joining with class interval

[greg-botwin]

This is my first time posting on GitHub so I apologize in advance if I am missing needed information or if I am doing anything incorrectly. I also want to thank the developers for making an amazing package!

I ran into a possible bug when joining a data frame that contained an interval class column (lubridate). The interval class seems to be joined by position, not via the "by" variable. I think this is related to 2432, but I did not see this specific case described. Using merge produces the correct data frame.

Appreciate your support!

library(lubridate)
library(dplyr)
library(reprex)

df1 <- data.frame(id = seq(from = 1, to = 10, by = 1),
                  interval_start = seq.Date(from = ymd("2017-03-01"), to = ymd("2017-03-10"), by = 1))

df1 <- df1 %>%
  mutate(interval_end = interval_start + days(10)) %>%
  mutate(ten_d_interval = interval(start = interval_start, end = interval_end))

df2 <- data.frame(id = c(1,2,3,4,5,10,9,8,7,6),
                  letters = letters[1:10])

df2_join_df1 <- inner_join(df2, df1, by = "id")

# incorrectly joined df, see bottom five rows
df2_join_df1
#>    id letters interval_start interval_end                 ten_d_interval
#> 1   1       a     2017-03-01   2017-03-11 2017-03-01 UTC--2017-03-11 UTC
#> 2   2       b     2017-03-02   2017-03-12 2017-03-02 UTC--2017-03-12 UTC
#> 3   3       c     2017-03-03   2017-03-13 2017-03-03 UTC--2017-03-13 UTC
#> 4   4       d     2017-03-04   2017-03-14 2017-03-04 UTC--2017-03-14 UTC
#> 5   5       e     2017-03-05   2017-03-15 2017-03-05 UTC--2017-03-15 UTC
#> 6  10       f     2017-03-10   2017-03-20 2017-03-06 UTC--2017-03-16 UTC
#> 7   9       g     2017-03-09   2017-03-19 2017-03-07 UTC--2017-03-17 UTC
#> 8   8       h     2017-03-08   2017-03-18 2017-03-08 UTC--2017-03-18 UTC
#> 9   7       i     2017-03-07   2017-03-17 2017-03-09 UTC--2017-03-19 UTC
#> 10  6       j     2017-03-06   2017-03-16 2017-03-10 UTC--2017-03-20 UTC

df2_merge_df1 <- merge(df2, df1, by = "id")

# correcty merged df
df2_merge_df1
#>    id letters interval_start interval_end                 ten_d_interval
#> 1   1       a     2017-03-01   2017-03-11 2017-03-01 UTC--2017-03-11 UTC
#> 2   2       b     2017-03-02   2017-03-12 2017-03-02 UTC--2017-03-12 UTC
#> 3   3       c     2017-03-03   2017-03-13 2017-03-03 UTC--2017-03-13 UTC
#> 4   4       d     2017-03-04   2017-03-14 2017-03-04 UTC--2017-03-14 UTC
#> 5   5       e     2017-03-05   2017-03-15 2017-03-05 UTC--2017-03-15 UTC
#> 6   6       j     2017-03-06   2017-03-16 2017-03-06 UTC--2017-03-16 UTC
#> 7   7       i     2017-03-07   2017-03-17 2017-03-07 UTC--2017-03-17 UTC
#> 8   8       h     2017-03-08   2017-03-18 2017-03-08 UTC--2017-03-18 UTC
#> 9   9       g     2017-03-09   2017-03-19 2017-03-09 UTC--2017-03-19 UTC
#> 10 10       f     2017-03-10   2017-03-20 2017-03-10 UTC--2017-03-20 UTC

[krlmlr]

Thanks! This doesn't seem to be related to #2432. As far as I can tell, the data frames differ only by the order of rows? Looks like inner_join() maintains the order of the lhs data frame, and merge() returns the data in the order of id.

[greg-botwin]

Thanks for looking into this!

The data frames do differ by the order of the rows, but I think that is expected output.

What is specifically different, and what I believe to be incorrect, is that in df2_join_d1 the actual value of the interval object changes during the join.

For example in df1 prior to the join, look at row id 10. The interval has a start date of 2017-03-10, end date of 2017-03-20, and a correct interval object of 2017-03-10 UTC--2017-03-20 UTC. But, after the join in df2_join_df1, the value of the interval object is changed and is reported as 2017-03-06 UTC--2017-03-16 UTC. This is incorrect. For some reason the class interval object changes during the join.

Hope this helps clarify! Greatly appreciative of your support.

[krlmlr]

Thanks for walking me through, I somehow missed the obvious. So: The row order is different, this is by design (and shouldn't be relevant). The ten_d_interval column is wrong, this will be fixed with #2432.

[greg-botwin]

great! thanks for looking into this.