[libvirt] [Qemu-devel] [PATCH 2/2] qapi: deprecate implicit filters

[John Snow]

On 8/15/19 12:48 PM, Kevin Wolf wrote:
> Am 15.08.2019 um 18:07 hat John Snow geschrieben:
>> On 8/15/19 6:49 AM, Kevin Wolf wrote:
>>> Am 14.08.2019 um 21:27 hat John Snow geschrieben:
>>>>
>>>> This might be OK to do right away, though.
>>>>
>>>> I asked Markus this not too long ago; do we want to amend the QAPI
>>>> schema specification to allow commands to return with "Warning" strings,
>>>> or "Deprecated" stings to allow in-band deprecation notices for cases
>>>> like these?
>>>>
>>>> example:
>>>>
>>>> { "return": {},
>>>>   "deprecated": True,
>>>>   "warning": "Omitting filter-node-name parameter is deprecated, it will
>>>> be required in the future"
>>>> }
>>>>
>>>> There's no "error" key, so this should be recognized as success by
>>>> compatible clients, but they'll definitely see the extra information.
>>>>
>>>> Part of my motivation is to facilitate a more aggressive deprecation of
>>>> legacy features by ensuring that we are able to rigorously notify users
>>>> through any means that they need to adjust their scripts.
>>>
>>> Who would read this, though? In the best case it ends up deep in a
>>> libvirt log that nobody will look at because there was no error. In the
>>> more common case, the debug level is configured so that QMP traffic
>>> isn't even logged.
>>>
>>> Kevin
>>>
>>
>> I believe you are right, but I also can't shake the feeling that this
>> attitude ensures that we'll never find a way to expose this information
>> to the end-user. Is this not too defeatist?
> 
> I think the discussed approach that seemed most likely to me to succeed
> was adding a command line option that makes QEMU just crash if you use a
> deprecated feature, and enable that in libvirt test cases (or possibly
> even any non-release builds, though maybe it's a bit harsh there).
> 
>> I think deprecation notices in the QMP stream has two benefits:
>>
>> 1) Any direct usages via qmp-shell or manual JSON connection are likely
>> to see this message in development or testing. I feel the usage of QEMU
>> directly is more likely to increase with time as other stacks seek to
>> work around libvirt.
>>
>> [Whether or not they should is another question, but I believe the
>> current reality to be that people are trying to.]
> 
> I don't know about other people, but as a human user, I don't care about
> deprecation notices. As long as something works, I use it, and once I
> get an error message back, I'll use something else.
> 
> If I manually enter drive_mirror and get a warning back, that doesn't
> tell me that libvirt still does the same thing and needs to be fixed. It
> just tells me that in the future I might need to change the commands
> that I use manually.
> 

That the message we return needs to be *useful* doesn't sound like a
count against it.

> I guess this would still prevent adding new libvirt features that build
> on deprecated QEMU features because some manual testing will be involved
> there. But was this ever a problem?
> 

No, because until recently we didn't deprecate anything.

>> 2) Programmatic deprecation notices can't be presented to a user at all
>> if we don't send them; at least this way it becomes libvirt's problem
>> over what to do with them. Perhaps even just in testing and regression
>> suites libvirt can assert that it sees no deprecation warnings (or
>> whitelist certain ones it knows about.)
>>
>> In the case of libvirt, it's not even necessarily about making sure the
>> end user sees it, because it isn't even necessarily the user's fault --
>> it's libvirt's. This is a sure-fire programmatic way to communicate
>> compatibility changes to libvirt.
> 
> If libvirt uses this to make test cases fail, it could work.
> 

Yeah, I think there's solid use there. I'll continue along in Markus's
thread.

> Kevin
>