[Rd] Wishlist: install.packages to look for the newest version (PR#13852)

Uwe Ligges ligges at statistik.tu-dortmund.de
Sat Jul 25 11:48:27 CEST 2009



Thomas Lumley wrote:
> 
> Uwe,
> 
> I think Ulrike is making a different suggestion, that install.packages() 
> should fetch the binary that has been built for the current version of R.
> 
> This would be a bad idea for a different reason -- in general it is not 
> possible to be sure that the package works with an older version of R. 
> The R version dependency isn't enough for two reasons. The first is that 
> the author may well not know that the package fails with an older 
> version of R and so would not have listed a dependency. The second is 
> that the binary versions may be incompatible even if the source versions 
> are compatible.
> 
> You can always download a binary package from CRAN in a browser and use 
> the option to install from a local zip file. Or, as Uwe suggests, get a 
> new version of R.
> 
> What I think might be useful if it's not too difficult is a warning from 
> install.packages() that a newer version of the package you were 
> installing is available for the current version of R.

That could be done, of course, but not that easy (see below * ). But 
then, should we also warn in general (say if 
[install,update,available].packages() is called) that a newer R version 
is available?

I thought that has been discussed some years ago and the decision was 
not to warn, but now it is 2009 and we may want something different.

(*) Not easy for former releases. For future releases, it is possible to 
always look additionally into the "R-release" respository, but it means 
that an additional packages file must be downloaded (which is not funny 
for modem connections) in order to be able to compare versions and just 
to generate a warning.

Best wishes,
Uwe



>      -thomas
> 
> 
> On Fri, 24 Jul 2009 ligges at statistik.tu-dortmund.de wrote:
> 
>> Ulrike,
>>
>> if you install from source, you always get the most recent version of
>> the package given it does not depend on a newer version of R.
>>
>> If you want a binary package, you also get the newest version - that was
>> newest at the time we stopped building binaries for that version of R.
>> We (or better I if we only talk about Windows, but similar for all other
>> platforms) cannot build for each R version any more. In that case we'd
>> have to build even 11 binary versions for Windows just for the R-2.x.y
>> series now. Binary repositories are fixed at some time (for Windows once
>> the first patchlevel release of the next R version is out, e.g. at the
>> time of the R-2.9.1 release the binary builds for R-2.8.x had been 
>> stopped).
>>
>> So please upgrade your version of R or compile yourself from sources for
>> the R version you need the particular package for.
>>
>> Best wishes,
>> Uwe Ligges
>>
>>
>>
>>
>>
>> groemping at bht-berlin.de wrote:
>>> Full_Name: Ulrike Groemping
>>> Version: 2.9.0 (and older)
>>> OS: Windows
>>> Submission from: (NULL) (84.190.173.190)
>>>
>>>
>>> When using an older version of R, packages are not found although 
>>> they are
>>> available for newer versions of R and do work when installed with the 
>>> old
>>> version. For example, installing DoE.base on R 2.8.1 installs version 
>>> 0.2, while
>>> CRAN is at version 0.4-1 currently. It would be nice if the install 
>>> process
>>> would per default look for the newest version of the package and 
>>> install this
>>> one if its R-version request allows this. (I recently found a help 
>>> list entry by
>>> Uwe Ligges that explains how to manually install from a repository 
>>> for a newer
>>> CRAN version, but I did not bookmark it and cannot find it any more. The
>>> documentation does not enlighten me there.)
>>>
>>> Regards, Ulrike
>>>
>>> ______________________________________________
>>> R-devel at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-devel
>>
>> ______________________________________________
>> R-devel at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel
>>
> 
> Thomas Lumley            Assoc. Professor, Biostatistics
> tlumley at u.washington.edu    University of Washington, Seattle
> 
> 
>



More information about the R-devel mailing list