Using do.call with dplyr standard evaluation version

Question

How can I get a do.call with a variable list of arguments and functions to work with the standard evaluation version of summarise_ in dplyr?

## Some sample data, function, and variables to interpolate
set.seed(0)
dat <- data.frame(a=runif(10), b=runif(10))
fn <- function(x, y) IQR(x / y, na.rm = TRUE)
funs <- list(fn="fn")
targs <- list("a", "b")

This is the lazyeval::interp I'm trying to make work

library(dplyr)
interp(~do.call(fn, xs), .values=list(fn=funs$fn, xs=targs))
# ~do.call("fn", list("a", "b"))

but it doesnt work,

dat %>%
  summarise_(out = interp(~do.call(fn, xs), .values=list(fn=funs$fn, xs=targs)))

Expected result

dat %>%
  summarise(out = do.call(fn, list(a, b)))
#        out
# 1 1.084402

If I add in some print statements, I know the problem is that the "a" and "b" aren't being interpreted properly, but I haven't been able to figure out how to quote them properly.

fn <- function(x, y) { print(x); print(y); IQR(x / y, na.rm = TRUE) }
dat %>%
  summarise_(out = interp(~do.call(fn, xs), fn=funs$fn, xs=targs))
# [1] "a"
# [1] "b"
# Error: non-numeric argument to binary operator

Answer 1

The targs argument needs to be a call class. The variables in the call (a and b) need to be a name class. All this is done in the second (and third) line below. ?call, ?as.name, and ?is.language might make the line more understandable.

dat <- data.frame(a=runif(10), b=runif(10), grp=rep(1:2, each=5))
targs_quoted = do.call(call, c("list", lapply(targs, as.name)), quote=TRUE)
# In hardcoded form, targs_quoted = quote(list(a, b))
dat %>%
  group_by(grp) %>%
  summarise_(out = interp(~do.call(fn, xs), 
                          .values=list(fn=funs$fn, xs=targs_quoted)))

# Source: local data frame [2 x 2]
#     
#       grp       out
#     (int)     (dbl)
#  1     1  1.0754497
#  2     2  0.9892201

dplyr's "nse" (non-standard evaluation) vignette was very helpful here. I found that the . always referred to the entire table, not the grouped table. That's why some of the recommendations in the comments didn't "work" as wanted.