Re: [Qemu-devel] [PATCH 1/4] Add socket/xnet libs to configure for Solaris

[Andreas Färber]

Am 27.03.2012 14:01, schrieb Lee Essen:
> On 27/03/2012 12:31, Andreas Färber wrote:
>> Am 27.03.2012 09:23, schrieb Stefan Hajnoczi:
>>> On Sat, Mar 24, 2012 at 04:26:27PM +0000, Lee Essen wrote:
>>>> libsocket and libxnet are required for base network functionality
>>>> used in os_dep.c, qemu-socket.c, qga/commands-posix.c and cutils.c
>>>>
>>>> Signed-off-by: Lee Essen<address@hidden>
>>>> ---
>>>>   configure |    1 +
>>>>   1 files changed, 1 insertions(+), 0 deletions(-)
>>>>
>>>> diff --git a/configure b/configure
>>>> index 8b4e3c1..152adaa 100755
>>>> --- a/configure
>>>> +++ b/configure
>>>> @@ -471,6 +471,7 @@ SunOS)
>>>>     QEMU_CFLAGS="-D__EXTENSIONS__ $QEMU_CFLAGS"
>>>>     QEMU_CFLAGS="-std=gnu99 $QEMU_CFLAGS"
>>>>     LIBS="-lsocket -lnsl -lresolv $LIBS"
>>>> +  libs_qga="-lsocket -lxnet $lib_qga"
>>>
>>> s/lib_qga/libs_qga/
>>>
>>> BTW this typo is also present in mingw32 libs_qga, I have sent a patch
>>> to fix it.
>>>
>>> So -lxnet isn't required in plain old LIBS?
>>
>> It's a question of generation AFAIU, I didn't like it either. By using
>> the old libs, then due to Solaris' backwards compatibility we are able
>> to run them on older Solaris versions in theory. We should be using the
>> same libs consistently in QEMU, and I don't like double-coding them.
>> Those comments were not yet addressed, just as my suggested subject for
>> the timer patch and the ordering of the patches was deliberately
>> ignored. :/ Since my patience is limited, I plan to fix them up myself
>> before applying them to my Solaris branch and sending a PULL.
> 
> <rant>
> 
> What?  I'm trying here ... I don't understand the ordering comment, your
> suggestion was about putting more meaningful titles, I've tried to do that.
> 
> Blimey ... this isn't my job, this is my own time ... I'm doing this
> because I want to try to make things better and it feels like I'm having
> to jump through ever decreasing hoops.
> 
> I'm new to the whole git patch submission thing (as is obviously
> apparent) ... so give me a break.
> 
> And let's be clear here ... at the moment there is no support for
> Solaris, there are countless fundamental fixes that need to go in before
> it will even get close ... let alone thinking about kvm.
> 
> I've tried very hard not to break any other platform, but still I can't
> even get a single thing applied.
> 
> </rant>
> 
> Ok, since I'm obviously incapable of providing patches in the right
> form, let me know if I can help in any other way. For now I will just
> maintain a separate tree.

Sorry if this was harsh for you, you have indeed been trying and
improving things, but my issue is this:

<rant>

Apart from the C99 patch that has been committed now, QEMU has been
working fine for me as inofficial maintainer of Solaris host support.

KVM was never supported on illumos in upstream QEMU and it's not even in
upstream KVM AFAIK. It might even never be merged due to licensing
issues. So this is a new, optional feature and not a breakage.

Yet you keep pushing for this. You send patches on Friday afternoon and
on Monday noon do a slightly improved repost. This is my job now and I
do not work on it every weekend. I would rather see you not rush things
so much and put more emphasis on quality of submission and investigation
of why, what and how.
People like you have occasionally appeared out of nowhere, submitted a
few patches and left again, leaving two hands full of core contributors
with the code. So it must be easily maintainable for us.

Especially code that does #if oneplatform||anotherplatform is really bad
because it will mean that someone else will soon come and want to add
||thirdplatform.

My main point however is that you keep sending patches in an
egocentrical rather than maintainer-centric way, which we have already
discussed recently with David for pseries. I would've preferred that you
not send everything *you* need for your goal of SmartOS support in one
large series, but a patch to Paolo about qemu-timer (and I was serious
about the prefix notation, there's many good example on the list and I
made it really easy for you to just copy&paste) that I could just ack
and maybe apply through qemu-trivial, a patch about the KVM stuff that
Jan/Marcello et al. could handle, and qemu-ga in a small series that
Michael could handle and I would ack (qemu-ga being unneeded for most
use cases, easy to disable and therefore even less inconvenient than our
broken Darwin host support).

Your saying that you will maintain this in a separate tree now shows me
even more that you have not yet understood what the problem with your
submissions is that I have been trying to guide you to tackle. Maybe
someone else can explain better, e.g. on IRC where some of the
discussions would be much easier to conduct.

</rant>

Andreas