Most current programming libraries for time calculations are based on essentially the same model, JodaTime (now java.time in Java 8 and later) being one of the better examples. Core concepts include instants, intervals, periods, and a few other concepts. Time values are wrappers around a Unix Time (a number of milliseconds since January 1970). Chronologies are configurable in principle, but this doesn't usually happen
Time Count is an exploration of alternative ways of thinking about time for business software. It is NOT currently recommended for use in production code.
For examples in code, and more detailed explanations, see files /time-count/explainers/time_count/explainer/* (probably best read in order).
For a discussion of the broad exploration of time models this was part of, check out the video here: https://youtu.be/Zm95cYAtAa8
- Time is Countable
- Everything is an Interval
- Allen's Interval Algebra
- Canonical String Representation (~ ISO 8601)
- Composible Transformations
Calendars and time are (for most business software) weird ways of counting. E.g. at the most basic level, dates are a count of days:
(t-> "2017-04-09" next-t)
=> "2017-04-10"
Viewing time as sequences nested within other sequences allows us to use basic sequence logic to express many common time operations. E.g. Days can be nested within months, and months within years.
(t-> "2017-04-09"
(enclosing :month))
=> "2017-04"
End of month could be expressed as:
(t-> "2017-04-09
(enclosing :month)
(nest :day) t-sequence
last)) ;Note that 'last' is the normal Clojure sequence operation.
=> "2017-04-30"
In time-count we avoid thinking of time units as a measure, as when we might talk about "addition" of a period to a time. There are domains where time as a measure is important (e.g. keeping records for a track-meet), but that might be better handled with a whole different time library focused on that.
See more examples in this file: time_count/explainer/b_counting_and_nesting.clj
All time values are intervals. There are no "Instants". So, for example, a date is an interval one day long. A typical meeting invitation specifies the start time to the minute. A time value could be a specific second, hour, month, year or quarter.
The sequences of countable times are sequences of consecutive intervals, where each interval meets the interval following it.
An important case is the timestamp, which is usually regarded as an "instant". But a timestamp is actually a reading from a physical clock. If that clock ticks each millisecond, then there is an approximately one millisecond-long interval during which you would get a particular reading from the clock. Then there is another millisecond interval in which you'd get the next value, and so on. In time-count, a typical timestamp is represented as an interval, one millisecond long. If the precision of the clock finer or coarser than this, the appropriate scale could be chosen. (Of course, most time stamps come from outside, so this is the time-count interpretation of what they mean.
Allen's Interval Algebra is an elegant formalism for describing and inferring the relations between different time intervals, and it is used in time-count for comparing and defining time values.
Allen defined 13 basic relations that distinctly and exhaustively relate any two intervals. These include the familiar "before" and "after", but defined unambiguously. It also has relations like "meets" and overlaps.
Example pseudocode (where 2016 is substituted for a value for the year 2016, etc.):
(relation 2016 2018) => :before
(relation 2016 2017) => :meets
Any interval can be compared with any other to produce one of Allen's relations. Intervals are defined either as part of a sequence or as a "RelationBoundedInterval".
In a sequence of countable times, each time is an interval that meets the next interval in the sequence.
A relation-bounded interval, is specified by two smaller intervals, one that starts it and one that finishes it.
(relation 2016 2016/2018) => :starts
(A relation-bounded interval may have only one of these relations, and be unbounded on the other side. In time-count, the lack of a bound is interpreted as an unknown bound -- not as an infinite interval.)
See more examples in file: time_count/explainer/c_allens_interval_algebra.clj And a good general explanation at https://www.ics.uci.edu/~alspaugh/cls/shr/allen.html
Having a string representation of a time is first-class concern in time-count, not just display formatting issue. There is a string representation of every time value, with bidirectional operations to convert between the string and any other representation used in computations (isomorphic). time-count uses ISO 8601, whenever there is a suitable representation,
(-> "2017-12-29T17:46"
from-iso
to-iso)
=> "2017-12-29T17:46"
For convenience, two threading macros are provided that accept the string representations as input to time functions and produce strings from the results.
(t-> "2017-04-09" next-t)
=> "2017-04-10"
See more examples in file: time_count/stringifying_tests.clj
time-count is not dependent on a specific representation of time-values, but it includes three:
- canonical strings (based on ISO 8601)
- metajoda
- place-values
Most of the time, in this prototype, the computation-friendly representation is MetaJodaTime, which just adds some metadata to a JodaTime DateTime. This metadata represents a nesting of scales intended to be significant.
MetaJodaTime (in metajoda namespace) is a place-holder implementation. It could be any representation that supports a few low-level functions specified in the CountableTime protocol and the Interval protocol.
Using lets us use a lot of work the JodaTime people have done! E.g.
- How many days are in February in 2017?
- When does New York switch to Daylight Savings Time in 2017? This embedded knowledge is accessed through the basic sequence and nesting functions.
Unintuitively, this means that to use operations like next-t you must require the core namespace (where CountableTime is defined) and also the metajoda namespace, because there must be some implementation of the protocol. (You could provide another implementation instead.)
Another representation is "place values". For example,
(place-values (from-iso "2017-12-28"))
=> [[:day 28] [:month 12] [:year 2017]]
(This representation might actually be an alternative for implementing nest and enclosing.)
See more examples in this file: time_count/explainer/a_representations.clj
This is not a very unusual goal. It is important, though, and not always adhered to in creating time libraries. Instead of creating all the operations people will want when using the library, time-count tries to create a small set of low-level operations that compose well. Each project could then create their own higher-level operations.
For example, "end-of-month" might be a function provided in a library, but it is not provided in time-count. It can be composed (as described in the Countable section above) from the basic sequence and nesting operations:
(defn end-of-month [t]
(-> t (enclosing :month)
(nest :day)
t-sequence
last))
(t-> "2017-04-09" end-of-month)
=> "2017-04-30"
(t-> "2017-04-09T17:46" end-of-month)
=> "2017-04-30"
This is easier partly because time-count connects time with the standard sequence abstraction of the programming language, which supports all sorts of rich operations and compositions. (This is true of many modern languages.)
Also, concepts like "end-of-month" may be defined differently on different projects. If a business defines end-of-month to mean the last business day of the month, they could define a predicate, business-day?, and end-of-month as
(defn end-of-month [t]
(-> t (enclosing :month)
(nest :day) t-sequence
(#(filter business-day? %)
last))
or in some other way suited to their business rules.
See more examples in this file: time_count/explainer/d_composing_operations.clj
No firm plan, but probably,
- Millisecond as a single-level nesting representing a Unix time (milliseconds since the Epoch), and mapable to a millisecond-scale time with a UTC-offset.
- Representing timezones and offsets with the place-value representation (which currently only represents the local time component).