[Rd] Pb with contrib.url() (PR#9131)

[Simon Urbanek]

On Aug 8, 2006, at 2:52 AM, Prof Brian Ripley wrote:

> On Mon, 7 Aug 2006, Simon Urbanek wrote:
>
>> Herve,
>>
>> On Aug 7, 2006, at 11:34 AM, hpages at fhcrc.org wrote:
>>
>>> Recently I needed to download a few R packages for Unix, Windows
>>> and Mac OS X. The idea was to put them all together on a USB key in
>>> order to be able to install them on systems without network
>>> connection.
>>>
>>> It worked fine for the "src" and "win.binary" packages but I had
>>> the following problems with the "mac.binary" packages:
>>>
>>>> download.packages("XML", destdir=".", type="mac.binary")
>>>     Warning: unable to access index for repository
>>>     http://cran.fhcrc.org/bin/macosx/x86_64/contrib/2.3
>>
>> ^^ we don't build binaries for x86_64 architecture, so you cannot use
>> that (Apple didn't release Leopard yet, so there is no 64-bit Intel
>> binaries around anyway). The only architectures supported by our
>> binaries are powerpc, i686 and universal.
>
> Note that ?download.packages says (from Linux)
>
>     type: character, indicating the type of package to download and
>           install. Possible values are '"source"' (the default except
>           under the CRAN Mac OS X build), '"mac.binary"' and
>           '"win.binary"' (which can be downloaded but not installed).
>
> so we do appear to say we can download MacOS binaries on non-MacOS
> platforms.  There in lies the problem as contrib.url 
> (type="mac.binary")
> uses the architecture of the current platform.  I think you need types
> "mac.binary.powerpc" etc to now allow this.
>

Yes, I see, thanks. Given that we supply universal binaries (=should  
work on all supported archs), we could define mac.binary as macosx/ 
universal and mac.binary.xxx as macosx/xxx. The default would still  
be mac.binary.
I'm not sure if we really need separate architecture binaries. The  
only benefit I see is that other repositories which don't have the  
resources to build binaries for all platforms could provide separate  
binaries. On CRAN the other directories are just symlinks to  
universal anyway. Any comments on that? Does anyone use the separate  
binaries feature?

Cheers,
Simon