or contain comments.
>>>
>>> This fixes all the examples command and return data, making them
>>> proper JSON, as they would be received and generated by QEMU's
>>> QMP monitor.
>>>
>>> Signed-off-by: Cleber Rosa <crosa@redhat.com>
>>> ---
>>>  qapi-schema.json     |  9 ++++-----
>>>  qapi/block-core.json | 32 ++++++++++++++++----------------
>>>  qapi/rocker.json     |  5 +----
>>>  3 files changed, 21 insertions(+), 25 deletions(-)
>> 
>> 
>>> +++ b/qapi-schema.json
>>> @@ -2000,8 +2000,7 @@
>>>  #                "host": "127.0.0.1",
>>>  #                "channel-id": 0,
>>>  #                "tls": false
>>> -#             },
>>> -#             [ ... more channels follow ... ]
>>> +#             }
>> 
>> I still wonder if we want SOME sort of markup to make it obvious where
>> we are compressing the example for the sake of brevity, where whatever
>> we use to automate tests based on the docs would know how to recognize
>> that the actual values given in reply to the test can be longer than the
>> documented example.  But I guess we can cross that when we have an
>> automated test where it matters.
>> 
>
> I wonder the same.  Also, we seem to agree that it's a separate and more
> complex problem, to be tackled later.

We can cross that bridge when we get to it.

Any particular reason not to keep the [ ... more channels follow ... ]
until then?

>>> @@ -2039,7 +2038,7 @@
>>>  #
>>>  # -> { "execute": "query-balloon" }
>>>  # <- { "return": {
>>> -#          "actual": 1073741824,
>>> +#          "actual": 1073741824
>>>  #       }

This is a straighforward doc fix.

>> I also suspect that test automation will have to do a lot of filtering,
>> even for commands that don't need to be abbreviated, since some of the
>> examples have pretty arbitrary numbers that will be difficult to
>> reliably reproduce any particular number.
>> 
>
> Yes.  I'm already aware of a couple of use cases that will require
> different types of comparison, including pretty relaxed ones.  Expect
> more about that in a later thread.
>
>> This is a documentation fix, so it could still go in 2.10 - but since we
>> are past -rc2, it's probably just as easy to save it for 2.11.  Either way,
>> 
>> Reviewed-by: Eric Blake <eblake@redhat.com>
>> 
>
> Thanks for the prompt review!