Change in behavior using summarise with max(. ,na.rm=TRUE) from 0.4.2 -> 0.4.3

[felasa]

Example:

toy data

library(dplyr)

foo <- 
  data.frame(id = c(1,1,2,2,3,3), 
             date_of_x = as.Date(c("2015-01-01", "2015-01-02", NA, "2015-02-02",
                           NA,NA)),
             type = c("A","B","A","B","A","B")) %>% tbl_df

foo
Source: local data frame [6 x 3]

     id  date_of_x   type
  (dbl)     (date) (fctr)
1     1 2015-01-01      A
2     1 2015-01-02      B
3     2       <NA>      A
4     2 2015-02-02      B
5     3       <NA>      A
6     3       <NA>      B

0.4.2

foo %>% group_by(id) %>% 
  summarise(most_recent = max(date_of_x, na.rm=TRUE),
            confirmed = !is.na(most_recent))

Source: local data frame [3 x 3]

  id most_recent confirmed
1  1  2015-01-02      TRUE
2  2  2015-02-02      TRUE
3  3        <NA>     FALSE

in 0.4.3:

foo %>% group_by(id) %>% 
  summarise(most_recent = max(date_of_x, na.rm=TRUE),
            confirmed = !is.na(most_recent))

Source: local data frame [3 x 3]

    id most_recent confirmed
(dbl)      (date)     (lgl)
1     1  2015-01-02      TRUE
2     2  2015-02-02      TRUE
3     3        <NA>      TRUE
Warning message:
In max.default(c(NA_real_, NA_real_), na.rm = TRUE) :
 no non-missing arguments to max; returning -Inf

in 0.4.3, it returns a value of class Inf and in 0.4.2 one of class NA

Change is consistent with the behavior of max I guess but it broke some workflows (maybe just mine but I am unsure if i could just replace is.na with is.infinite and fix everything).

May be working as it should, although im confused why it changed at all (i.e. why the previous one worked).

[hadley]

This seems like a bug in max.Date() in R. I'll post to R-devel

[felasa]

However bizarre the base R methods handling dates, this doesn't account for the difference in outputs with dplyr versions (since nothing changed in base R in bewteen).

The issue I was pointing out is the difference in the 'confirmed' columns from the examples above.

(having some issues readapting code for .3, so i'm venting here)