[Spice-devel] [PATCH 4/9] build-sys: bump current, as ABI is going to be broken next

[Alon Levy]

>>>>>>> I'd rather we avoid breaking ABI 'just' for the NBD channel.
>>>>>>
>>>>>> What do you think is the correct criteria?
>>>>>
>>>>> My criteria would be to never break it, so not really useful. I'd
>>>>> tend to return the question, if we break the ABI now (which hasn't
>>>>> been done in recent times), where do we stop? Some stuff in the
>>>>> Opus patches could also be made easier by breaking ABI, should we
>>>>> break ABI a second time there?
>
> I don't know. We can always keep both, in this case add a second
> add_watch function (add_watch_ext..) and possibly deprecate the former
> later. This is of course ugly, but doesn't break the ABI.
>>>
>>> It's changing the size of SpiceCoreInterface, that would break it too.
>>> We would probably need a different struct interface.
>>>
>>> Another problem is SpiceBaseInterface, where I added a user_data
>>> pointer. This could be hidden in a private struct and accessors.
>
>> I meant BaseInstance.

Well, we suck. This instance could have been passed to qemu/users via
pointer and not embedded in any structs, and we wouldn't have this
problem. We could do it now, which would also break the ABI. Or we could
have another set of API's,
spice_new_{mouse,char_device,kbd,tablet,migrate,playback,record}_instance,
to be used with the new watch_add & timer_add.

Not sure it is worth the work.

>
>>>>>
>>>>> Christophe
>>>>>
>
>>
>>
>>
>> --
>> Marc-André Lureau
>
>
>