[R-gui] GUI for require(package)!

[John Fox]

Dear James,

I'll be interested to see what you come up with. I think that my solution is
probably sufficient for the Rcmdr package (actually, I can eliminate the
Bioconductor radio button since at present I only use CRAN packages) and
should be pretty easily adapted. As usual, I'll let things percolate for a
few days to see if I can think of improvements.

Thanks for raising the issue.
 John 

> -----Original Message-----
> From: r-sig-gui-bounces at stat.math.ethz.ch 
> [mailto:r-sig-gui-bounces at stat.math.ethz.ch] On Behalf Of 
> James Wettenhall
> Sent: Thursday, July 15, 2004 7:24 PM
> To: John Fox
> Cc: r-sig-gui at stat.math.ethz.ch
> Subject: RE: [R-gui] GUI for require(package)!
> 
> Hi John,
> 
> Great work!
> 
> Yes I agree that it's not too difficult to provide a PACKAGES 
> file on a CD.  On the other hand, I do still like the idea of 
> having a GUI option to install a missing R package from an 
> aribtrary place on your hard disk, e.g. your desktop (without 
> necessarily having a PACKAGES file).  I'll look at your code 
> and think about this.  (Maybe this is coming from a Windows 
> point of view - unlike Mac, it's easy to have multiple R 
> installations, e.g. C:\rw1090\ and C:\rw1091\ , so you could 
> download an R package and install it within your R 1.9.0 
> installation but not your R 1.9.1 installation, and then 
> later decide that you want it in R 1.9.1)
> 
> In my R packages, I'm not sure if .onLoad() is the right 
> place to install _all_ missing R packages, because some of 
> the R packages are really only 'recommended' rather than 
> 'required', i.e. a complete microarray analysis can be 
> performed using my GUI(s), without the 'recommended'
> packages, as long as you don't choose the 'advanced' or 
> 'extra-memory-hungry' normalization options.
> 
> So I would still like users to be able to run my GUIs without 
> some of the 'recommended' packages, but there are certainly 
> some packages which are critically required, so maybe they 
> could be installed in .onLoad() or .First.lib() if missing.
> For the 'recommended' packages, .onLoad() should not fail if 
> they are unavailable, but it could give a warning.
> 
> Another thing which could be incorporated into this is the 
> new reposTools / install.packages2() / getBioC() methods of 
> installation for the case of Bioconductor packages.  This may 
> not be of general interest to people on this list, but 
> obviously I am interested in Bioconductor.  The advantage of 
> the Bioconductor package-installation methods, is that they 
> try to take care of dependencies automatically.  I would 
> assume that the "Install package(s) from Bioconductor" menu 
> item in Rgui for Windows, doesn't yet use this, i.e. it 
> probably uses the original install.packages() and a PACKAGES file.
> 
> For an example of dependencies, if my limmaGUI package 
> depends on limma, but I forget that limma depends on statmod, 
> then if limma is missing, install.packages2() in 
> Bioconductor's reposTools packages should automatically see 
> that statmod is required for limma, and find it on CRAN.
> 
> But the reposTools / install.packages2() method can be slow 
> sometimes, because it searches multiple repositories for an R 
> package.  This may be fixed now, but in the past, I found 
> that (using the default options), it often searched the wrong 
> repository first, e.g. CRAN instead of Bioconductor.
> 
> Jeff Gentry is the expert on reposTools, install.packages2(), 
> and getBioC().  He also has a widgetInvoke package on 
> Bioconductor which could be of interest to Gtk fans on this 
> mailing list.
> 
> Regards,
> James