[Rd] Re: [R] Creating a minimal package

[Duncan Murdoch]

On Tue, 13 Jul 2004 11:20:36 -0500, Dirk Eddelbuettel <edd at debian.org>
wrote :

>On Tue, Jul 13, 2004 at 10:02:16AM -0400, Duncan Murdoch wrote:
>> Yes, I think that's right.  Currently the roadblocks are:
>> 
>>  - Perl 
>> 
>> I think we could write around this.
>
>Is that worth it?  ActiveState Perl is easy to install. 

I think it's worth it if someone else does the work, but so far I
haven't thought it's worthwhile enough to do the work myself.

The reason is that it would simplify support and maintenance if the R
installer could install a complete R system.  We can't do that, but
the fewer "and then you need to..." instructions we have, the better.

>>  - Microsoft Help Compiler
>> 
>> Users who don't have it could do without CHM help files.
>
>I built local packages for years, and only added this tool fairly recently
>(mostly to suppress the error/warning during build).  As I don't use html
>help, I obviously didn't care much.  
>
>Non-issue in my biased book, but I acknowledge that this is probably a
>minority view. That said, this is likely to be non-redistributable, so a
>case could be made that as a user needs to download this anyway, how much
>harder is it to download the free-as-in-beer Perl?

The difference is that package installation works without it, it just
does a more complete job if you have it.  Currently installation
doesn't work at all without Perl.
 
>>  - Lots of Cygwin (and other) tools involved in making packages.
>> 
>> We could possibly rewrite the build process in R, but I don't think we
>> should.  I think a better solution would be to distribute the toolset
>> with R.  The full toolset doesn't add up to much (maybe 1.5 M).  A
>
>Neat idea. I like that a lot. I can see how some users would have Perl and
>Cygwin, and even MinGW, anyway, but not the BDR tools collection. This helps.
>
>> problem is that multiple Cygwin versions don't coexist nicely, so we'd
>> have to be careful during installation.
>> 
>>  - MinGW gcc
>> 
>> Users who want to compile code would still have to install the
>> compiler.
>
>And that's where it gets hairy. So we do all this to let lusers build R-only
>packages, but not compiled packages?  Isn't that both a little inconsistent
>and confusing?

Lots and lots of users have compilers other than MinGW.  This would
allow them to install their own packages built with those compilers.

>But we could rewrite MinGW gcc in R too.

I just tried this:

> gcc <- function(file) { # compile it 
+ }
> package.skeleton('Rgcc', 'gcc')
Creating directories ...
Creating DESCRIPTION ...
Creating READMEs ...
Saving functions and data ...
Making help files ...
Done.
Further steps are described in ./Rgcc/README 

When I have time I'll work through the README and finish it off.

Duncan Murdoch