Dates
API reference
- Dates and Time Types
- Dates Functions
  - Accessor Functions

Dates

The Dates module provides two types for working with dates: Date and DateTime, representing day and millisecond precision, respectively; both are subtypes of the abstract TimeType. The motivation for distinct types is simple: some operations are much simpler, both in terms of code and mental reasoning, when the complexities of greater precision don’t have to be dealt with. For example, since the Date type only resolves to the precision of a single date (i.e. no hours, minutes, or seconds), normal considerations for time zones, daylight savings/summer time, and leap seconds are unnecessary and avoided.

Both Date and DateTime are basically immutable Int64 wrappers. The single instant field of either type is actually a UTInstant{P} type, which represents a continuously increasing machine timeline based on the UT second [1]. The DateTime type is not aware of time zones (naive, in Python parlance), analogous to a LocalDateTime in Java 8. Additional time zone functionality can be added through the TimeZones.jl package, which compiles the IANA time zone database. Both Date and DateTime are based on the ISO 8601 standard, which follows the proleptic Gregorian calendar. One note is that the ISO 8601 standard is particular about BC/BCE dates. In general, the last day of the BC/BCE era, 1-12-31 BC/BCE, was followed by 1-1-1 AD/CE, thus no year zero exists. The ISO standard, however, states that 1 BC/BCE is year zero, so 0000-12-31 is the day before 0001-01-01, and year -0001 (yes, negative one for the year) is 2 BC/BCE, year -0002 is 3 BC/BCE, etc.

Constructors

Date and DateTime types can be constructed by integer or Period types, by parsing, or through adjusters (more on those later):

1. julia> DateTime(2013)
2. 2013-01-01T00:00:00

4. julia> DateTime(2013,7)
5. 2013-07-01T00:00:00

7. julia> DateTime(2013,7,1)
8. 2013-07-01T00:00:00

10. julia> DateTime(2013,7,1,12)
11. 2013-07-01T12:00:00

13. julia> DateTime(2013,7,1,12,30)
14. 2013-07-01T12:30:00

16. julia> DateTime(2013,7,1,12,30,59)
17. 2013-07-01T12:30:59

19. julia> DateTime(2013,7,1,12,30,59,1)
20. 2013-07-01T12:30:59.001

22. julia> Date(2013)
23. 2013-01-01

25. julia> Date(2013,7)
26. 2013-07-01

28. julia> Date(2013,7,1)
29. 2013-07-01

31. julia> Date(Dates.Year(2013),Dates.Month(7),Dates.Day(1))
32. 2013-07-01

34. julia> Date(Dates.Month(7),Dates.Year(2013))
35. 2013-07-01

Date or DateTime parsing is accomplished by the use of format strings. Format strings work by the notion of defining delimited or fixed-width “slots” that contain a period to parse and passing the text to parse and format string to a Date or DateTime constructor, of the form Date("2015-01-01","y-m-d") or DateTime("20150101","yyyymmdd").

Delimited slots are marked by specifying the delimiter the parser should expect between two subsequent periods; so "y-m-d" lets the parser know that between the first and second slots in a date string like "2014-07-16", it should find the - character. The y, m, and d characters let the parser know which periods to parse in each slot.

Fixed-width slots are specified by repeating the period character the number of times corresponding to the width with no delimiter between characters. So "yyyymmdd" would correspond to a date string like "20140716". The parser distinguishes a fixed-width slot by the absence of a delimiter, noting the transition "yyyymm" from one period character to the next.

Support for text-form month parsing is also supported through the u and U characters, for abbreviated and full-length month names, respectively. By default, only English month names are supported, so u corresponds to “Jan”, “Feb”, “Mar”, etc. And U corresponds to “January”, “February”, “March”, etc. Similar to other name=>value mapping functions dayname and monthname, custom locales can be loaded by passing in the locale=>Dict{String,Int} mapping to the MONTHTOVALUEABBR and MONTHTOVALUE dicts for abbreviated and full-name month names, respectively.

One note on parsing performance: using the Date(date_string,format_string) function is fine if only called a few times. If there are many similarly formatted date strings to parse however, it is much more efficient to first create a Dates.DateFormat, and pass it instead of a raw format string.

1. julia> df = DateFormat("y-m-d");

3. julia> dt = Date("2015-01-01",df)
4. 2015-01-01

6. julia> dt2 = Date("2015-01-02",df)
7. 2015-01-02

You can also use the dateformat"" string macro. This macro creates the DateFormat object once when the macro is expanded and uses the same DateFormat object even if a code snippet is run multiple times.

1. julia> for i = 1:10^5
2. Date("2015-01-01", dateformat"y-m-d")
3. end

A full suite of parsing and formatting tests and examples is available in stdlib/Dates/test/io.jl.

Durations/Comparisons

Finding the length of time between two Date or DateTime is straightforward given their underlying representation as UTInstant{Day} and UTInstant{Millisecond}, respectively. The difference between Date is returned in the number of Day, and DateTime in the number of Millisecond. Similarly, comparing TimeType is a simple matter of comparing the underlying machine instants (which in turn compares the internal Int64 values).

1. julia> dt = Date(2012,2,29)
2. 2012-02-29

4. julia> dt2 = Date(2000,2,1)
5. 2000-02-01

7. julia> dump(dt)
8. Date
9. instant: Dates.UTInstant{Day}
10. periods: Day
11. value: Int64 734562

13. julia> dump(dt2)
14. Date
15. instant: Dates.UTInstant{Day}
16. periods: Day
17. value: Int64 730151

19. julia> dt > dt2
20. true

22. julia> dt != dt2
23. true

25. julia> dt + dt2
26. ERROR: MethodError: no method matching +(::Date, ::Date)
27. [...]

29. julia> dt * dt2
30. ERROR: MethodError: no method matching *(::Date, ::Date)
31. [...]

33. julia> dt / dt2
34. ERROR: MethodError: no method matching /(::Date, ::Date)

36. julia> dt - dt2
37. 4411 days

39. julia> dt2 - dt
40. -4411 days

42. julia> dt = DateTime(2012,2,29)
43. 2012-02-29T00:00:00

45. julia> dt2 = DateTime(2000,2,1)
46. 2000-02-01T00:00:00

48. julia> dt - dt2
49. 381110400000 milliseconds

Accessor Functions

Because the Date and DateTime types are stored as single Int64 values, date parts or fields can be retrieved through accessor functions. The lowercase accessors return the field as an integer:

1. julia> t = Date(2014, 1, 31)
2. 2014-01-31

4. julia> Dates.year(t)
5. 2014

7. julia> Dates.month(t)
8. 1

10. julia> Dates.week(t)
11. 5

13. julia> Dates.day(t)
14. 31

While propercase return the same value in the corresponding Period type:

1. julia> Dates.Year(t)
2. 2014 years

4. julia> Dates.Day(t)
5. 31 days

Compound methods are provided because it is more efficient to access multiple fields at the same time than individually:

1. julia> Dates.yearmonth(t)
2. (2014, 1)

4. julia> Dates.monthday(t)
5. (1, 31)

7. julia> Dates.yearmonthday(t)
8. (2014, 1, 31)

One may also access the underlying UTInstant or integer value:

1. julia> dump(t)
2. Date
3. instant: Dates.UTInstant{Day}
4. periods: Day
5. value: Int64 735264

7. julia> t.instant
8. Dates.UTInstant{Day}(Day(735264))

10. julia> Dates.value(t)
11. 735264

Query Functions

Query functions provide calendrical information about a TimeType. They include information about the day of the week:

1. julia> t = Date(2014, 1, 31)
2. 2014-01-31

4. julia> Dates.dayofweek(t)
5. 5

7. julia> Dates.dayname(t)
8. "Friday"

10. julia> Dates.dayofweekofmonth(t) # 5th Friday of January
11. 5

Month of the year:

1. julia> Dates.monthname(t)
2. "January"

4. julia> Dates.daysinmonth(t)
5. 31

As well as information about the TimeType‘s year and quarter:

1. julia> Dates.isleapyear(t)
2. false

4. julia> Dates.dayofyear(t)
5. 31

7. julia> Dates.quarterofyear(t)
8. 1

10. julia> Dates.dayofquarter(t)
11. 31

The dayname and monthname methods can also take an optional locale keyword that can be used to return the name of the day or month of the year for other languages/locales. There are also versions of these functions returning the abbreviated names, namely dayabbr and monthabbr. First the mapping is loaded into the LOCALES variable:

1. julia> french_months = ["janvier", "février", "mars", "avril", "mai", "juin",
2. "juillet", "août", "septembre", "octobre", "novembre", "décembre"];

4. julia> french_monts_abbrev = ["janv","févr","mars","avril","mai","juin",
5. "juil","août","sept","oct","nov","déc"];

7. julia> french_days = ["lundi","mardi","mercredi","jeudi","vendredi","samedi","dimanche"];

9. julia> Dates.LOCALES["french"] = Dates.DateLocale(french_months, french_monts_abbrev, french_days, [""]);

The above mentioned functions can then be used to perform the queries:

1. julia> Dates.dayname(t;locale="french")
2. "vendredi"

4. julia> Dates.monthname(t;locale="french")
5. "janvier"

7. julia> Dates.monthabbr(t;locale="french")
8. "janv"

Since the abbreviated versions of the days are not loaded, trying to use the function dayabbr will error.

1. julia> Dates.dayabbr(t;locale="french")
2. ERROR: BoundsError: attempt to access 1-element Vector{String} at index [5]
3. Stacktrace:
4. [...]

TimeType-Period Arithmetic

It’s good practice when using any language/date framework to be familiar with how date-period arithmetic is handled as there are some tricky issues to deal with (though much less so for day-precision types).

The Dates module approach tries to follow the simple principle of trying to change as little as possible when doing Period arithmetic. This approach is also often known as calendrical arithmetic or what you would probably guess if someone were to ask you the same calculation in a conversation. Why all the fuss about this? Let’s take a classic example: add 1 month to January 31st, 2014. What’s the answer? Javascript will say March 3 (assumes 31 days). PHP says March 2 (assumes 30 days). The fact is, there is no right answer. In the Dates module, it gives the result of February 28th. How does it figure that out? Consider the classic 7-7-7 gambling game in casinos.

Now just imagine that instead of 7-7-7, the slots are Year-Month-Day, or in our example, 2014-01-31. When you ask to add 1 month to this date, the month slot is incremented, so now we have 2014-02-31. Then the day number is checked if it is greater than the last valid day of the new month; if it is (as in the case above), the day number is adjusted down to the last valid day (28). What are the ramifications with this approach? Go ahead and add another month to our date, 2014-02-28 + Month(1) == 2014-03-28. What? Were you expecting the last day of March? Nope, sorry, remember the 7-7-7 slots. As few slots as possible are going to change, so we first increment the month slot by 1, 2014-03-28, and boom, we’re done because that’s a valid date. On the other hand, if we were to add 2 months to our original date, 2014-01-31, then we end up with 2014-03-31, as expected. The other ramification of this approach is a loss in associativity when a specific ordering is forced (i.e. adding things in different orders results in different outcomes). For example:

1. julia> (Date(2014,1,29)+Dates.Day(1)) + Dates.Month(1)
2. 2014-02-28

4. julia> (Date(2014,1,29)+Dates.Month(1)) + Dates.Day(1)
5. 2014-03-01

What’s going on there? In the first line, we’re adding 1 day to January 29th, which results in 2014-01-30; then we add 1 month, so we get 2014-02-30, which then adjusts down to 2014-02-28. In the second example, we add 1 month first, where we get 2014-02-29, which adjusts down to 2014-02-28, and then add 1 day, which results in 2014-03-01. One design principle that helps in this case is that, in the presence of multiple Periods, the operations will be ordered by the Periods’ types, not their value or positional order; this means Year will always be added first, then Month, then Week, etc. Hence the following does result in associativity and Just Works:

1. julia> Date(2014,1,29) + Dates.Day(1) + Dates.Month(1)
2. 2014-03-01

4. julia> Date(2014,1,29) + Dates.Month(1) + Dates.Day(1)
5. 2014-03-01

Tricky? Perhaps. What is an innocent Dates user to do? The bottom line is to be aware that explicitly forcing a certain associativity, when dealing with months, may lead to some unexpected results, but otherwise, everything should work as expected. Thankfully, that’s pretty much the extent of the odd cases in date-period arithmetic when dealing with time in UT (avoiding the “joys” of dealing with daylight savings, leap seconds, etc.).

As a bonus, all period arithmetic objects work directly with ranges:

1. julia> dr = Date(2014,1,29):Day(1):Date(2014,2,3)
2. Date("2014-01-29"):Day(1):Date("2014-02-03")

4. julia> collect(dr)
5. 6-element Vector{Date}:
6. 2014-01-29
7. 2014-01-30
8. 2014-01-31
9. 2014-02-01
10. 2014-02-02
11. 2014-02-03

13. julia> dr = Date(2014,1,29):Dates.Month(1):Date(2014,07,29)
14. Date("2014-01-29"):Month(1):Date("2014-07-29")

16. julia> collect(dr)
17. 7-element Vector{Date}:
18. 2014-01-29
19. 2014-02-28
20. 2014-03-29
21. 2014-04-29
22. 2014-05-29
23. 2014-06-29
24. 2014-07-29

Adjuster Functions

As convenient as date-period arithmetic is, often the kinds of calculations needed on dates take on a calendrical or temporal nature rather than a fixed number of periods. Holidays are a perfect example; most follow rules such as “Memorial Day = Last Monday of May”, or “Thanksgiving = 4th Thursday of November”. These kinds of temporal expressions deal with rules relative to the calendar, like first or last of the month, next Tuesday, or the first and third Wednesdays, etc.

The Dates module provides the adjuster API through several convenient methods that aid in simply and succinctly expressing temporal rules. The first group of adjuster methods deal with the first and last of weeks, months, quarters, and years. They each take a single TimeType as input and return or adjust to the first or last of the desired period relative to the input.

1. julia> Dates.firstdayofweek(Date(2014,7,16)) # Adjusts the input to the Monday of the input's week
2. 2014-07-14

4. julia> Dates.lastdayofmonth(Date(2014,7,16)) # Adjusts to the last day of the input's month
5. 2014-07-31

7. julia> Dates.lastdayofquarter(Date(2014,7,16)) # Adjusts to the last day of the input's quarter
8. 2014-09-30

The next two higher-order methods, tonext, and toprev, generalize working with temporal expressions by taking a DateFunction as first argument, along with a starting TimeType. A DateFunction is just a function, usually anonymous, that takes a single TimeType as input and returns a Bool, true indicating a satisfied adjustment criterion. For example:

1. julia> istuesday = x->Dates.dayofweek(x) == Dates.Tuesday; # Returns true if the day of the week of x is Tuesday

3. julia> Dates.tonext(istuesday, Date(2014,7,13)) # 2014-07-13 is a Sunday
4. 2014-07-15

6. julia> Dates.tonext(Date(2014,7,13), Dates.Tuesday) # Convenience method provided for day of the week adjustments
7. 2014-07-15

This is useful with the do-block syntax for more complex temporal expressions:

1. julia> Dates.tonext(Date(2014,7,13)) do x
2. # Return true on the 4th Thursday of November (Thanksgiving)
3. Dates.dayofweek(x) == Dates.Thursday &&
4. Dates.dayofweekofmonth(x) == 4 &&
5. Dates.month(x) == Dates.November
6. end
7. 2014-11-27

The Base.filter method can be used to obtain all valid dates/moments in a specified range:

1. # Pittsburgh street cleaning; Every 2nd Tuesday from April to November
2. # Date range from January 1st, 2014 to January 1st, 2015
3. julia> dr = Dates.Date(2014):Day(1):Dates.Date(2015);

5. julia> filter(dr) do x
6. Dates.dayofweek(x) == Dates.Tue &&
7. Dates.April <= Dates.month(x) <= Dates.Nov &&
8. Dates.dayofweekofmonth(x) == 2
9. end
10. 8-element Vector{Date}:
11. 2014-04-08
12. 2014-05-13
13. 2014-06-10
14. 2014-07-08
15. 2014-08-12
16. 2014-09-09
17. 2014-10-14
18. 2014-11-11

Additional examples and tests are available in stdlib/Dates/test/adjusters.jl.

Period Types

Periods are a human view of discrete, sometimes irregular durations of time. Consider 1 month; it could represent, in days, a value of 28, 29, 30, or 31 depending on the year and month context. Or a year could represent 365 or 366 days in the case of a leap year. Period types are simple Int64 wrappers and are constructed by wrapping any Int64 convertible type, i.e. Year(1) or Month(3.0). Arithmetic between Period of the same type behave like integers, and limited Period-Real arithmetic is available. You can extract the underlying integer with Dates.value.

1. julia> y1 = Dates.Year(1)
2. 1 year

4. julia> y2 = Dates.Year(2)
5. 2 years

7. julia> y3 = Dates.Year(10)
8. 10 years

10. julia> y1 + y2
11. 3 years

13. julia> div(y3,y2)
14. 5

16. julia> y3 - y2
17. 8 years

19. julia> y3 % y2
20. 0 years

22. julia> div(y3,3) # mirrors integer division
23. 3 years

25. julia> Dates.value(Dates.Millisecond(10))
26. 10

Rounding

Date and DateTime values can be rounded to a specified resolution (e.g., 1 month or 15 minutes) with floor, ceil, or round:

1. julia> floor(Date(1985, 8, 16), Dates.Month)
2. 1985-08-01

4. julia> ceil(DateTime(2013, 2, 13, 0, 31, 20), Dates.Minute(15))
5. 2013-02-13T00:45:00

7. julia> round(DateTime(2016, 8, 6, 20, 15), Dates.Day)
8. 2016-08-07T00:00:00

Unlike the numeric round method, which breaks ties toward the even number by default, the TimeTyperound method uses the RoundNearestTiesUp rounding mode. (It’s difficult to guess what breaking ties to nearest “even” TimeType would entail.) Further details on the available RoundingMode s can be found in the API reference.

Rounding should generally behave as expected, but there are a few cases in which the expected behaviour is not obvious.

Rounding Epoch

In many cases, the resolution specified for rounding (e.g., Dates.Second(30)) divides evenly into the next largest period (in this case, Dates.Minute(1)). But rounding behaviour in cases in which this is not true may lead to confusion. What is the expected result of rounding a DateTime to the nearest 10 hours?

1. julia> round(DateTime(2016, 7, 17, 11, 55), Dates.Hour(10))
2. 2016-07-17T12:00:00

That may seem confusing, given that the hour (12) is not divisible by 10. The reason that 2016-07-17T12:00:00 was chosen is that it is 17,676,660 hours after 0000-01-01T00:00:00, and 17,676,660 is divisible by 10.

As Julia Date and DateTime values are represented according to the ISO 8601 standard, 0000-01-01T00:00:00 was chosen as base (or “rounding epoch”) from which to begin the count of days (and milliseconds) used in rounding calculations. (Note that this differs slightly from Julia’s internal representation of Date s using Rata Die notation; but since the ISO 8601 standard is most visible to the end user, 0000-01-01T00:00:00 was chosen as the rounding epoch instead of the 0000-12-31T00:00:00 used internally to minimize confusion.)

The only exception to the use of 0000-01-01T00:00:00 as the rounding epoch is when rounding to weeks. Rounding to the nearest week will always return a Monday (the first day of the week as specified by ISO 8601). For this reason, we use 0000-01-03T00:00:00 (the first day of the first week of year 0000, as defined by ISO 8601) as the base when rounding to a number of weeks.

Here is a related case in which the expected behaviour is not necessarily obvious: What happens when we round to the nearest P(2), where P is a Period type? In some cases (specifically, when P <: Dates.TimePeriod) the answer is clear:

1. julia> round(DateTime(2016, 7, 17, 8, 55, 30), Dates.Hour(2))
2. 2016-07-17T08:00:00

4. julia> round(DateTime(2016, 7, 17, 8, 55, 30), Dates.Minute(2))
5. 2016-07-17T08:56:00

This seems obvious, because two of each of these periods still divides evenly into the next larger order period. But in the case of two months (which still divides evenly into one year), the answer may be surprising:

1. julia> round(DateTime(2016, 7, 17, 8, 55, 30), Dates.Month(2))
2. 2016-07-01T00:00:00

Why round to the first day in July, even though it is month 7 (an odd number)? The key is that months are 1-indexed (the first month is assigned 1), unlike hours, minutes, seconds, and milliseconds (the first of which are assigned 0).

This means that rounding a DateTime to an even multiple of seconds, minutes, hours, or years (because the ISO 8601 specification includes a year zero) will result in a DateTime with an even value in that field, while rounding a DateTime to an even multiple of months will result in the months field having an odd value. Because both months and years may contain an irregular number of days, whether rounding to an even number of days will result in an even value in the days field is uncertain.

See the API reference for additional information on methods exported from the Dates module.

API reference

Dates and Time Types

Dates.Period — Type

1. Period
2. Year
3. Quarter
4. Month
5. Week
6. Day
7. Hour
8. Minute
9. Second
10. Millisecond
11. Microsecond
12. Nanosecond

Period types represent discrete, human representations of time.

Code	Examples	Comment
`y`	6	Numeric year with a fixed width
`Y`	1996	Numeric year with a minimum width
`m`	1, 12	Numeric month with a minimum width
`u`	Jan	Month name shortened to 3-chars according to the `locale`
`U`	January	Full month name according to the `locale` keyword
`d`	1, 31	Day of the month with a minimum width
`H`	0, 23	Hour (24-hour clock) with a minimum width
`M`	0, 59	Minute with a minimum width
`S`	0, 59	Second with a minimum width
`s`	000, 500	Millisecond with a minimum width of 3
`e`	Mon, Tue	Abbreviated days of the week
`E`	Monday	Full day of week name

Code	Matches	Comment
`y`	1996, 96	Returns year of 1996, 0096
`Y`	1996, 96	Returns year of 1996, 0096. Equivalent to `y`
`m`	1, 01	Matches 1 or 2-digit months
`u`	Jan	Matches abbreviated months according to the `locale` keyword
`U`	January	Matches full month names according to the `locale` keyword
`d`	1, 01	Matches 1 or 2-digit days
`H`	00	Matches hours (24-hour clock)
`I`	00	For outputting hours with 12-hour clock
`M`	00	Matches minutes
`S`	00	Matches seconds
`s`	.500	Matches milliseconds
`e`	Mon, Tues	Matches abbreviated days of the week
`E`	Monday	Matches full name days of the week
`p`	AM	Matches AM/PM (case-insensitive)
`yyyymmdd`	19960101	Matches fixed-width year, month, and day