seq.Date giving problems - r

Question

I have recently updated R to version 3.2.3 and now I found a problem using seq with dates:
date1<-as.POSIXct("2014-01-30 02:00:00")
date2<-as.POSIXct("2014-12-24 11:00:00")
seq(date1,date2,by="month")
#[1] "2014-01-30 02:00:00 CET" "2014-03-02 02:00:00 CET"
#[3] NA "2014-04-30 02:00:00 CEST"
#[5] "2014-05-30 02:00:00 CEST" "2014-06-30 02:00:00 CEST"
#[7] "2014-07-30 02:00:00 CEST" "2014-08-30 02:00:00 CEST"
#[9] "2014-09-30 02:00:00 CEST" "2014-10-30 02:00:00 CET"
#[11] "2014-11-30 02:00:00 CET"
I don't understand where the NA comes from. I have tried on different machines with both the same R version as mine or a previous one and in the place of that NA they correctly give "2014-03-30". Furthermore, if I change the year in the dates from 2014 to 2015, no NAs are returned!
I guess that during the installation something in my locale was modified but I cannot understand how to fix the problem.
Sys.getlocale() returns:
"en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8"
and my system is a Mac Book Pro with Maverick.
Thanks for any help!

Answer 1

I was guessing Germany and here's what the situation was in that CET timezone on Mar 30 (a Sunday)
http://www.timetemperature.com/utc-converter/utc-20140330-germany-12.html
UTC or GMT Time Germany
Sunday 30th March 2014 00:00:00 Sun 01:00 AM
Sunday 30th March 2014 01:00:00 Sun 03:00 AM*
Changing the setting to Italy, I get the same result:
UTC or GMT Time Italy
Sunday 30th March 2014 00:00:00 Sun 01:00 AM
Sunday 30th March 2014 01:00:00 Sun 03:00 AM*
The key here is to be suspicious of weirdness when the time is in the early morning hours of a Spring or Fall date, or when calculations of intervals crosses such dates. The rules change from year to year, and since countries often do the switch on a Sunday or Saturday morning, the exact dates jump around.
The changes vary by country and in the US they may vary by state or even "sub-state" boundaries: In Washington State in 2014 you find the change happening onhte second Sunday of March:
http://www.timetemperature.com/utc-converter/utc-20140309-us-washington+state-12.html
UTC or GMT Time US-Washington State
snipped several times
Sunday 9th March 2014 07:00:00 Sat 11:00 PM
Sunday 9th March 2014 08:00:00 Sun 12:00 AM
Sunday 9th March 2014 09:00:00 Sun 01:00 AM
Sunday 9th March 2014 10:00:00 Sun 03:00 AM*
Sunday 9th March 2014 11:00:00 Sun 04:00 AM*
I'm in the same TZ as Washington state. With a Sys.timezone set, one can reproduce the NA, at least on a Mac. The implementation of times and timezones is OS-specific, so it's possible to see variations in how these weirdities get visible:
> Sys.timezone(location = TRUE)
[1] "America/Los_Angeles"
> date1<-as.POSIXct("2014-01-09 02:00:00")
> date2<-as.POSIXct("2014-12-09 11:00:00")
> seq(date1,date2,by="month")
[1] "2014-01-09 02:00:00 PST" "2014-02-09 02:00:00 PST"
[3] NA "2014-04-09 02:00:00 PDT"
[5] "2014-05-09 02:00:00 PDT" "2014-06-09 02:00:00 PDT"
[7] "2014-07-09 02:00:00 PDT" "2014-08-09 02:00:00 PDT"
[9] "2014-09-09 02:00:00 PDT" "2014-10-09 02:00:00 PDT"
[11] "2014-11-09 02:00:00 PST" "2014-12-09 02:00:00 PST"

Answer 2

By inspecting the relevant code in seq.POSIXt there appears that a call to seq with by="month" works as follows
[some manipulation of the data]
conversion of data1 & data2 to POSIXlt
creation of a sequence of months numbers spanning the interval from data1 to data2 (in this case 0,...,11)
manual update of data1$mon to this sequence of months (and up to this point the dates are all properly handled)
finally, the resulting dates are converted to POSIXct and here the NA shows up
while the resulting NA is technically correct, since it is trying to convert an invalid date ("2014-01-30 02:00:00 CET", which does not exist) to POSIXct, could the issue be possibly worked around by passing through difftimes? [*]
not sure it is worth, though...
[*] here by difftimes I mean to add the correct number of seconds to the dates instead of just adding the months...