Re: [Qemu-devel] Should we auto-generate IDs?

[Programmingkid]

On Aug 27, 2015, at 1:39 AM, Markus Armbruster wrote:

> Programmingkid <address@hidden> writes:
> 
>> On Aug 26, 2015, at 6:08 PM, John Snow wrote:
>> 
>>> 
>>> 
>>> On 08/26/2015 05:48 PM, Programmingkid wrote:
>>>> 
>>>> On Aug 26, 2015, at 2:45 PM, Peter Maydell wrote:
>>>> 
>>>>> On 26 August 2015 at 18:16, Programmingkid
>>>>> <address@hidden> wrote:
>>>>>> That is assuming they have the time and/or the interest in
>>>>>> solving this problem. I
>>>>>> suppose giving them some time to respond would be reasonable. I'm
>>>>>> thinking if
>>>>>> no consensus has been reached in one weeks time (starting today),
>>>>>> we turn to
>>>>>> Peter Maydell for the answer. Hopefully he will just pick which
>>>>>> of the patches he
>>>>>> likes the best. Judging by how long this problem has been ongoing, 
>>>>>> someone
>>>>>> pick the answer is probably the best we can expect.
>>>>> 
>>>>> This is the kind of thing I strongly prefer to leave to the
>>>>> relevant subsystem maintainer(s). My opinion is not worth
>>>>> a great deal since I don't have a strong familiarity with
>>>>> this bit of QEMU.
>>>> 
>>>> It looks unreasonable to assume any consensus can be reached over this 
>>>> issue.
>>>> The easy thing to do is to just let each maintainer deal with this problem
>>>> his own way. 
>>>> 
>>> 
>>> What feedback was there that seemed insurmountable? Last I talked to
>>> Jeff Cody he said there was no "overwhelming pushback" against his
>>> patches, just a list of concerns.
>> 
>> Markus Armbruster sent me four different threads each trying to solve
>> this problem.
>> Some of those threads were many years old. The situation is the same then as 
>> it
>> is now. There is no judicator to decide how this problem is to be
>> solved. Expecting
>> all the maintainers to agree on one patch is unrealistic. 
>> 
>>> This doesn't sound like a dead end so much as it sounds like we haven't
>>> planned the feature enough yet.
>> 
>> The threads did have some really good patches that did seem to solve
>> the problem.
>> I could send you the threads if you haven't read them yet.
> 
> Back then was then and now is now.  I understand your impatience to get
> stuff done, but things may have changed, people may have changed.  We
> really need to talk it over one more time.  If we can reach rough
> consensus, swell.  If we can't, we can still narrow the scope to
> subsystems where we can.
> 
>>>> Markus:
>>>> I know you really wanted a single ID generating system, but it just
>>>> isn't going
>>>> to happen. I will make a patch that would only effect USB devices. All 
>>>> other
>>>> devices would be untouched. At least the device_del problem will be solved.
>>>> 
>>> 
>>> I think this is being unnecessarily hasty. We should make sure that an
>>> auto-generated ID system does not create problems for other areas of
>>> code before we rush ahead with one to solve a single problem.
> 
> Right.
> 
>> I would make sure my patch only affects USB devices. No other systems
>> would be affected. 
> 
> Device IDs are in the qdev/QOM subsystem, so that's the subsystem you
> have to attack should QEMU-wide consensus remain elusive.  The USB
> subsystem is merely a user of qdev/QOM here.

I could make a patch that detects if the device is a USB device, then
give it an ID if none was specified. This patch would have no affect on other
non-USB devices. We could implement this patch for the time being. Then *if*
an universal device ID system does become available, then just start using 
it instead.

> 
>>> Let's give the universal approach some more time before we jump to the
>>> conclusion that it's impossible.
>> 
>> I suppose 5 more years will do ;)
>> 
>> Maybe that's too soon...
> 
> I understand your frustration with our collective inability to solve
> this problem for 5+ years.  Heck, I share it!  I certainly don't want
> the problem to be shelved *again*.  But a proper discussion should take
> us perhaps days, at worst weeks (KVM Forum was last week, several folks
> are taking time off right now), certainly not months, let alone years.

Ok. I know you don't want a deadline. What about a goal. Our goal could be
solving this universal device ID problem within two weeks, or we switch over
to plan b. Plan b would be giving only USB devices an ID if none were
specified. That way device_del would always work, and we can finally move
on from this device ID problem.

> We can afford time for discussion.  That's how we work.  Rough consensus
> and running code.  You're welcome to provide running code early, just be
> prepared for it to be thrown out and redone :)


Is it possible you would accept a patch that *only* gives USB devices an ID if 
none were specified. If you would, I will gladly make this patch and send it 
in. 

Given all the different views expressed so far, this universal device ID problem
will not likely be solved soon.