filter on lubridate interval does not filter start slot

[ajholguin]

It looks like filter is sub-setting the ".Data" slot, but not the corresponding "start" slot.

library(dplyr)
library(lubridate)

dat <- data.frame(id = 1:2,
                  intervals = c(interval(ymd(20090201), ymd(20090202)), interval(ymd(20090202), ymd(20090203))))

dat_filt <- dat %>% 
  filter(id == 1)

# data slot is filtered, but start slot is not
str(dat_filt$intervals)

result:

Formal class 'Interval' [package "lubridate"] with 3 slots
  ..@ .Data: num 86400
  ..@ start: POSIXct[1:2], format: "2009-02-01" "2009-02-02"
  ..@ tzone: chr "UTC"

length(dat_filt$intervals)         # --> 1
length(dat_filt$intervals@start)   # --> 2

[hadley]

@romainfrancois could you please take a look at what it would take to fix this?

[krlmlr]

Perhaps we can call [ from the C++ code in the general case, this will also get us POSIXlt support for free.

[hadley]

I think we need specialised methods for performance, but maybe we can fall back for unknown types. I'm planning on moving this code in r-lib/vctrs#7 (at least the low level vector stuff, then the data frame assembly can remain in dplyr, or possibly move to tibble)

[krlmlr]

r-lib/vctrs#7 is about coercion, but I agree it's good to have that logic available in C++ and fall back to R only as a last resort.

[lelandbr]

Just fyi, what I'm assuming is this same issue also affects slice, and arrange, as well as summarize with nth, first, or last

Here is a test df I made for #2214, which I'll include here bc I think I might be seeing a slightly different phenotype from @ajholguin above.

Each column is a different class. Either a duration (1:6 days) or a date (1st through 6th), so that incorrect row subsetting or ordering will be easily apparent. lubridate columns are on the right.

df <- tibble(grps          = c("a","a","b","b","c","c"), 
             to.fill       = c(1,NA,3,NA,5,NA),
             to.arrange    = 6:1,
             base.Date     = as.Date(0:5,origin = "1970-01-01"),
             base.POSIXct  = as.POSIXct(as.character(base.Date)),
             # base.POSIXlt: tibble does not allow POSIXlt columns
             base.difftime = as.difftime(1:6,units="days"),
             date.date     = date::as.date(1:6), #doesn't display very well in a data.frame
             chron.dates   = chron::chron(0:5),
             chron.times   = chron::chron(times. = (1:6)),
             lubr.Interval = lubridate::interval( as_date(0:5), as_date(10:15) ),
             lubr.Period   = lubridate::period(days=1:6),
             lubr.Duration = lubridate::ddays(1:6)
             )

And here are what I assume must be the same or at least very similar to the filter issue:

# slice is also affected, in addition to filter:
df %>% filter(to.fill %in% c(1,3,5))
df %>% slice(c(1,3,5))
df[c(1,3,5),] # this should be the correct output

# same with arrange:
df %>% arrange(to.arrange)
df[6:1,] # this should be the correct output

# sample_n seems to work though
df %>% sample_n(3)

# group_by() %>% summarize(nth), same problem?
df %>% group_by(grps) %>% summarize_all( funs(nth(.,1)) )

# Similar problem with first() and last(), 
# but with a twist ... there are some NAs and 0s introduced
df %>% group_by(grps) %>% summarize_all(first) 
df %>% group_by(grps) %>% summarize_all(last) 

[krlmlr]

Should be taken care of in r-lib/vctrs#27.

[hadley]

Now part of #2432