grouply.R

grouply <- function(f, ...) {
  groups <- lazyeval::lazy_dots(...)
  
  function(tbl, ...) {
    dplyr::group_by_(tbl, .dots = groups) %>%
      f(...) %>%
      dplyr::ungroup()
  }
}

mtcars %>% grouply(mutate, cyl, am)(n = n())

mutate_by.R

#' Mutate within (transient) groups
#'
#' @params ...    Variables to group by. See \link[dplyr]{group_by} for details.
#' @params .dots  Used to work around non-standard evaluation. See \code{vignette("nse")} for details.
#' @params add    See note below
#'
#' @return a function (wrapping \link[dplyr]{mutate_})
#'
#' @note Should `add` default to `TRUE`?
#' @note Should we completely ungroup the result,
#'       or restore the original grouping?
#'       (as much as possible, if `add` = `FALSE`?)
#'
#' @importFrom dplyr group_by_ mutate_
#'
#' @examples
#'  library(dplyr)
#'  mtcars %>% mutate_by(cyl, am)(n = n())
#'  mtcars %>% mutate_by_(.dots = c("cyl", "am"))(n = n())
#'
#' @export
mutate_by <- function (..., add = TRUE) {
  by_dots <- lazyeval::lazy_dots(...)
  mutate_by_(.dots = by_dots, add = add)
}

#' @describeIn mutate_by
#' @export
mutate_by_ <- function (..., .dots, add = TRUE) {
  all_by_dots <- lazyeval::all_dots(.dots, ..., all_named = TRUE)
  function (.data, ..., .dots) {
    grouped <- group_by_(.data, .dots = all_by_dots, add = add)

    # why u no work?
    # mutations <- lazyeval::all_dots(..., .dots)
    mutations <- lazyeval::lazy_dots(...)

    mutated <- mutate_(grouped, .dots = mutations)
    group_by_(mutated, .dots = as.character(groups(.data))) # restore grouping of .data
  }
}

if (interactive()) {
  mtcars %>% mutate_by_("cyl", "am")(n = n(), .dots = c(foo = "bar"))
  mtcars %>% mutate_by(cyl, am)(n = n())
}

Upon some consideration, I realized there's another useful "preposition" here: where. mutate_where would be handy and reduce the need for ifelse, and this question lays out a need for summarise_where, though I'm not quite as sold on it.