Scoping of variables in aes(…) inside a function i

问题:

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

回答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())