[Rd] An exercise in the use of 'substitute'

[Douglas Bates]

I would like to create a method for the generic function "with" applied 
to a class of fitted models.  The method should do two things:

1. Substitute the name of the first argument for '.' throughout the 
expression

2. Evaluate the modified expression using the data argument to the 
fitted model as the first element of the search list.

The second part is relatively easy.  The default method for "with" has body
   eval(substitute(expr), data, enclos = parent.frame())
and you just change this to
   eval(substitute(expr), eval(data$call$data), enclos = parent.frame())

So, for example

 > fm <- lm(optden ~ carb, Formaldehyde)
 > with.lm <- function(data, expr, ...) eval(substitute(expr), 
eval(data$call$data), enclos = parent.frame())
 > with(fm, carb)
[1] 0.1 0.3 0.5 0.6 0.7 0.9

However, I haven't been able to work out a clever way of using 
substitute to get the first part.  I would like to be able to call, e.g.

with(fm, xyplot(resid(.) ~ carb))

and get a plot of resid(fm) ~ Formaldehyde$carb

It is possible to do the first part by deparsing, substituting, and 
parsing but that's inelegant.  Can anyone suggest a more elegant method?

BTW, the example of an lm model is just for illustration.  The actual 
use I have in mind is for lme (now lmer) models.  The plot method for 
the lme class in the nlme package does something very similar to this.