Stamped date format using lubridate is selecting the wrong fixed format

[ekstroem]

I am trying to format date output using stamp from the lubridate package. I would like the end format to be similar to

Sunday, November 1, 23:15

My problem is getting the unabbreviated month to be printed

library(lubridate)
x <- ymd_hm("2017-11-20 15:15")

Now if I use stamp as below then I almost get the right output

stamp("Sunday, November 30, 23:15")(x)
Multiple formats matched: "%A, %B %d, %y:%H"(1), "%A, %B %y, %d:%H"(1), "Sunday, %B %d, %y:%H"(1), "Sunday, %B %y, %d:%H"(1)
Using: "%A, %B %d, %y:%H"
[1] "Monday, November 20, 17:15"

However, November is interpreted as fixed text and the guessed format uses the year and hour in place of the hour and minutes. If I add the orders argument then I can try to force the order of the inputs

stamp("Sunday, Nov 30, 23:15", orders="AbdHM")(x)
Multiple formats matched: "%A, %Om %d, %H:%M"(0), "%A, %b %d, %H:%M"(1)
Using: "%A, %Om %d, %H:%M"
[1] "Monday, 11 20, 15:15"

Here the output is correct except I get the month as a number and not as text. Note that stamp does provide the correct format but ends up using the not-quite-correct-format despite being given an identical match.

What can I do to force stamp to use the exact format that I supplied?

[There are other options than using stamp. I just cannot figure out why the code above isn't working]

[cderv]

Here is my understanding of the situation base on some small test.
I use last dev version of lubridate

library(lubridate)
#> 
#> Attachement du package : 'lubridate'
#> The following object is masked from 'package:base':
#> 
#>     date
devtools::session_info("lubridate")$packages
#>  package   * version    date       source                           
#>  lubridate * 1.6.0.9009 2017-05-22 Github (hadley/lubridate@2608730)
#>  magrittr    1.5        2014-11-22 CRAN (R 3.3.2)                   
#>  Rcpp        0.12.10    2017-03-19 CRAN (R 3.3.3)                   
#>  stringi     1.1.5      2017-04-07 CRAN (R 3.3.3)                   
#>  stringr     1.2.0      2017-02-18 CRAN (R 3.3.2)

I choose another month than the stamp one. here October and save my locale LC_TIME setting for later

x <- ymd_hm("2017-10-20 15:15")
(ori <- Sys.getlocale("LC_TIME"))
#> [1] "French_France.1252"

Results is depending on my locale setting

# In French as my locale
stamp("Dimanche, Novembre 30, 23:15")(x)
#> Multiple formats matched: "%A, %B %d, %y:%H"(1), "%A, %B %y, %d:%H"(1), "Dimanche, %B %d, %y:%H"(1), "Dimanche, %B %y, %d:%H"(1)
#> Using: "%A, %B %d, %y:%H"
#> [1] "vendredi, octobre 20, 17:15"
# In English, different from my locale
stamp("Sunday, November 30, 23:15")(x)
#> Error in stamp("Sunday, November 30, 23:15"): Couldn't guess formats of: Sunday, November 30, 23:15
# Setting stamp with local English
stamp("Sunday, November 30, 23:15", locale = "English")(x)
#> Multiple formats matched: "%A, %B %d, %y:%H"(0), "%A, %B %y, %d:%H"(0), "Sunday, %B %d, %y:%H"(0), "Sunday, %B %y, %d:%H"(0)
#> Using: "%A, %B %d, %y:%H"
#> [1] "vendredi, octobre 20, 17:15"
# Changing completely my Locale
Sys.setlocale("LC_TIME", "English")
#> [1] "English_United States.1252"
stamp("Sunday, November 30, 23:15")(x)
#> Multiple formats matched: "%A, %B %d, %y:%H"(1), "%A, %B %y, %d:%H"(1), "Sunday, %B %d, %y:%H"(1), "Sunday, %B %y, %d:%H"(1)
#> Using: "%A, %B %d, %y:%H"
#> [1] "Friday, October 20, 17:15"

What is your local setting ? Could it be the source of the problem ?

---

Regarding the second problem about forcing orders in stamp, here's a few lines of test to understand

# Getting my original locale back
Sys.setlocale("LC_TIME", ori)
#> [1] "French_France.1252"
# Do not work as stamp can't guess format
stamp("Sunday, Nov 30, 23:15", orders="AbdHM")(x)
#> Error in stamp("Sunday, Nov 30, 23:15", orders = "AbdHM"): Couldn't guess formats of: Sunday, Nov 30, 23:15
# Stamp guess format but select the wrong format `%Om` giving us back a month in number
stamp("Sunday, Nov 30, 23:15", orders="AbdHM", locale = "English")(x)
#> Multiple formats matched: "%A, %Om %d, %H:%M"(0), "%A, %b %d, %H:%M"(0)
#> Using: "%A, %Om %d, %H:%M"
#> [1] "vendredi, 10 20, 15:15"
# Changing my local gives us the desired result
Sys.setlocale("LC_TIME", "English")
#> [1] "English_United States.1252"
stamp("Sunday, Nov 30, 23:15", orders="AbdHM")(x)
#> Multiple formats matched: "%A, %b %d, %H:%M"(1), "%A, %Om %d, %H:%M"(0)
#> Using: "%A, %b %d, %H:%M"
#> [1] "Friday, Oct 20, 15:15"

This behaviour comes from the trained part of stamp function. Internal .trained_format function is not impacted by local argument in stamp. Formats are trained with the system LC_TIME local, it is why you get this behaviour.

Maybe we could change this behaviour as lubridate:::.trained_format have an unused local argument. Could be better if training was coherent with the date provided in stamp

[vspinu]

This is a bug and it comes from the fact that fromat doesn't accept locale argument. We need to wrap it into system call to set the locale before the format.

[vspinu]

Should be working now:

> Sys.setlocale("LC_TIME", "it_IT.utf8")
[1] "it_IT.utf8"
> formater <- stamp("Sunday, November 30, 23:15", locale = "en_DK.utf8")
Multiple formats matched: "%A, %B %d, %y:%H"(0), "%A, %B %y, %d:%H"(0), "Sunday, %B %d, %y:%H"(0), "Sunday, %B %y, %d:%H"(0)
Using: "%A, %B %d, %y:%H"
> x <- ymd_hm(c("2017-01-20 15:15", "2017-02-10 10:10"))
> formater(x)
[1] "Friday, January 20, 17:15"  "Friday, February 10, 17:10"

Thanks for investigating this.
Could you please check again with your data?

[vspinu]

Note that only the locale issue is fixed. There is no easy way to fix the the misinterpretation of %H:%M. You need to supply your own orders which currently don't quite work as you reported:

> formater <- stamp("Sunday, November 30, 23:15", "ABdHM", locale = "en_DK.utf8")
Multiple formats matched: "%A, %Om %d, %H:%M"(0), "%A, %B %d, %H:%M"(0)
Using: "%A, %Om %d, %H:%M"
> x <- ymd_hm(c("2017-01-20 15:15", "2017-02-11 10:10"))
> formater(x)
[1] "Friday, 01 20, 15:15"   "Saturday, 02 11, 10:10"

[ekstroem]

Ah good (in the sense that it wasn't just me being baffled)!

With the updated version of lubridate I now get (my locale is C, and I'm running OSX).

stamp("Sunday, November 30, 23:15")(x)
Multiple formats matched: "%A, %B %d, %y:%H"(1), "%A, %B %y, %d:%H"(1), "Sunday, %B %d, %y:%H"(1), "Sunday, %B %y, %d:%H"(1)
Using: "%A, %B %d, %y:%H"
[1] "Monday, November 20, 17:15"

Date is okay, the time is due to year and hour just as @vspinu mentioned. If I run the shorter version of the months, %b, I get the following

stamp("Sunday, Nov 30, 23:15", orders="AbdHM")(x)
Multiple formats matched: "%A, %b %d, %H:%M"(1), "%A, %Om %d, %H:%M"(0)
Using: "%A, %b %d, %H:%M"
[1] "Monday, Nov 20, 15:15"

where all is good. Following my initial approach I get something that is almost right using B instead of b.

stamp("Sunday, Nov 30, 23:15", orders="ABdHM")(x)
Using: "%A, %Om %d, %H:%M"
[1] "Monday, 11 20, 15:15"

This is quite surprising (in a positive way) since I'm specifying an order that doesn't match the format and it still guesses the month although in numbers.

If I force the time zone and run

stamp("Sunday, November 1, 16:15", orders="ABdHM")(force_tz(x,  "Europe/Copenhagen"))
Multiple formats matched: "%A, %B %d, %H:%M"(1), "%A, %Om %d, %H:%M"(0)
Using: "%A, %B %d, %H:%M"
[1] "Monday, November 20, 15:15"

then I get the desired output format. Also works with

stamp("Sunday, November 30, 23:15", "ABdHM", locale = "en_DK.utf8")

as mentioned above.

Cheers!

[vspinu]

This is quite surprising (in a positive way)

Well. This is a bug which is now fixed. It works now as expected with no need to reset the global locale:

> Sys.setlocale("LC_TIME", "it_IT.utf8")
[1] "it_IT.utf8"
> 
> formater <- stamp("Sunday, November 30, 23:15", "ABdHM", locale = "C")
Multiple formats matched: "%A, %B %d, %H:%M"(1), "%A, %Om %d, %H:%M"(0)
Using: "%A, %B %d, %H:%M"
> x <- ymd_hm(c("2017-01-20 15:15", "2017-02-11 10:10"))
> formater(x)
[1] "Friday, January 20, 15:15"    "Saturday, February 11, 10:10"

The locale issue wasn't stamp specific, so yes, your report revealed a bug in parsing as well. Thanks.