timezone parameter for scale_x_datetime has no effect

[swebs]

I am trying to use the timezone parameter to scale_x_datetime to control how the tick labels are displayed, but I can't get it to work. Changing the timezone does not seem to make a difference to plot output. This probably applies equally to scale_y_datetime. Potentially I am not using the correct syntax and the documentation could be expanded.

library(tidyverse)
library(lubridate)

plotdata <- tribble(
  ~t_stamp, ~value,
  "2019-08-17T06:00:00", 1,
  "2019-08-17T07:00:00", 2,
  "2019-08-17T08:00:00", 3
)

plotdata <- plotdata %>% 
  mutate(t_stamp=force_tz(ymd_hms(t_stamp),tzone="US/Mountain"))

ggplot(plotdata) +
  geom_point(mapping=aes(x=t_stamp, y=value)) +
  scale_x_datetime(timezone = "US/Mountain")

ggplot(plotdata) +
  geom_point(mapping=aes(x=t_stamp, y=value)) +
  scale_x_datetime(timezone = "US/Eastern")

# both plots are the same

[smouksassi]

seems timezone not applied when the date_labels are a waiver ?
below work

ggplot(plotdata) +
  geom_point(mapping=aes(x=t_stamp, y=value)) +
  scale_x_datetime(date_labels= "%H:%M",
                   timezone = "US/Eastern")

ggplot(plotdata) +
  geom_point(mapping=aes(x=t_stamp, y=value)) +
  scale_x_continuous(trans = time_trans(tz = "US/Eastern"))

[yutannihilation]

Thanks, it actually processed differently depending on whether data_labels() is supplied or not:

ggplot2/R/scale-date.r

Lines 285 to 290 in f802c5e

	if (!is.waive(date_labels)) {
	labels <- function(self, x) {
	tz <- if (is.null(self$timezone)) "UTC" else self$timezone
	date_format(date_labels, tz)(x)
	}
	}

But, even if it's waive(), it should be handled as trans:

ggplot2/R/scale-date.r

Line 336 in f802c5e

self$trans <- time_trans(self$timezone)

However, scales::time_trans() seems to ignore the timezone...? I'm not yet sure what is the expected behaviour here...

library(lubridate)
#> 
#> Attaching package: 'lubridate'
#> The following objects are masked from 'package:base':
#> 
#>     date, intersect, setdiff, union

x <- c(
  "2019-08-17T06:00:00",
  "2019-08-17T07:00:00",
  "2019-08-17T08:00:00"
)

x <- force_tz(ymd_hms(x), tzone = "US/Mountain")

# all returns the same results
t1 <- scales::time_trans("US/Mountain")
t1$transform(x)
#> [1] 1566043200 1566046800 1566050400
t1$format(t1$breaks(range(x)))
#> [1] "06:00" "06:30" "07:00" "07:30" "08:00"

t2 <- scales::time_trans("US/Eastern")
t2$transform(x)
#> [1] 1566043200 1566046800 1566050400
t2$format(t2$breaks(range(x)))
#> [1] "06:00" "06:30" "07:00" "07:30" "08:00"

^{Created on 2020-05-19 by the reprex package (v0.3.0)}

[swebs]

Aha, but look at this, these are different:

> t1$inverse(t1$transform(x))
[1] "2019-08-17 06:00:00 MDT" "2019-08-17 07:00:00 MDT" "2019-08-17 08:00:00 MDT"
> t2$inverse(t2$transform(x))
[1] "2019-08-17 08:00:00 EDT" "2019-08-17 09:00:00 EDT" "2019-08-17 10:00:00 EDT"

[swebs]

Ok, so, I think to summarize: I believe the intention is that when you pass the timezone parameter, the tick labels will use that timezone. This works when you also specify date_labels but not otherwise, so I think this is an "actual problem." It seems like the issue is either in scales or is due to a misinterpretation of the way scales::time_trans is supposed to work by the ggplot2 authors.

Can anyone confirm if my above interpretation is correct? If so, any thoughts as to whether this is a ggplot2 issue or scales issue?

[thomasp85]

This is a bug in ggplot2

[thomasp85]

Just leaving some comments for whoever ends up fixing this:

The issue is that the trans object is never constructed with the given timezone. The code pointed to by @yutannihilation is only for fetching a timezone from the data if none is given.
Second problem is that the breaks function in the transformation returns a named vector and thus the names of the vector is used unquestioned in get_labels() so even if the trans object was constructed correctly it wouldn't matter