[R-SIG-Mac] Building packages under R >= 2.7.0 (issue with gcc-4.2)

[Simon Urbanek]

On Jun 2, 2008, at 10:10 AM, Vincent Goulet wrote:

> Simon,
>
> Thanks for your reply. Please see inline.
>
> Le lun. 02 juin à 09:16, Simon Urbanek a écrit :
>
>> Vincent,
>>
>> I'm sorry but somehow I have no idea what you're talking about.  
>> Your package "actuar" (in the most recent version 0.9-7 on CRAN)  
>> passes "make check" with R 2.7.0 and it is available in binary form  
>> from CRAN as well, see:
>>
>> http://cran.r-project.org/web/packages/actuar/index.html
>>
>> more comments inline:
>>
>> On Jun 1, 2008, at 1:50 AM, Vincent Goulet wrote:
>>
>>> My package checked and buit fine under 2.6.2 on Leopard 10.5.2.  
>>> Under 2.7.0, 2.7.0 patched (2008-05-29) and R-devel (2008-05-30)  
>>> it always fails at the "checking whether package 'actuar' can be  
>>> installed"  step. The 00install.log tells me that R tried to call  
>>> gcc-4.2, which I don't have. It is my understanding from reading  
>>> the documentation that this version of gcc is not mandatory. I  
>>> have XCode v. 3.0 installed.
>>>
>>
>> The compiler used in the released versions is gcc (which is by  
>> default gcc-4.0 even if you installed gcc-4.2). You can map gcc to  
>> gcc-4.2 if you desire so, but that is a non-standard setup.
>
> I started again from scratch (deleting everything in Library/ 
> Frameworks/R.framework and reinstalling 2.7.0) and indeed,  
> everything is fine. It appears I had the troublesome Resources/etc/ 
> x86_64/Makeconf file in my installation. How it got there I have no  
> idea. Perhaps a momentary lapse of reason in the installation  
> process at some point.
>

Ah, that makes sense. If you installed the Leopard experimental build  
at any time, it adds the 64-bit architectures to the R tree which are  
retained even if you re-install 32-bit R over it. The main reason is  
that the Leopard build is not supplied as Installer package, so the  
installation system has no idea that the 64-bit files were installed.  
The Leopard build has to use gcc-4.2 due to a few bugs in gcc-4.0  
which are triggered by the 64-bit builds.


>>> Furthermore, I noticed that, for 2.7.0, the /Library/Frameworks/ 
>>> R.framework/Resources/etc/*/Makeconf files refer to gcc for i386  
>>> and ppc, and to gcc-4.2 on x86_64. For 2.7.0 patched and R-devel,  
>>> the Makeconf files refer to gcc-4.2 on all architectures.
>>>
>>
>> Not in the official binaries that I am releasing:
>>
>> /Library/Frameworks/R.framework/Resources/etc/ppc/Makeconf:
>> CC = gcc -arch ppc -isysroot /Developer/SDKs/MacOSX10.4u.sdk - 
>> mmacosx-version-min=10.4 -std=gnu99
>>
>> /Library/Frameworks/R.framework/Resources/etc/i386/Makeconf:
>> CC = gcc -arch i386 -isysroot /Developer/SDKs/MacOSX10.4u.sdk - 
>> mmacosx-version-min=10.4 -std=gnu99
>
> However, as I originally stated, all Makeconf files in R-2.7-branch- 
> leopard-universal.tar.gz and R-devel-leopard-universal.tar.gz on  
> R.research.att.com refer to gcc-4.2. But then these are not official  
> releases, I guess.
>

Yes, those are experimental and Leopard users can easily get gcc-4.2  
so it's ok to require it (and Xcode 3.1 contains it anyway which mean  
that they'll likely be included by default soon).

Cheers,
Simon


>>> I can't see what I did wrong, but then I'm certainly not the first  
>>> one to compile a package from source since the release of 2.7.0.  
>>> Any help would be appreciated.
>>>
>>
>> You'll have to tell us exactly which R build are you using. CRAN  
>> binaries are clearly not configured the way you are describing, so  
>> you must be using some custom or experimental build in which case  
>> you're on you own ...
>>
>> Nonetheless, even in the case of using gcc-4.2 your package  
>> installs, so I have no idea what you are concerned about.
>
> Again, this was most likely caused by a partially oozed system.  
> Sorry for the inconvenience and thank you again for your help and  
> great work.
>
> Best,
>
> Vincent
>
>>
>>
>> Cheers,
>> Simon
>>
>
> _______________________________________________
> R-SIG-Mac mailing list
> R-SIG-Mac at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-mac
>
>