Adding 10 seconds to any date with 20:59:50 results in NA

[mdagost]

I encountered something odd today, which looks to me like a bug. I had been running R 2.15.2 with lubridate_1.2.0, I can do the following normal thing:

ymd_hms('2013-05-01 20:59:50')+seconds(10)
[1] "2013-05-01 21:00:00 UTC"

However, with R 3.0.0 and lubridate_1.3.0, I get the following:

ymd_hms('2013-05-01 20:59:50')+seconds(10)
[1] NA

I get an NA for any date that I try as long as the timestamp is 20:59:50 and I add seconds(10). Actually, it looks like I get an NA for any timestamp (like 20:59:30) and interval combo in seconds (like seconds(30)) that would have a resulting timestamp of 21:00:00. If the interval is in minutes or hours, it works though:

ymd_hms('2013-05-01 20:00:00')+hours(1)
[1] "2013-05-01 21:00:00 UTC"
ymd_hms('2013-05-01 20:50:00')+minutes(10)
[1] "2013-05-01 21:00:00 UTC"

Any ideas?

Michelangelo

[mdagost]

So I've done a little digging. The bug persists in the latest R and the lubridate master.

It looks to me like the bug is coming from this piece of code:

https://github.com/hadley/lubridate/blob/master/R/update.r#L141-L148

For the first example above, chours=21 and lhours=20. That causes this to run:

ct[chours != lhours] <- NA

If I comment that line out, the first example works with the latest version of R and the lubridate master.

Looks like that code was added in with this commit last month by @garrettgman to fix a DST issue:

8056471

I don't know enough about what the DST code is doing, but I think I've at least localized the bug to that bit.

[garrettgman]

Looks like a bug. Thanks.

[vspinu]

Finally fixed in 7941df8. This report was duplicated in #217 and #313.