In R, making packages with functions that are by() and with() compliant - r

Question

Working on improving my package summarytools, I'm looking for a way to use the information on each of by()'s groups to integrate this info in some function's output. To give a little bit more of a context, the functions in this package print out the dataframe name and variable name(s) being summarized. Functions like by() make it difficult because they use generic names such as dd[x, ] when slicing the data and feeding it to functions. substitute() is thus not an option to get at the x parameter in that case, and the values of the IND variable(s) are also hidden (to a certain level).
To illustrate, in the following example, the group information (c.g. "gender: F" and "smoker: No") is simply printed out with cat() when print.by() is invoked, using attributes of the object of class "by":
dat <- data.frame(gender=rep(c("F","M"),each=15),
smoker=rep(c("Yes", "No")),
someQty=runif(n = 30,min = 0, max = 10))
by(dat$someQty, INDICES = list(gender=dat$gender, smoker=dat$smoker), FUN = mean)
## gender: F
## smoker: No
## [1] 5.560505
## -------------------------------------------------------------------------------
## gender: M
## smoker: No
## [1] 2.568055
## -------------------------------------------------------------------------------
## gender: F
## smoker: Yes
## [1] 4.057938
## -------------------------------------------------------------------------------
## gender: M
## smoker: Yes
## [1] 3.416027
Now what I need is to get the info for each group during the by-group processing (as opposed to recuperating them after the "by" object has been created).
I worked on a solution, but before I repeat a similar work for making functions comply with with(), %>%, and possibly others... and their combinations, I'm wondering if there might be a simpler approach to this.
Here's what I have so far do deal with by():
# Initialise variable in package-specific environment that
# will help keeping track of the by-processing
myenv <- new.env()
myenv$byInfo <- list()
# Declare some function that will return the `by` variables values
# at each iteration (it's a sort of dummy function that does just that)
myfunc <- function(x) {
sc <- sys.calls()
sf <- sys.frames()
# Find position of by.default() and tapply() in the sys.calls list
by_pos <- which(as.character(lapply(sc, head, 1))=="by.default()")
tapply_pos <- which(as.character(lapply(sc, head, 1))=="tapply()")
if (length(by_pos) == 1) {
# check if this is the first "by" iteration
if(length(myenv$byInfo) == 0) {
# Standardise the call (adds argument names)
by_call <- as.list(pryr::standardise_call(sc[[by_pos]]))
# Extract the data argument
by_data <- deparse(by_call$data)
# Extract the IND variable names
by_IND <- as.character(by_call$IND)
by_IND <- by_IND[-which(by_IND=="list")]
# Get the levels of these IND variables
by_levels <- sf[[tapply_pos]]$namelist
levels_df <- expand.grid(by_levels, stringsAsFactors = FALSE)
# Store the info in the package-specific environment
myenv$byInfo$iter <- 1
myenv$byInfo$levels_df <- levels_df
myenv$byInfo$nb_iter <- nrow(levels_df)
}
levels_df <- myenv$byInfo$levels_df
info <- paste(colnames(myenv$byInfo$levels_df),
as.character(myenv$byInfo$levels_df[myenv$byInfo$iter,]),
sep=" = ", collapse = ", ")
if (myenv$byInfo$iter == myenv$byInfo$nb_iter)
myenv$byInfo <- list()
else
myenv$byInfo$iter = myenv$byInfo$iter + 1
return(info)
}
return()
}
b <- by(data = dat$someQty,
INDICES = list(gender = dat$gender, smoker = dat$smoker),
FUN = myfunc)
b[1:4]
## [1] "gender = F, smoker = No" "gender = M, smoker = No"
## [3] "gender = F, smoker = Yes" "gender = M, smoker = Yes"
So yes, it does give me what I want, but I'd like to know if I'm missing something more straightforward here.
Note: I thought adding a by= parameter to some functions and just ignore base R's by() altogether but I'd rather use the preexisting base functions people are accustomed to.