terminology for frames and environments

Peter Dalgaard <p.dalgaard at biostat.ku.dk>

 
Gabor Grothendieck <ggrothendieck at myway.com> writes:

Thomas Lumley <tlumley <at> u.washington.edu> writes:

The distinction between "environment" and "frame" is important. The frame
is what you find things in with get(, inherits=FALSE) and the environment
uses get(, environment=TRUE).

The thing I find odd about this one is that if we have:

e <- new.env()
e$x <- 1
f <- new.env(parent=e)
f$x # gives an error

(Actually not. I get NULL)

Sure.

then I would have expected x to be returned since f is an environment
and x is in that environment. On the other hand, if an environment
is defined to be the same as a frame (and there is some other word for 
an environment and its ancestors) then the above notation makes sense.

Why? f$x is documented to be basically equivalent to
get("x",f,inherits=FALSE). In contrast,

evalq(x,f)

[1] 1

works fine.

Well this certainly is more environment-like if one defines environment
as a frame plus ancestors but it does not address the issue that 
if f is an environment in that sense then $ "should" pick out its
components.  One can define + to mean subtraction but that's not natural.

I agree that it would have been nice if environment had been used to
mean a frame plus its ancestors and class(f) was "frame", not "environment"
but, of course, that's not how it turned out.

I think its better to maintain backward compatiblity as far as possible and
make the terms suit rather than use environment in a way which is
inconsistent with the way R works.

In fact, it might have been nice if there were both a frame class and an
environment class so one could have either behavior.  With this new
setup R might have worked like this:

	# create parent environment 
        # (in the sense of frame+ancestors) & populate
	Pe <- new.env(): Pe$x <- 1

	# create child environment
	Ce <- new.env(parent=Pe)

	# create corresponding frames
	Pf <- as.frame(Pe); Cf <- as.frame(Ce)

	# x exists in child environment
	exists(Ce$x) # TRUE

	# but not in child frame
	exists(Cf$x) # FALSE

Unfortunately this can't be done without breaking backward compatibility
although it would still be possible to come up with a new word for
environment plus ancestors (call it the envlist class) using environment 
as a synonym for frame.  In that case this example would become:

	# create parent using envlist class & populate
	Pel <- new.envlist(): Pela$x <- 1

	# create child environment+ancestors class
	Cel <- new.envlist(parent=Pel)

	# create corresponding environments (in the sense of frames)
	Pe <- as.env(Pel); Ce <- as.env(Cel)

	# x exists in child environment
	exists(Cel$x) # TRUE

	# but not in child frame
	exists(Ce$x) # FALSE

and this one is upwardly compatible with the current R.