'nth' inside 'summarize' returns an error when 'which' is used to calculate the argument 'n'

[CameronBieganek]

Here's a toy example:

> set.seed(11)
> df <- data.frame(x = c(1:4, 7:9, 13:19), y = sample(100:999, 14))
> df
    x   y
1   1 349
2   2 100
3   3 558
4   4 112
5   7 157
6   8 954
7   9 177
8  13 358
9  14 885
10 15 209
11 16 255
12 17 491
13 18 905
14 19 854
> which(df$x == 16)
[1] 11
> summarize(
+     df,
+     abc = nth(y, n = which(x == 16))
+ )
Error: 'n' should be a scalar integer

[hadley]

Why are you doing that and not just y[which(x == 16)]?

[CameronBieganek]

Because I want to use this feature of nth:

> vec <- 5:8
> vec[0]
integer(0)
> nth(vec, n = 0)
[1] NA

My data frame has some missing data, so sometimes the boolean inside the which is all false, and then which returns integer(0). So my actual use case looks something like this:

summarize(
    df,
    abc = nth(y, n = sum(which(x == 16)))
)

That way when the which fails to find anything and returns integer(0), the sum will return 0, and the nth should return NA.

[romainfrancois]

What I've done is fallback to R evaluation whenever n is not a length one integer or numeric, e.g. an expression for R to evaluate.