Joins are too strict about tz attributes

[hadley]

I think this is going to cause major pain for people

library(dplyr, warn.conflicts = FALSE)
library(nycflights13)

hourly_delay <- flights %>% 
  group_by(time_hour, origin) %>% 
  summarise(delay = mean(dep_delay, na.rm = TRUE))
    
hourly_delay %>% 
  inner_join(weather, by = c("time_hour", "origin"))
#> Error in inner_join_impl(x, y, by$x, by$y, suffix$x, suffix$y, accept_na_match): attributes are different


attributes(flights$time_hour)
#> $class
#> [1] "POSIXct" "POSIXt" 
#> 
#> $tzone
#> [1] "UTC"
attributes(weather$time_hour)
#> $class
#> [1] "POSIXct" "POSIXt" 
#> 
#> $tzone
#> [1] ""

@zeehio would this be easy to fix?

@krlmlr could we get a better error message here? Ideally needs to give variable names on both sides of the join

[zeehio]

The join logic goes through a different code path than the code paths I have worked with in the past (this is the third way I see to merge vectors in dplyr... I hope vctrs comes up soon!).

A quick fix can be done at (check_attribute_compatibility)[https://github.com/tidyverse/dplyr/blob/master/src/join.cpp#L44] so if both left and right inherit POSIXct and have the same time zone then the function returns without error. Consider "GMT", "UTC", and "" as equivalent. With that fix the POSIXct case should work fine. If nobody can push this patch I may do it, but I don't know when I'll have the time.

I believe that the best fix would be to make join_visitor use the Collecter.h logic, but that would be more work.

[zeehio]

Correction: ignore the "have the same time zone" constraint. If left and right inherit POSIXct then check_attribute_compatibility can return safely without error.

The timezone is just used for visualization, as far as I understand. So if someone joined different timezones the merge would be fine and the result would be printed in UTC (the default).

[zeehio]

I found the time to fix this. I added a test case. I did not work on the error messages though