R devel - May 2020 - issues with environment handling in model.frame()

Dear all,

model.frame behaves in a way I don't expect when both its formula and
subset argument are passed through a function call.

This works as expected:

model.frame(~wool, warpbreaks, breaks < 15)
#>    wool
#> 14    A
#> 23    A
#> 29    B
#> 50    B
fun1 <- function(y) model.frame(~wool, warpbreaks, y)
fun1(with(warpbreaks, breaks < 15))
#>    wool
#> 14    A
#> 23    A
#> 29    B
#> 50    B

but this doesn't:

fun2 <- function(x, y) model.frame(x, warpbreaks, y)
fun2(~wool, with(warpbreaks, breaks < 15))
#> Error in eval(substitute(subset), data, env): object 'y' not found

model.frame is used by xtabs() and aggregate() so the following won't work
either:
fun3 <- function(x, y) xtabs(x, warpbreaks, y)
fun3(~wool, with(warpbreaks, breaks < 15))
#> Error in eval(substitute(subset), data, env): object 'y' not found

fun4 <- function(x, y) aggregate(x, warpbreaks, length, subset = y)
fun4(breaks ~ wool, with(warpbreaks, breaks < 15))
#> Error in eval(substitute(subset), data, env): object 'y' not found

The subset argument uses NSE, though no help file among aggregate, xtabs or
model.frame suggests it as far as I can see, and I assume the expectation
that NSE will be used led to overlook the problematic case above.

The SO question prompted this bug report :
https://stackoverflow.com/questions/61145404

Best regards,

Antoine

	[[alternative HTML version deleted]]

> On May 2, 2020, at 5:30 AM, Antoine Fabri <antoine.fabri at
gmail.com> wrote:
> 
> Dear all,
> 
> model.frame behaves in a way I don't expect when both its formula and
> subset argument are passed through a function call.
> 
See the help page

?formula

in particular the section headed 'Environments'.

Then look at the help for 

?model.frame

which says

"All the variables in formula, subset and in ... are looked for first in
data and then in the environment of formula (see the help for formula for
further details) and collected into a data frame."

> This works as expected:
> 
> model.frame(~wool, warpbreaks, breaks < 15)
> #>    wool
> #> 14    A
> #> 23    A
> #> 29    B
> #> 50    B
> fun1 <- function(y) model.frame(~wool, warpbreaks, y)
> fun1(with(warpbreaks, breaks < 15))
> #>    wool
> #> 14    A
> #> 23    A
> #> 29    B
> #> 50    B
> 
Here the fornmula `~wool' has the environment created when fun1 is called
and so does `y'.

So model.frame finds them both after first looking for them in `warpbreaks'.

> but this doesn't:
> 
> fun2 <- function(x, y) model.frame(x, warpbreaks, y)
> fun2(~wool, with(warpbreaks, breaks < 15))
> #> Error in eval(substitute(subset), data, env): object 'y' not
found
> 

Here the formula has environment `<environment: R_GlobalEnv>' and
`y' has the environment created when fun2 is called.

So model.frame looks in warpbreaks, then in `<environment:
R_GlobalEnv>' and doesn't find y in either.

Similar issues apply in the use of formulas below.
> model.frame is used by xtabs() and aggregate() so the following won't
work
> either:
> fun3 <- function(x, y) xtabs(x, warpbreaks, y)
> fun3(~wool, with(warpbreaks, breaks < 15))
> #> Error in eval(substitute(subset), data, env): object 'y' not
found
> 
> fun4 <- function(x, y) aggregate(x, warpbreaks, length, subset = y)
> fun4(breaks ~ wool, with(warpbreaks, breaks < 15))
> #> Error in eval(substitute(subset), data, env): object 'y' not
found
> 

Each of these cases can be made to work by inserting

	environment(x) <- environment()

before the call to model.frame in each of your fun[2-4] functions.

However, this can lead to headaches downstream if you need to save the formula
and use it in another function.  If this is where you are headed some time spent
studying lm() and methods for "lm" objects may help.

[rest deleted]

HTH,

Chuck