Segfault caused by min function in summarise function

[Fija]

When apply min function on summarizing an empty data.frame, it would lead to segfault.

df <- tbl_df(data.frame(A=c(1)))
df <- df %>% filter(A > 1)
df %>% summarise(Min=min(A, na.rm=T))

more details see this StackOverflow Question

[romainfrancois]

@hadley should I also throw a warning like R does ?

[hadley]

I don't think so - R isn't consistent anyway compare sum() and max()

[bwlewis]

+/- Inf is not always expected; sorry I'm not good enough w/c++ to write a sol'n but I can give an example:

min(1[-)],na.rm=TRUE)
[1] Inf
Warning message:
In min(1[-1], na.rm = TRUE) :
  no non-missing arguments to min; returning Inf

>  min("a"[-1],na.rm=TRUE)
[1] NA
Warning message:
In min("a"[-1], na.rm = TRUE) : no non-missing arguments, returning NA

But in dplyr with the current commit:

> summarise(data.frame(A=1[-1]), Min=min(na.rm=TRUE))
  Min
1 Inf
Warning message:
In min(na.rm = TRUE) : no non-missing arguments to min; returning Inf

> summarise(data.frame(A="a"[-1]), Min=min(na.rm=TRUE))
  Min
1 Inf
Warning message:
In min(na.rm = TRUE) : no non-missing arguments to min; returning Inf

[bwlewis]

For the record I think R is idiosyncratic here. I kind of wish min(numeric()) returned numeric(0), and so on for other types. Oh well.

[hadley]

I think the contract for min() (and all other summary functions is that they always return a numeric vector of length 1, so I think returning a 0-length vector would be less helpful, not more.

[bwlewis]

Yes, that's what I meant by idiosyncratic. Mathematically, the minimum of a set S is the intersection of S with infimum(S), a.k.a. the smallest value in the set. Mathematica, Matlab, etc. return min(empty set) = empty set, consistent with that definition, R does not.

--Sorry, this is an R issue really, not a dplyr one. Not the best place to discuss it!