Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
The error is: https://github.com/ursacomputing/crossbow/runs/3696604088#step:7:15209
Timezone not present, cannot convert to string with timezone: %Y-%m-%d%z
Ok, I think I’ve got close to the source of the issue (though let me tell you neither the source nor the docs for R are accurate [1] on this…)
3.4 (and earlier):
> attributes(c(lubridate::ymd_hms("2018-10-07 19:04:05", tz = "Etc/GMT+6"), NA)) $class [1] "POSIXct" "POSIXt"
3.5 (and later):
> attributes(c(lubridate::ymd_hms("2018-10-07 19:04:05", tz = "Etc/GMT+6"), NA)) $class [1] "POSIXct" "POSIXt" $tzone [1] "Etc/GMT+6"
So R itself is dropping the tzone attribute when we use c(), that is being passed to Arrow as such, and then when Arrow goes to print a timezone it (rightfully!) complains that there is no timezone to be formatted into the string.
This behavior actually sounds right (given the inputs), so I propose that we catch the error R <=3.4 (or skip the test in r<=3.4)
[1] - The documented behavior is current, but it didn't change at the same time as the actual behavior.
The docs starting in 4.1.0 state:
Using \code{\link{c}} on \code
current time zone, and on \code{"POSIXct"} objects drops any
\code{"tzone"} attributes, unless they are all marked with the same
time zone.
https://github.com/wch/r-source/blob/tags/R-4-1-0/src/library/base/man/DateTimeClasses.Rd#L180-L183
The docs before that state:
Using \code{\link{c}} on \code{"POSIXlt"}
objects converts them to the
current time zone, and on \code
objects drops any
\code
attributes (even if they are all marked with the same
time zone).
https://github.com/wch/r-source/blob/tags/R-4-0-5/src/library/base/man/DateTimeClasses.Rd
Attachments
Issue Links
- links to
