[Development] 4.8 Build failing on Mac OS 10.7 / XCode 4.2

[eike.ziller at nokia.com]

On 31 Oct 2011, at 23:24, ext Alexis Menard wrote:

> On Mon, Oct 31, 2011 at 12:08 PM,  <eike.ziller at nokia.com> wrote:
>> 
>> On 30 Oct 2011, at 11:08, ext Alexis Menard wrote:
>> 
>>> 2011/10/30 Thiago Macieira <thiago at kde.org>:
>>>> On Saturday, 29 de October de 2011 09:22:07 Chris Meyer wrote:
>>>>> I would like to call attention to the following problem which is that
>>>>> the Qt 4.8 build fails using the latest Mac OS 10.7 / Xcode 4.2 tools.
>>> 
>>> For *your* use case. Normal and default cases work as the CI and other
>>> reports say.
>>> 
>>>>> 
>>>>> https://bugreports.qt.nokia.com/browse/QTBUG-20619
>>>>> 
>>>>> Somehow the RC 1 candidates are being built... so there must be some
>>>>> workaround. Is it possible to post the configure commands,
>>>>> environment, and tools used to build the official 4.8 RC's?
>>>>> 
>>>>> Also, I would hope that unilaterally closing a bug that affects a
>>>>> primary platform (the Mac is still a primary platform, right?) with a
>>>>> 'Won't Fix'  and closing down all discussion on the issue is a trend
>>>>> that won't continue.
>>> 
>>> It's a primary platform but an exotic build setup.
>>> 
>>>> 
>>>> You're over-reacting. That just means he (the assignee) will not fix the bug.
>>>> If he had the power to close down all discussion, your email wouldn't exist
>>>> and I wouldn't be typing this reply right now.
>>>> 
>>>> Also do note that comments after the bug being closed are still received just
>>>> as they were before. You can contribute more information to the report so that
>>>> a solution can be found.
>>>> 
>>>>> Zeno Albisser (the one who closed the bug) made some suggestions, but
>>>>> since the bug is closed, there is no place for developers to post
>>>>> information about what works and what doesn't work. My first thought
>>>>> is to turn off debug libs; but I'm guessing other developers have done
>>>>> it already so it would be nice if the bug were to remain open so that
>>>>> we could all benefit from trial and error until a solution is found.
>>>> 
>>>> Sure there is: the bug report. Comments are not disabled in closed bug
>>>> reports. It just means Zeno is no longer looking for a solution. But if you
>>>> provide one that works for everyone...
>>> 
>>> Which I believe it is hard to find. I would have close the bug myself.
>>> The command line option that the test case passes -arch x86 -arch
>>> x86_64 is part of the so called "combination of command lines we can't
>>> support". It perhaps worked before but with WebCore growing everyday
>>> that was pure luck.
>> 
>> Calling the abitlity to build 32+64bit fat binaries "pure luck" is a bit funny (and feels a bit offensive to me), since we always provided universal 32bit+64bit binary packages for Mac on qt.nokia.com/downloads, up to 4.7.4.
> 
> Was not meant to be be offensive. You have to keep in mind that what
> ships in 4.7 is at least a good year old WebKit and it is much bigger
> now.
> 
>> 
>> Also, universal binaries are quite common on Mac. Btw, even Safari comes in the form of a universal 32+64bit binary (at least on snow leopard) ;)
>> 
>> Anyhow, it looks like the problem is not the combination of "-arch x86 -arch x86_64" but the combination of "-debug_and_release -arch x86 -arch x86_64". Which I think would be acceptable. (I just checked that "-release -arch x86 -arch x86_64" works for me.)
> 
> Debug build of WebKit are always problematic as it requires a lot of
> RAM to link for example (read at least 8GB), on Linux it may kill the
> linker sometimes because it runs out of memory.


> Debug build of WebKit
> are to use only if really needed.

That reminds me: Wasn't there a time where debug was off by default for QtWebKit even when doing a debug build of Qt? Where's that gone?
I suppose the main problem is that doing the perceived trivial thing "./configure -arch x86 -arch x86_64 && make" fails to build.
Afair Qt didn't build webkit in debug by default, and you could enable debug for webkit with a separate configure switch.
Default compile/link time and disk space/RAM consumption would be much reduced too, in that case.

Btw, here's a visualization of *how* big webkit actually got within Qt (debug): http://imagebin.org/181891

++ Eike

>>> Splitting the build of WebCore into two pieces so
>>> that ranlib could work and your test case work is just stupid because
>>> it will add complexity for us to support/implement this, and we don't
>>> want.  In the other hand if it's that important for you, you can try
>>> to have two builds and make a mixture of the library in one of the
>>> framework if that's possible.
>> 
>> 
>> 
>> --
>> Eike Ziller
>> Principal Software Engineer
>> 
>> Nokia, Qt Development Frameworks
>> 
>> Nokia gate5 GmbH
>> Firmensitz: Invalidenstr. 117, 10115 Berlin, Germany
>> Registergericht: Amtsgericht Charlottenburg, Berlin: HRB 106443 B
>> Umsatzsteueridentifikationsnummer: DE 812 845 193
>> Geschäftsführer: Dr. Michael Halbherr, Karim Tähtivuori
>> 
>> 
> 
> 
> 
> -- 
> Alexis Menard (darktears)
> Software Engineer
> INdT Recife Brazil

-- 
Eike Ziller
Principal Software Engineer

Nokia, Qt Development Frameworks

Nokia gate5 GmbH
Firmensitz: Invalidenstr. 117, 10115 Berlin, Germany
Registergericht: Amtsgericht Charlottenburg, Berlin: HRB 106443 B
Umsatzsteueridentifikationsnummer: DE 812 845 193
Geschäftsführer: Dr. Michael Halbherr, Karim Tähtivuori