interval() and dplyr::filter() display incorrect intervals

[giovannotti]

Lubridate's interval() function seems to display incorrect negativ conditional results when I combine it with dplyr::filter().
This is the data frame to test:

library(lubridate)
library(dplyr)
test <- tibble::tibble(id = 1:2,
                       date1 = c(as.Date("1990-01-01", origin = "1970-01-01"), 
                                 as.Date("1993-01-01", origin = "1970-01-01")),
                       date2 = c(as.Date("1991-02-02", origin = "1970-01-01"), 
                                 as.Date("1992-02-02", origin = "1970-01-01")))

When filtered by negative interval condition with dplyr::filter() it displays wrong interval but interestingly right interval start:

test %>%
  mutate(int = interval(date1, date2), intstart = int_start(interval(date1, date2))) %>%
  filter(as.numeric(int) < 0)
  
#   # A tibble: 1 x 5
#      id      date1      date2                            int   intstart
#   <int>     <date>     <date>                 <S4: Interval>     <dttm>
# 1     2 1993-01-01 1992-02-02 1990-01-01 UTC--1989-02-01 UTC 1993-01-01

Without filter everything is fine:

test %>%
  mutate(int = interval(date1, date2), inststart = int_start(interval(date1, date2)))
  
#  A tibble: 2 x 5
#      id      date1      date2                            int  inststart
#   <int>     <date>     <date>                 <S4: Interval>     <dttm>
# 1     1 1990-01-01 1991-02-02 1990-01-01 UTC--1991-02-02 UTC 1990-01-01
# 2     2 1993-01-01 1992-02-02 1993-01-01 UTC--1992-02-02 UTC 1993-01-01

As well when using generic subsetting:

test %>%
  mutate(int = interval(date1, date2), inststart = int_start(interval(date1, date2))) %>%
  .[as.numeric(.$int) < 0,]
  
#  A tibble: 1 x 4
#      id      date1      date2                            int  inststart
#   <int>     <date>     <date>                 <S4: Interval>     <dttm>
# 1     2 1993-01-01 1992-02-02 1993-01-01 UTC--1992-02-02 UTC 1993-01-01

When I then try to extract the interval start with int_start() after filtering with dpylr::filter() I get an error suggesting that filter() doesn't filter the interval variable int.

test %>%
  mutate(int = interval(date1, date2)) %>%
  filter(as.numeric(int) < 0) %>%
  mutate(intstart = int_start(int))
  
  # Error in mutate_impl(.data, dots) : 
  # Column `intstart` must be length 1 (the number of rows), not 2

Session Output:

sessionInfo()

# R version 3.4.3 (2017-11-30)
# Platform: x86_64-w64-mingw32/x64 (64-bit)
# Running under: Windows >= 8 x64 (build 9200)

# Matrix products: default

# locale:
#  [1] LC_COLLATE=German_Germany.1252  LC_CTYPE=German_Germany.1252    LC_MONETARY=German_Germany.1252 LC_NUMERIC=C                   
#  [5] LC_TIME=German_Germany.1252    

# attached base packages:
#  [1] stats     graphics  grDevices utils     datasets  methods   base     

# other attached packages:
#  [1] dplyr_0.7.4     lubridate_1.7.1

# loaded via a namespace (and not attached):
#  [1] compiler_3.4.3   assertthat_0.2.0 R6_2.2.2         magrittr_1.5     tools_3.4.3      bindrcpp_0.2     glue_1.2.0       tibble_1.3.4    
#  [9] yaml_2.1.16      Rcpp_0.12.14     stringi_1.1.6    stringr_1.2.0    pkgconfig_2.0.1  rlang_0.1.4      bindr_0.1

[yutannihilation]

This is the issue on the dplyr's side. tidyverse/dplyr#3206

[giovannotti]

Ok, sry, I didn't realized it. Thanks.

[vspinu]

Yes. It's a well known issue not specific to lubridate (tidyverse/dplyr#2432)

[yutannihilation]

It's a well known issue

I feel this is a lessor-known known issue, yet the issue quite a few people may encounter, unfortunately... Considering the situation the issue will not going to be solved soon, some efforts are needed to make this really "known" (e.g. "Known Issues" section on some documentation?). I'm not saying lubridate is responsible for that, tough. ¯\_(ツ)_/¯