nanotime: Nanosecond-Resolution Time Objects for R

CI License CRAN r-universe Dependencies Downloads Code Coverage Last Commit Documentation

Motivation

R has excellent tools for dates and times. The Date and POSIXct classes (as well as the ‘wide’ representation in POSIXlt) are versatile, and a lot of useful tooling has been built around them.

However, POSIXct is implemented as a double with fractional seconds since the epoch. Given the 53 bits accuracy, it leaves just a bit less than microsecond resolution. Furthermore, using floating-point arithmetic for an integer concept opens the door to painful issues of error accumulation.

More and more performance measurements, latency statistics, etc., are now measured more finely, and we need nanosecond resolution for which commonly an integer64 is used to represent nanoseconds since the epoch.

And while R does not have a native type for this, the bit64 package by Jens Oehlschlägel offers a performant one implemented as a lightweight S3 class. So this package uses the integer64 class, along with multiple helper functions for parsing and formatting at nano-second resolution from the RcppCCTZ package which wraps the CCTZ library from Google. CCTZ is a modern C++11 library extending the (C++11-native) chrono type.

In addition to the point-in-time type nanotime, this package also provides an interval type nanoival which may have open or closed start/end, a period type nanoperiod that is a human representation of time, such as day, month, etc., and a duration type nanoduration. These types are similar to what the lubridate package proposes.

Set and arithmetic operations on these types are available. All functionality is designed to correctly handle instances across different time zones. Because these temporal types are based on R built-in types, most functions have an efficient implementation and the types are suitable for use in data.frame and data.table. nanotime is also a better choice than the native POSIXct in most of the cases where fractional seconds are needed because it avoids floating point issues.

Documentation

Package documentation, help pages, a vignette, and more is available here.

Demo

See the included demo script nanosecondDelayExample.R for a (completely simulated and hence made-up) study of network latency measured in nanoseconds resulting in the figure below

Examples

Simple Parsing and Arithmetic

R> x <- as.nanotime("1970-01-01T00:00:00.000000001+00:00")
R> x
[1] "1970-01-01T00:00:00.000000001+00:00"
R> x + 1e9
[1] "1970-01-01T00:00:01.000000001+00:00"
R> as.nanotime("2020-03-21 Europe/London")
[1] 2020-03-21T00:00:00+00:00

Vectorised

R> options("width"=60)
R> v <- nanotime(Sys.time()) + 1:5
R> v
[1] 2020-03-22T03:09:20.732122001+00:00
[2] 2020-03-22T03:09:20.732122002+00:00
[3] 2020-03-22T03:09:20.732122003+00:00
[4] 2020-03-22T03:09:20.732122004+00:00
[5] 2020-03-22T03:09:20.732122005+00:00
R>

Use with zoo

R> library(zoo)
R> z <- zoo(cbind(A=1:5, B=5:1), v)
R> options(nanotimeFormat="%H:%M:%E*S")  ## override default format
R> z
R> options(nanotimeFormat=NULL)  ## go back to default format
R> z

Use with data.table

R> library(data.table)
R> dt <- data.table(v, cbind(A=1:5, B=5:1))
R> fwrite(dt, file="datatableTest.csv")  # write out
R> dtcheck <- fread("datatableTest.csv") # read back
R> dtcheck
R> dtcheck[, v:=nanotime(v)]             # read as a string, need to re-class as nanotime
R> fread("../datatableTest.csv", colClasses=c("nanotime", "integer", "integer"))

Use with data.frame

This requires version 0.0.2 or later.

R> data.frame(cbind(A=1:5, B=5:1), v=v)

Intervals

R> ival <- as.nanoival("+2009-01-01 13:12:00 America/New_York -> 2009-02-01 15:11:03 America/New_York-")
R> ival
[1] +2009-01-01T18:12:00+00:00 -> 2009-02-01T20:11:03+00:00-

R> start <- nanotime("2009-01-01 13:12:00 America/New_York")
R> end   <- nanotime("2009-02-01 15:11:00 America/New_York")
R> nanoival(start, end)                   # by default sopen=F,eopen=T
[1] +2009-01-01T18:12:00+00:00 -> 2009-02-01T20:11:00+00:00-
R> nanoival(start, end, sopen=FALSE, eopen=TRUE)
[1] +2009-01-01T18:12:00+00:00 -> 2009-02-01T20:11:00+00:00-

R> intersect(as.nanoival("+2019-03-01 UTC -> 2020-03-01 UTC-"),
             as.nanoival("+2020-01-01 UTC -> 2020-06-01 UTC-"))
[1] +2020-01-01T00:00:00+00:00 -> 2020-03-01T00:00:00+00:00-

R> union(as.nanoival("+2019-03-01 UTC -> 2020-03-01 UTC-"),
         as.nanoival("+2020-01-01 UTC -> 2020-06-01 UTC-"))
[1] +2019-03-01T00:00:00+00:00 -> 2020-06-01T00:00:00+00:00-

R> setdiff(as.nanoival("+2019-03-01 UTC -> 2020-03-01 UTC-"),
           as.nanoival("+2020-01-01 UTC -> 2020-06-01 UTC-"))
[1] +2019-03-01T00:00:00+00:00 -> 2020-01-01T00:00:00+00:00-

Periods

R> as.nanoperiod("1y1m1w1d/01:01:01.000_000_001")
[1] 13m8d/01:01:01.000_000_001
R> nanoperiod(months=13, days=-1, duration="01:00:00")
[1] 13m-1d/01:00:00

R> ones <- as.nanoperiod("1y1m1w1d/01:01:01.000_000_001")
R> nanoperiod.month(ones); nanoperiod.day(ones); nanoperiod.nanoduration(ones)
[1] 13
[1] 8
[1] 01:01:01.000_000_001

R> plus(v, as.nanoperiod("1y1m"), tz="UTC")
[1] 2021-04-22T03:09:20.732122001+00:00
[2] 2021-04-22T03:09:20.732122002+00:00
[3] 2021-04-22T03:09:20.732122003+00:00
[4] 2021-04-22T03:09:20.732122004+00:00
[5] 2021-04-22T03:09:20.732122005+00:00

Durations

R> nanoduration(hours=1, minutes=1, seconds=1, nanoseconds=1)
R> as.nanoduration("00:00:01")
R> as.nanoduration("-00:00:01")
R> as.nanoduration("100:00:00")
R> as.nanoduration("00:00:00.000_000_001")

Sequences

R> from <- as.nanotime("2018-09-14T12:44:00+00:00")
R> seq(from, by=as.nanoperiod("1y"), length.out=4, tz="Europe/London")
[1] 2018-09-14T12:44:00+00:00
[2] 2019-09-14T12:44:00+00:00
[3] 2020-09-14T12:44:00+00:00
[4] 2021-09-14T12:44:00+00:00

Technical Details

The bit64 package (by Jens Oehlschlägel) supplies the integer64 type used to store the nanosecond resolution time as (positive or negative) offsets to the epoch of January 1, 1970. The RcppCCTZ package supplies the formatting and parsing routines based on the (modern C++) library CCTZ from Google, when the parsing cannot be done using a fast built-in parser. integer64 is also used for the type nanoduration, whereas nanoival and nanoperiod are stored in a complex, i.e. over 128 bits.

Status

The package is by now fairly mature, has been rewritten once (to go from S3 to S4) and has recently received a sizeable feature extension. There may still be changes, though there should generally never be breaking ones. The package also has an extensive test suite, and very good code coverage.

See the issue tickets for an up to date list of potentially desirable, possibly planned, or at least discussed items.

Installation

The package is on CRAN and can be installed via a standard

install.packages("nanotime")

whereas in order to install development versions a

remotes::install_github("eddelbuettel/nanotime")  # dev version

should suffice.

Authors

Dirk Eddelbuettel and Leonardo Silvestri

License

GPL (>= 2)