[Qemu-devel] QEMU build broken

[Qemu-devel] QEMU build broken

[Brad Smith]

Commit 0056ae24bc36798fdd96d0b31e217e9f73896736 broke the build of QEMU 
and this affects the 2.0.0 release too.

http://buildbot.b1-systems.de/qemu/builders/default_openbsd_current/builds/700/steps/compile/logs/stdio
http://buildbot.b1-systems.de/qemu/builders/default_openbsd_4.9/builds/891/steps/compile/logs/stdio

MACHINE and MACHINE_ARCH are used to indicate the meta arch and
hardware architecture.

vl.c: In function 'main':
vl.c:3914: error: called object '"amd64"' is not a function

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

Re: [Qemu-devel] QEMU build broken

[Peter Maydell]

On 16 March 2014 13:22, Brad Smith <brad@comstyle.com> wrote:
> Commit 0056ae24bc36798fdd96d0b31e217e9f73896736 broke the build of QEMU and
> this affects the 2.0.0 release too.
>
> http://buildbot.b1-systems.de/qemu/builders/default_openbsd_current/builds/700/steps/compile/logs/stdio
> http://buildbot.b1-systems.de/qemu/builders/default_openbsd_4.9/builds/891/steps/compile/logs/stdio
>
> MACHINE and MACHINE_ARCH are used to indicate the meta arch and
> hardware architecture.

Can we limit the set of system headers we're pulling in to
avoid the header where OpenBSD is defining this? Generally
for specific headers that clash it's nicer to just limit those to
whatever file really needed those defines. I'm assuming it's
not pulled in by one of the POSIX headers...

thanks
-- PMM

Re: [Qemu-devel] QEMU build broken

[Marcel Apfelbaum]

On Sun, 2014-03-16 at 13:26 +0000, Peter Maydell wrote:
> On 16 March 2014 13:22, Brad Smith <brad@comstyle.com> wrote:
> > Commit 0056ae24bc36798fdd96d0b31e217e9f73896736 broke the build of QEMU and
> > this affects the 2.0.0 release too.
> >
> > http://buildbot.b1-systems.de/qemu/builders/default_openbsd_current/builds/700/steps/compile/logs/stdio
> > http://buildbot.b1-systems.de/qemu/builders/default_openbsd_4.9/builds/891/steps/compile/logs/stdio
> >
> > MACHINE and MACHINE_ARCH are used to indicate the meta arch and
> > hardware architecture.
> 
> Can we limit the set of system headers we're pulling in to
> avoid the header where OpenBSD is defining this? Generally
> for specific headers that clash it's nicer to just limit those to
> whatever file really needed those defines. I'm assuming it's
> not pulled in by one of the POSIX headers...
It will be really nice if we could do that,
otherwise I need to get creative and find a suitable name for the machine type.
By the way, do we have some standard QOM prefixes/suffixes for collisions like this?

Thanks,
Marcel

> 
> thanks
> -- PMM

Re: [Qemu-devel] QEMU build broken

[Brad Smith]

On 16/03/14 9:26 AM, Peter Maydell wrote:
> On 16 March 2014 13:22, Brad Smith <brad@comstyle.com> wrote:
>> Commit 0056ae24bc36798fdd96d0b31e217e9f73896736 broke the build of QEMU and
>> this affects the 2.0.0 release too.
>>
>> http://buildbot.b1-systems.de/qemu/builders/default_openbsd_current/builds/700/steps/compile/logs/stdio
>> http://buildbot.b1-systems.de/qemu/builders/default_openbsd_4.9/builds/891/steps/compile/logs/stdio
>>
>> MACHINE and MACHINE_ARCH are used to indicate the meta arch and
>> hardware architecture.
>
> Can we limit the set of system headers we're pulling in to
> avoid the header where OpenBSD is defining this? Generally
> for specific headers that clash it's nicer to just limit those to
> whatever file really needed those defines. I'm assuming it's
> not pulled in by one of the POSIX headers...

It looks like it is being pulled in via include/qemu/sockets.h
which includes netdb.h.

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

Re: [Qemu-devel] QEMU build broken

[Peter Maydell]

On 16 March 2014 14:33, Brad Smith <brad@comstyle.com> wrote:
> On 16/03/14 9:26 AM, Peter Maydell wrote:
>> Can we limit the set of system headers we're pulling in to
>> avoid the header where OpenBSD is defining this? Generally
>> for specific headers that clash it's nicer to just limit those to
>> whatever file really needed those defines. I'm assuming it's
>> not pulled in by one of the POSIX headers...

> It looks like it is being pulled in via include/qemu/sockets.h
> which includes netdb.h.

Ouch, and netdb.h results in this macro getting defined?
That seems like a bit of a namespace grab by OpenBSD's
libc really :-(

-- PMM

Re: [Qemu-devel] QEMU build broken

[Brad Smith]

On 16/03/14 11:03 AM, Peter Maydell wrote:
> On 16 March 2014 14:33, Brad Smith <brad@comstyle.com> wrote:
>> On 16/03/14 9:26 AM, Peter Maydell wrote:
>>> Can we limit the set of system headers we're pulling in to
>>> avoid the header where OpenBSD is defining this? Generally
>>> for specific headers that clash it's nicer to just limit those to
>>> whatever file really needed those defines. I'm assuming it's
>>> not pulled in by one of the POSIX headers...
>
>> It looks like it is being pulled in via include/qemu/sockets.h
>> which includes netdb.h.
>
> Ouch, and netdb.h results in this macro getting defined?
> That seems like a bit of a namespace grab by OpenBSD's
> libc really :-(
>
> -- PMM

netdb.h pulls in sys/param.h which pulls in machine/param.h
which is a MD header and that would define something like..

#define	_MACHINE	amd64
#define	MACHINE		"amd64"
#define	_MACHINE_ARCH	amd64
#define	MACHINE_ARCH	"amd64"

There has been some relatively recent work towards improving
POSIX/SUS compliance and reducing namespace pollution with some
of the headers but it seems this has not been eliminated yet.
But most of the namespace pollution reduction has come from
theoretical issues that could come from namespace collisions.
I'm not sure if anything with this constant has come up before.
But I have mentioned this issue to the developer whom is working
on fixing / cleaning up the headers to see what he says.

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

[Qemu-devel] QEMU build broken

[Brad Smith]

The following commit broke the build of QEMU..

linux-user: remove configure option for setting uname release

http://git.qemu.org/?p=qemu.git;a=commit;h=e586822a58b6609edb5ea929e8a4aa394d32389f

http://buildbot.b1-systems.de/qemu/builders/default_openbsd_current/builds/752/steps/compile/logs/stdio

/buildbot-qemu/default_openbsd_current/build/bsd-user/main.c:46:34: 
error: use of undeclared identifier 'CONFIG_UNAME_RELEASE'


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

Re: [Qemu-devel] QEMU build broken

[Peter Maydell]

On 8 May 2014 15:47, Brad Smith <brad@comstyle.com> wrote:
> The following commit broke the build of QEMU..
>
> linux-user: remove configure option for setting uname release
>
> http://git.qemu.org/?p=qemu.git;a=commit;h=e586822a58b6609edb5ea929e8a4aa394d32389f
>
> http://buildbot.b1-systems.de/qemu/builders/default_openbsd_current/builds/752/steps/compile/logs/stdio
>
> /buildbot-qemu/default_openbsd_current/build/bsd-user/main.c:46:34: error:
> use of undeclared identifier 'CONFIG_UNAME_RELEASE'

Ah, bsd-user. Do you actually use it, or is it just
in the default compile that you're running?

In this case bsd-user makes no use at all of the
qemu_uname_release variable so we should probably
just rip it out (together with the useless command
line argument that lets the user tweak it).

thanks
-- PMM

Re: [Qemu-devel] QEMU build broken

[Riku Voipio]

[-- Attachment #1: Type: text/plain, Size: 1061 bytes --]

Hi,

On 8 May 2014 17:54, Peter Maydell <peter.maydell@linaro.org> wrote:

> On 8 May 2014 15:47, Brad Smith <brad@comstyle.com> wrote:
> > The following commit broke the build of QEMU..
> >
> > linux-user: remove configure option for setting uname release
> >
> >
> http://git.qemu.org/?p=qemu.git;a=commit;h=e586822a58b6609edb5ea929e8a4aa394d32389f
> >
> >
> http://buildbot.b1-systems.de/qemu/builders/default_openbsd_current/builds/752/steps/compile/logs/stdio
> >
> > /buildbot-qemu/default_openbsd_current/build/bsd-user/main.c:46:34:
> error:
> > use of undeclared identifier 'CONFIG_UNAME_RELEASE'
>
> Ah, bsd-user. Do you actually use it, or is it just
> in the default compile that you're running?
>

 One year since last bsd-user specific patch, I take we need a new
maintainer for bsd-user?

In this case bsd-user makes no use at all of the
> qemu_uname_release variable so we should probably
> just rip it out (together with the useless command
> line argument that lets the user tweak it).
>

Agreed, this is just copypaste from linux-user.

Riku

[-- Attachment #2: Type: text/html, Size: 2007 bytes --]

Re: [Qemu-devel] QEMU build broken

[Peter Maydell]

On 9 May 2014 09:57, Riku Voipio <riku.voipio@linaro.org> wrote:
> On 8 May 2014 17:54, Peter Maydell <peter.maydell@linaro.org> wrote:
>> Ah, bsd-user. Do you actually use it, or is it just
>> in the default compile that you're running?

>  One year since last bsd-user specific patch, I take we need a new
> maintainer for bsd-user?

Perhaps so. Stacey Son submitted a set of patches to
it back in January, but they were a very large series
which needed some restructuring to get through code
review and I don't think there's been a respin of those.

Personally I would like to see it either (a) actively
maintained upstream or (b) just removed from the tree;
the current situation doesn't seem very useful.

thanks
-- PMM

Re: [Qemu-devel] QEMU build broken

[Stacey Son]

On May 9, 2014, at 4:55 AM, Peter Maydell <peter.maydell@linaro.org> wrote:

> On 9 May 2014 09:57, Riku Voipio <riku.voipio@linaro.org> wrote:
>> On 8 May 2014 17:54, Peter Maydell <peter.maydell@linaro.org> wrote:
>>> Ah, bsd-user. Do you actually use it, or is it just
>>> in the default compile that you're running?
> 
>> One year since last bsd-user specific patch, I take we need a new
>> maintainer for bsd-user?
> 
> Perhaps so. Stacey Son submitted a set of patches to
> it back in January, but they were a very large series
> which needed some restructuring to get through code
> review and I don't think there's been a respin of those.
> 
> Personally I would like to see it either (a) actively
> maintained upstream or (b) just removed from the tree;
> the current situation doesn't seem very useful.

There is some interest from FreeBSD folks to use qemu bsd-user mode in package cross building.  In fact, support was added recently to poudriere to used qemu to cross build packages for armv6 and mips64 (see https://fossil.etoilebsd.net/poudriere/tktview/eb7afa654cec95e107905c64b78fb030abc1437b for example).  Sean Bruno added the missing bits to make mips32 work correctly.  Juergen Lock (nox@) has also contributed a lot of bug fixes and maintains the FreeBSD port.

As for restructuring the (25,000+ lines of) patches Sean has agreed to help with that given that I am busy with some other things at the moment.

Regards,

-stacey. 

Re: [Qemu-devel] QEMU build broken

[Brad Smith]

On 08/05/14 10:54 AM, Peter Maydell wrote:
> On 8 May 2014 15:47, Brad Smith <brad@comstyle.com> wrote:
>> The following commit broke the build of QEMU..
>>
>> linux-user: remove configure option for setting uname release
>>
>> http://git.qemu.org/?p=qemu.git;a=commit;h=e586822a58b6609edb5ea929e8a4aa394d32389f
>>
>> http://buildbot.b1-systems.de/qemu/builders/default_openbsd_current/builds/752/steps/compile/logs/stdio
>>
>> /buildbot-qemu/default_openbsd_current/build/bsd-user/main.c:46:34: error:
>> use of undeclared identifier 'CONFIG_UNAME_RELEASE'
>
> Ah, bsd-user. Do you actually use it, or is it just
> in the default compile that you're running?

I do not use it personally but it is common sense that commits
must not be breaking the build.

> In this case bsd-user makes no use at all of the
> qemu_uname_release variable so we should probably
> just rip it out (together with the useless command
> line argument that lets the user tweak it).
>
> thanks
> -- PMM


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

Re: [Qemu-devel] QEMU build broken

[Peter Maydell]

On 10 May 2014 00:02, Brad Smith <brad@comstyle.com> wrote:
> On 08/05/14 10:54 AM, Peter Maydell wrote:
>> Ah, bsd-user. Do you actually use it, or is it just
>> in the default compile that you're running?

> I do not use it personally but it is common sense that commits
> must not be breaking the build.

I generally agree, but the minor ports (roughly, anything
not x86 Linux) are inevitably going to get broken from time
to time, because not everybody has access to all those
systems to test on. bsd-user is particularly bad because
it is a large chunk of code only built for BSD and it's
not really maintained right now. (Hence my interest in
whether it actually has users or if we're just carrying
around a big lump of dead weight code.)

If we had a system that automatically tested all pull
requests on a full set of build systems and mailed the
list with the results, that would be nice, but we don't...

thanks
-- PMM

Re: [Qemu-devel] QEMU build broken

[Brad Smith]

On 09/05/14 7:35 PM, Peter Maydell wrote:
> On 10 May 2014 00:02, Brad Smith <brad@comstyle.com> wrote:
>> On 08/05/14 10:54 AM, Peter Maydell wrote:
>>> Ah, bsd-user. Do you actually use it, or is it just
>>> in the default compile that you're running?
>
>> I do not use it personally but it is common sense that commits
>> must not be breaking the build.
>
> I generally agree, but the minor ports (roughly, anything
> not x86 Linux) are inevitably going to get broken from time
> to time, because not everybody has access to all those
> systems to test on. bsd-user is particularly bad because
> it is a large chunk of code only built for BSD and it's
> not really maintained right now. (Hence my interest in
> whether it actually has users or if we're just carrying
> around a big lump of dead weight code.)

This is just excuses and points out poor project process.
There could easily be a staging branch to deal with this.

> If we had a system that automatically tested all pull
> requests on a full set of build systems and mailed the
> list with the results, that would be nice, but we don't...
>
> thanks
> -- PMM
>


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

Re: [Qemu-devel] QEMU build broken

[Peter Maydell]

On 10 May 2014 00:49, Brad Smith <brad@comstyle.com> wrote:
> This is just excuses and points out poor project process.
> There could easily be a staging branch to deal with this.

I have no objection if you'd like to sort out our build
and test infrastructure.

thanks
-- PMM

Re: [Qemu-devel] QEMU build broken

[Markus Armbruster]

Brad Smith <brad@comstyle.com> writes:

> On 09/05/14 7:35 PM, Peter Maydell wrote:
>> On 10 May 2014 00:02, Brad Smith <brad@comstyle.com> wrote:
>>> On 08/05/14 10:54 AM, Peter Maydell wrote:
>>>> Ah, bsd-user. Do you actually use it, or is it just
>>>> in the default compile that you're running?
>>
>>> I do not use it personally but it is common sense that commits
>>> must not be breaking the build.
>>
>> I generally agree, but the minor ports (roughly, anything
>> not x86 Linux) are inevitably going to get broken from time
>> to time, because not everybody has access to all those
>> systems to test on. bsd-user is particularly bad because
>> it is a large chunk of code only built for BSD and it's
>> not really maintained right now. (Hence my interest in
>> whether it actually has users or if we're just carrying
>> around a big lump of dead weight code.)
>
> This is just excuses and points out poor project process.
> There could easily be a staging branch to deal with this.

Having your feature in-tree is a privilege, not a right.  You earn it by
helping to maintain it.  "it's not really maintained right now" means it
has not been earning its keep.  You're encouraged to remedy that.

Re: [Qemu-devel] QEMU build broken

[Brad Smith]

On 10/05/14 2:25 AM, Markus Armbruster wrote:
> Brad Smith <brad@comstyle.com> writes:
>
>> On 09/05/14 7:35 PM, Peter Maydell wrote:
>>> On 10 May 2014 00:02, Brad Smith <brad@comstyle.com> wrote:
>>>> On 08/05/14 10:54 AM, Peter Maydell wrote:
>>>>> Ah, bsd-user. Do you actually use it, or is it just
>>>>> in the default compile that you're running?
>>>
>>>> I do not use it personally but it is common sense that commits
>>>> must not be breaking the build.
>>>
>>> I generally agree, but the minor ports (roughly, anything
>>> not x86 Linux) are inevitably going to get broken from time
>>> to time, because not everybody has access to all those
>>> systems to test on. bsd-user is particularly bad because
>>> it is a large chunk of code only built for BSD and it's
>>> not really maintained right now. (Hence my interest in
>>> whether it actually has users or if we're just carrying
>>> around a big lump of dead weight code.)
>>
>> This is just excuses and points out poor project process.
>> There could easily be a staging branch to deal with this.
>
> Having your feature in-tree is a privilege, not a right.  You earn it by
> helping to maintain it.  "it's not really maintained right now" means it
> has not been earning its keep.  You're encouraged to remedy that.

Huh? "my feature"? I have nothing to do with this. What kind of crazy
is this? How to misdirect and not take responsibility for breaking
something. If there wasn't sloppy irresponsible development in the
first place it wouldn't be an issue.

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

Re: [Qemu-devel] QEMU build broken

[Paolo Bonzini]

Il 10/05/2014 08:45, Brad Smith ha scritto:
>>
>> Having your feature in-tree is a privilege, not a right.  You earn it by
>> helping to maintain it.  "it's not really maintained right now" means it
>> has not been earning its keep.  You're encouraged to remedy that.
>
> Huh? "my feature"? I have nothing to do with this. What kind of crazy
> is this? How to misdirect and not take responsibility for breaking
> something. If there wasn't sloppy irresponsible development in the
> first place it wouldn't be an issue.

Brad,

all this is doing, is convincing people that bsd-user is not worth 
keeping in the tree.  It's a fact that in a million-line codebase not 
all patches can be tested by all people.

Why don't you send a patch instead of whining?

Paolo

Re: [Qemu-devel] QEMU build broken

[Peter Maydell]

On 10 May 2014 08:07, Paolo Bonzini <pbonzini@redhat.com> wrote:
> Why don't you send a patch?

Or just test the one I sent yesterday:
http://patchwork.ozlabs.org/patch/347443/

thanks
-- PMM

Re: [Qemu-devel] QEMU build broken

[Brad Smith]

On 10/05/14 3:07 AM, Paolo Bonzini wrote:
> Il 10/05/2014 08:45, Brad Smith ha scritto:
>>>
>>> Having your feature in-tree is a privilege, not a right.  You earn it by
>>> helping to maintain it.  "it's not really maintained right now" means it
>>> has not been earning its keep.  You're encouraged to remedy that.
>>
>> Huh? "my feature"? I have nothing to do with this. What kind of crazy
>> is this? How to misdirect and not take responsibility for breaking
>> something. If there wasn't sloppy irresponsible development in the
>> first place it wouldn't be an issue.
>
> Brad,
>
> all this is doing, is convincing people that bsd-user is not worth
> keeping in the tree.  It's a fact that in a million-line codebase not
> all patches can be tested by all people.

My posts have nothing to do with bsd-user. I don't give a shit about it.
The real issue is the process and the fact that someone removed a 
constant from the configure script and didn't even grep the tree to see
if it existed anywhere else. That is very sloppy.

> Why don't you send a patch instead of whining?

Constantly trying to deflect from the real issues.


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.

Re: [Qemu-devel] QEMU build broken

[Markus Armbruster]

Brad Smith <brad@comstyle.com> writes:

> On 10/05/14 3:07 AM, Paolo Bonzini wrote:
>> Il 10/05/2014 08:45, Brad Smith ha scritto:
>>>>
>>>> Having your feature in-tree is a privilege, not a right.  You earn it by
>>>> helping to maintain it.  "it's not really maintained right now" means it
>>>> has not been earning its keep.  You're encouraged to remedy that.
>>>
>>> Huh? "my feature"? I have nothing to do with this. What kind of crazy
>>> is this? How to misdirect and not take responsibility for breaking
>>> something. If there wasn't sloppy irresponsible development in the
>>> first place it wouldn't be an issue.
>>
>> Brad,
>>
>> all this is doing, is convincing people that bsd-user is not worth
>> keeping in the tree.  It's a fact that in a million-line codebase not
>> all patches can be tested by all people.
>
> My posts have nothing to do with bsd-user. I don't give a shit about it.
> The real issue is the process and the fact that someone removed a
> constant from the configure script and didn't even grep the tree to
> see
> if it existed anywhere else. That is very sloppy.
>
>> Why don't you send a patch instead of whining?
>
> Constantly trying to deflect from the real issues.

No, this *is* the real issue: process gets improved by *people*.  People
like *you*.  Pointing out flaws in the process is only marginally
helpful, as we're painfully aware of them already.  Actually doing
something about it is what counts, because that's where we're short.

As Peter said, code people can't easily build build locally *will* get
broken once in a while, and the solution for that are build bots
shooting down pull requests that don't compile, or fail automated tests.
You're very much invited to help with providing such services.