Scoping of variables in aes(...) inside a function in ggplot - r

Question

Consider this use of ggplot(...) inside a function.
x <- seq(1,10,by=0.1)
df <- data.frame(x,y1=x, y2=cos(2*x)/(1+x))
library(ggplot2)
gg.fun <- function(){
i=2
plot(ggplot(df,aes(x=x,y=df[,i]))+geom_line())
}
if(exists("i")) remove(i)
gg.fun()
# Error in `[.data.frame`(df, , i) : object 'i' not found
i=3
gg.fun() # plots df[,3] vs. x
It looks like ggplot does not recognize the variable i defined inside the function, but does recognize i if it is defined in the global environment. Why is that?
Note that this gives the expected result.
gg.new <- function(){
i=2
plot(ggplot(data.frame(x=df$x,y=df[,i]),aes(x,y)) + geom_line())
}
if(exists("i")) remove(i)
gg.new() # plots df[,2] vs. x
i=3
gg.new() # also plots df[,2] vs. x

Answer 1

Let's return a non-rendered ggplot object to see what's going on:
gg.str <- function() {
i=2
str(ggplot(df,aes(x=x,y=df[,i]))+geom_line())
}
gg.str()
List of 9
$ data :'data.frame': 91 obs. of 3 variables:
..$ x : num [1:91] 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 ...
..$ y1: num [1:91] 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 ...
..$ y2: num [1:91] -0.208 -0.28 -0.335 -0.373 -0.393 ...
$ layers :List of 1
..$ :Classes 'proto', 'environment' <environment: 0x0000000009886ca0>
$ scales :Reference class 'Scales' [package "ggplot2"] with 1 fields
..$ scales: list()
..and 21 methods, of which 9 are possibly relevant:
.. add, clone, find, get_scales, has_scale, initialize, input, n, non_position_scales
$ mapping :List of 2
..$ x: symbol x
..$ y: language df[, i]
$ theme : list()
$ coordinates:List of 1
..$ limits:List of 2
.. ..$ x: NULL
.. ..$ y: NULL
..- attr(*, "class")= chr [1:2] "cartesian" "coord"
$ facet :List of 1
..$ shrink: logi TRUE
..- attr(*, "class")= chr [1:2] "null" "facet"
$ plot_env :<environment: R_GlobalEnv>
$ labels :List of 2
..$ x: chr "x"
..$ y: chr "df[, i]"
- attr(*, "class")= chr [1:2] "gg" "ggplot"
As we can see, mapping for y is simply an unevaluated expression. Now, when we ask to do the actual plotting, the expression is evaluated within plot_env, which is global. I do not know why it is done so; I believe there are reasons for that.
Here's a demo that can override this behaviour:
gg.envir <- function(envir=environment()) {
i=2
p <- ggplot(df,aes(x=x,y=df[,i]))+geom_line()
p$plot_env <- envir
plot(p)
}
# evaluation in local environment; ok
gg.envir()
# evaluation in global environment (same as default); fails if no i
gg.envir(environment())