building packages using S4 methods - R-help

Wed, Jul 30, 2003 5:15 AM #

I can only find the option "--save" under install.
I'm using "Rcmd build --binary pkgName" on a Windows NT platform is there
anyway I can pass a similar directive?

Is there any more detailed documentation for Rcmd beyond the "Writing R
Extensions" manual or should I look at the R source code?

Thanks,

John Marsland

-----Original Message-----
From: Prof Brian Ripley [mailto:ripley at stats.ox.ac.uk]
Sent: 30 July 2003 13:00
To: Marsland, John
Cc: 'r-help at lists.R-project.org'
Subject: Re: [R] building packages using S4 methods


On Wed, 30 Jul 2003, Marsland, John wrote:

I have been building a package around a sequence of S4

classes which I have

coded in separate *.R files in the "./R" subdirectory of

the package.

The package builds without error, but when I load it in R I get:

[Incidentally, packages with R syntax errors `build without 
errors': that
is not a useful test.]

Error in reconcilePropertiesAndPrototype(name, slots, prototype,
superClasses) :
	Class "xxxx" extends an undefined class ("yyyyyy"

I guess R is trying to source the *.R files in the wrong

order? Since both

Or perhaps you didn't give them names in the right order?  
The files are
not in fact source-d, but they are concatenated in 
alphabetical order and 
the concatenated file is parsed and evaluated (at load time 
unless you 
used --save: see below).

`alphabetical order' is potentially locale-dependent, 
although we try to 
ensure that the C locale is used, I don't think this is bound 
to work on 
Windows.

classes are defined in the package and work fine at the

command prompt.

Is it just that this has not been an issue before owing to R's lazy
evaluation?
Is there anything I can do about this... maybe by putting

something in the

"./install.R"?

How are you building the package?  Are you using --save?  If 
not, that is 
probably the solution as it is highly recommended for 
packages using S4 
classes.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

********************************************************************** 
This is a commercial communication from Commerzbank AG.\ \ T...{{dropped}}

Brian Ripley

Wed, Jul 30, 2003 5:40 AM #

On Wed, 30 Jul 2003, Marsland, John wrote:

First point: Rcmd is JUST A FRONT END like R CMD on Unix: what do you want
detailed documentation on that for?  It's *build* you want to be looking
at.

Second point: build --binary is one way to distribute a fully-debugged
package, but INSTALL --build is a better way (as it allows more options).  
In either case you don't want to package up a non-functional package, do
you?

Third point: you can use install.R to provoke saving: that's described
in the help for INSTALL.

The moral is:  use R CMD/Rcmd INSTALL to install a package, and once it is
working, use R CMD/Rcmd build to make a source distribution or (on
Windows) Rcmd INSTALL --build to make a binary distribution

Unsurprisingly it is ?INSTALL that has help on the installation options.


[I presume the answer to my question was that you were not using --save,
but do please try to answer questions put to you by the helpers.]

[...]

Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595