Issue with midnight_to_midnight(): cascading effects of how lubridate::ymd_hms() handles midnight.

[etzkorn]

Hi Marta,
Please see the following reproducible example highlighting some problems we are experiencing with the package. As the function midnight_to_midnight() produces some unexpected results, the rest of the package (e.x. activity_stats()) seems broken as well.

The error seems to originate with the lubridate function ymd_hms, which fails to parse date-times at midnight, though it does parse character vectors at midnight. A potential solution would be to convert the user input to a character vector before re-converting to date-time. I've also "hacked" another solution in which I add one second to each date time, but I am not yet sure if that works for higher-level arc tools functions like activity_stats().

devtools::install_github("martakarass/arctools")
#install.packages("arctools")
library(arctools)
library(lubridate)

# generate a time vector 6am to 6pm next day
start.time <- ymd_hms("2024-02-07 06:00:00 UTC")
acc_ts <- start.time + minutes(0:1439)
class(acc_ts)
# Result is "POSIXct" "POSIXt" 

acc <- rnorm(length(acc_ts))

ts_mtm  <- midnight_to_midnight(acc = acc_ts, acc_ts = acc_ts)
# Warning message:
# 1 failed to parse.

# check that the length is the desired output
length(acc_ts) #1440
length(ts_mtm) #4320 
# in addition to the parsing error, the mtm function adds 1 extra day (1440 minutes) 
# at the end

# the error appears to happen when midnight_to_midnight calls lubridate::ymd_hms

acc_ts2 <- lubridate::ymd_hms(acc_ts)

# print original time stamps around the problematic one
acc_ts[(which(is.na(acc_ts2))-2):(which(is.na(acc_ts2))+2)]

################################################################################
# a "hack" solution: adding 1 second to date times
acc_ts <- start.time + minutes(0:1439) + seconds(1)
ts_mtm  <- midnight_to_midnight(acc = acc_ts, acc_ts = acc_ts)
# no warning

# check that the length is the desired output
length(acc_ts) #1440
length(ts_mtm) #2880 

Some extra notes on ymd_hms:

The issue has been noted here: tidyverse/lubridate#1124.

Essentially, ymd_hms only "works" for times at midnight which are coded as a character vector, not a date-time. Here is an illustration:

ymd_hms("2024-02-08 00:00:00 UTC")
# "2024-02-08 UTC", works

ymd_hms(ymd_hms("2024-02-08 00:00:00 UTC"))
# fails

ymd_hms("2024-02-08 00:01:00 UTC")
# "2024-02-08 00:01:00 UTC", works

ymd_hms(ymd_hms("2024-02-08 00:01:00 UTC"))
# "2024-02-08 00:01:00 UTC", works