[R-SIG-Mac] tools for compiling R from source on Leopard

[Simon Urbanek]

On Mar 2, 2008, at 11:40 AM, Michael Braun wrote:

> Simon:
>
> Thanks for clarifying all of this.  The statement in section 2.1.2  
> of the MacOS FAQ is what threw me off a bit (I thought when matching  
> 3.x and 4.x, the x had to match as well.  My mistake).  Certainly, I  
> don't need to be using experimental, unsupported tools.  Again, I  
> probably just read way too much into the instructions.
>
> But I do want to be sure that I have a 64-bit build of R  that  is  
> linked to the vecLib BLAS (similar to my current configuration on  
> Linux).  Is there a way to do this without compiling from source?  I  
> see there are instructions on the R for MacOS developer page, but  
> these are also flagged as "experimental."  If I don't want to be too  
> experimental, I'm not sure how to get to where I want to go otherwise.
>

Normally you can get 64-bit build from
http://r.research.att.com/
but right now the x86_64 build doesn't pass make check, so it's  
officially not available for download. The problem is a bug in the  
gcc-4.2 compiler that miscompiles some code.

Anyway, if you feel like it, you can try either
http://r.research.att.com/R-devel-leopard-universal.tar.gz
or
http://r.research.att.com/R-2.6-branch-leopard-universal.tar.gz

The latter is a 2.6-branch so it's a bit less experimental.
To install, just unpack to root,e.g.
sudo tar fxz R-devel-leopard-universal.tar.gz -C /

Let me know if that works for you.

As of vecLib - that's really easy - you can always symlink the vecLib  
framework as libRblas.dylib, but all CRAN builds use vecLib anyway.

Cheers,
Simon


> On Mar 2, 2008, at 11:04 AM, Simon Urbanek wrote:
>
>> Michael,
>>
>> the instructions you are referring to (gcc-4.2-based) are for  
>> experimental R builds using cutting-edge unreleased tools, so  
>> you're pretty much on your own there. So let me first address  
>> "offical" way as documented in the FAQ:
>> The gfortran compiler the FAQ refers to is on CRAN (http://cran.r-project.org/bin/macosx/tools 
>> ) - it is also the first download on the R for Mac Tools page. That  
>> is all you need (Xcode + the above Fortran) and both the FAQ and  
>> the tools page specifically say so.
>>
>> Now to your quest for experimental, unsupported tools (see also  
>> comments inline below). The only reason to use gcc-4.2 is if you  
>> want to play with OpenMP. Note that the gcc-4.2 is known to have  
>> bugs, so tread with care.
>>
>>
>>> Hi.  This week I received my very first Mac (I was previously a  
>>> Linux  user), and I am now trying to compile R from source.  I am  
>>> running Mac  OS 10.5.2, and I have downloaded and installed Xcode  
>>> 3.0.  Following  instructions on the R for Mac OS FAQ page I see  
>>> that I need to install  gfortran 4.2.1.  However, Xcode only comes  
>>> with gcc 4.0.1, and I  understand that the versions need to be the  
>>> same.
>>>
>>
>> That is not true (and I'm not sure why concluded it). Only the  
>> major version (3.x vs 4.x) has to match (as the FAQ tells you), so  
>> gcc 4.0.1 + gfortran 4.2.1 are just fine.
>>
>>
>>> The R for Mac tools site mentions that the Apple version for gcc  
>>> 4.2.1 is available from ADC, but I just cannot find it anywhere
>>
>> ADC: http://connect.apple.com/
>> Login, click on Downloads -> Developer Tools -> GCC 4.2 Developer  
>> Preview 1
>> If it's not there, please let me know.
>>
>>> (I Googled, searched, etc, with no luck--can someone please point  
>>> me in the right direction?).  So, I ran  the provided  
>>> gcc42build5531.pkg installation file.  The installer ran with no  
>>> errors, but it doesn't seem to install anywhere
>>
>> Try typing
>> gcc-4.2
>>
>>
>>> (I even created a test folder to install to, but after the  
>>> installation, that folder was empty).
>>
>> Which is expected - try enabling the installer log (Window- 
>> >Installer Log) if you're interested in the details. If you're more  
>> comfortable with command-line tools and want to learn more about  
>> packages on OS X, have a look at "man pkgutil" and "man installer".  
>> For old-style packages (which are bundles) see also "man pax" and  
>> "man lsbom".
>>
>>
>>> gcc --version still indicates that the version is 4.0.1.
>>
>> Yes, gcc is a link to gcc-4.0, not to gcc-4.2.
>>
>>
>>> The gfortran-42.pkg file installed correctly.
>>>
>>> I know this sounds more like a Mac OS problem than an R problem,
>>
>> There is no problem, really ;).
>>
>>
>>> but since I am using tools (and following instructions)
>>> designed specifically for R, I thought I'd ask for help here.   I  
>>> suppose that all I really need is to be able to effectively  
>>> install either gcc 4.2.1 (since I have the R-approved gfortran  
>>> 4.2.1 installed) or an R-approved gfortran 4.0.1 (since I have gcc  
>>> 4.0.1installed).
>>>
>>
>> No, that is wrong (see above). If you followed the instructions  
>> you'd use gcc 4.0.1 from Apple and gfortran 4.2.1 from CRAN. This  
>> is how the release is built.
>>
>> FWIW: Although it is a nice exercise, I'm not quite sure why you  
>> want to build R from sources unless you want to modify it - note  
>> that your result is likely to be incompatible with the packages we  
>> create (unless you build the full multi-arch universal build).
>>
>> One more side-note on the tools: now that you have installed  
>> several tools in your system (gcc 4.0, 4.2, two different gfortrans  
>> etc.), you should be aware that although you can use any  
>> combination of them, the library lookup is common to all, so unless  
>> you de-install the gfortran from /usr/local/bin (the CRAN verison)  
>> [via /usr/local/bin/gfortran-uninstall], even the binaries compiled  
>> with the Leopard gfotran (/usr/bin/gfortran-4.2] will be linked to  
>> the /usr/local/lib/libgfortran dynamic library, because it has  
>> higher precedence than the static library that the system gfortran  
>> supplies. This is not really a problem as long as you are aware of  
>> it, but it that reduces the portability of your binaries a bit  
>> (you'll have to ship the gfortran shared library with your binaries  
>> if they use Fortran). If you don't understand what I'm saying,  
>> don't worry too much :) it doesn't affect anything as long as you  
>> stay on the same machine.
>>
>> Cheers,
>> Simon
>>
>
>