Re: [Qemu-devel] [PATCH 1/2] Fix guest agent build with simpletrace

[Michael Roth <mdroth@linux.vnet.ibm.com>]

On 08/29/2011 07:28 AM, Stefan Hajnoczi wrote:
> On Sun, Aug 28, 2011 at 10:08 PM, Blue Swirl<blauwirbel@gmail.com>  wrote:
>> On Sun, Aug 28, 2011 at 6:13 PM, Lluís<xscript@gmx.net>  wrote:
>>> Blue Swirl writes:
>>>
>>>> On Sat, Aug 27, 2011 at 5:56 PM, Lluís<xscript@gmx.net>  wrote:
>>>>>>> I sent a patch that should fix it for everybody linking with the tracing
>>>>>>> objects:
>>>>>>>
>>>>>>> http://lists.gnu.org/archive/html/qemu-devel/2011-08/msg03150.html
>>>>>
>>>>>> With your patch, there are warnings from linker:
>>>>>> ../qemu-timer-common.o: warning: multiple common of `use_rt_clock'
>>>>>> ../qemu-timer-common.o: warning: previous common is here
>>>>>
>>>>> Ah, yes. These extra errors are fixed by the duplicate elimination patch
>>>>> :)
>>>>>
>>>>> http://lists.gnu.org/archive/html/qemu-devel/2011-08/msg02987.html
>>>>>
>>>>> So, you need both to keep it clean.
>>>
>>>> Using the sort function looks hackish to me. Maybe the linkage should
>>>> be controlled by configure instead?
>>>
>>> What do you mean? Moving the logic for selecting the object files to
>>> link with on each top-level target out into the configure?
>>
>> Add CONFIG_QEMU_TIMER, configure sets it to 'y' when it is needed by
>> simpletrace or other cases.
>
> The $(sort) approach is simpler because it is implicit.  I'm not sure
> that explicitly managing these dependencies is necessary.  But the

It also mirrors how most projects handle duplicate header includes using 
a guard.

Plus, it really sucks to not be able to create a self-sufficient group 
of objects just because someone that includes your group happens to also 
include a common object (in this case, common-obj-y). trace-obj-y should 
absolutely be able to note qemu-timer-common.o as an explicit dependency 
regardless of whether or not it happens to get linked in by something 
else in the common case.

Plus with talk of breaking up QEMU into shared objects I think we'd end 
up needing to have self-sufficient object groups anyway.

But, in the meantime, I broke something again so I put together a patch 
with Blue's suggestions that seems to do the trick fairly reasonably. 
I'll send it as a response for reference, though I'd really prefer the 
$(sort) approach.

> configure approach works for me too.
>
> Blue: Are you going to post the CONFIG_QEMU_TIMER patch?
>
> Stefan
>