dateformat does two things: turn datetime objects into strings, and turn strings into datetime objects.
It's goal is to do these things simply and well, and to satisfy the following criteria:
- Be fast enough (see below for benchmarks)
- Handle a variety of date formats from multiple sources
- Parse and format dates in many timezones and with many timezone offsets
- Represent the expected format in a way that a non-technical person may understand
- Be explicit about the expected format to prevent heuristic errors
"But why another date library?"
There isn't currently a python library I've been able to find that matches these
requirements well enough for my use-cases. Arrow
comes closest, but still isn't quite suitable performance-wise.
All functionality is based around DateFormat() objects:
create a dateformat object from the provided spec string.
>>> from dateformat import DateFormat
>>> date_format = DateFormat("YYYY-MM-DD hh:mm:ss.SSSS+HH:MM")
If is_24hour is not provided, the format will be in 12-hour mode if an am/pm
part is present in the spec, otherwise, dates will be in 24-hour mode.
DateFormat instances have two methods:
Parse a string(data) containing a date into a datetime object.
>>> date = date_format.parse("2017-06-03 15:32:00.2364-02:00")
datetime.datetime(2017, 6, 3, 15, 32, 0, 236400, tzinfo=datetime.timezone(datetime.timedelta(-1, 79200)))
Format the passed in datetime.datetime object (date) as a string:
>>> print(date_format.format(date))
2017-06-03 15:32:00.2364-02:00
If any part of the format provides a timezone, or UTC offset, then parsing produces dates with a timezone indicating the relevant UTC offset.
Likewise, if a dateformat has a timezone part, then dates passed to .format()
must include a tzinfo value.
If pytz is available, then some level of named timezone support is provided.
All numeric parts of the date format are zero-padded to the number of characters in the spec. I.e. 'DD' means that the day of the month is zero-padded to 2-digits.
During parsing, a missing leading zero is usually ignored, but if there is no separator between parts (for example: YYYYMMDD), then a missing leading zero will cause an error or bad value.
Currently, all formatted dates are zero-padded, in the future, this may be controllable.
| Part | Example | Description |
|---|---|---|
+HHMM |
-0515 | A UTC offset provided as a 2-digit hour, and 2-digit minute, with no separator |
+HH:MM |
-05:15 | A UTC offset provided as a 2-digit hour, and 2-digit minute, with a ':' separator |
+HH |
-05 | A UTC offset provided as a 2-digit hour only |
Dddddd |
Monday | The full locale-specific day of the week (Note, this value is ignored during date parsing, but added during date format) |
Ddd |
Mon | The locale-specific short name for the day of the week (Ignored during parsing) |
DD |
05 | The zero-padded day of the month. |
MMMMM |
September | Month as locale’s full name |
MMM |
Sep | Month as locale’s abbreviated name |
MM |
05 | The zero-padded month as number 1 - 12 |
YYYY |
2017 | Year with century as a number |
YY |
17 | Year without century as a zero-padded number |
hh |
09 | Hour as a zero-padded number |
mm |
06 | Minute as a zero-padded number |
ss |
45 | Second as a zero-padded number |
SSSSSS |
123456 | Microsecond as a zero-padded decimal number |
SSSS |
1234 | 100-microseconds as a zero-padded number |
SSS |
123 | milliseconds as a zero-padded number |
SS |
12 | 10-milliseconds as a zero-padded number |
am Am AM pm Pm PM |
am | either AM or PM depending on the hour. .format() matches the case of the spec. If present, the dateformat will default to 12-hour mode |
of |
Ignored during parsing, added during formtting | |
st |
th | The appropriate suffix for the day of the month, for example '1_st_ July', '2_nd_ March' |
␣ |
T | (Unicode OPEN BOX - U+2423) Matches either the character 'T' or a space ' '. During formatting, 'T' is always used (this is provided to improve flexibility when parsing iso8601 formats) |
| space | Matches one or more spaces during parsing. During formatting, one space will be output | |
any of :/-.,TZ() |
Ignored during parsing, output as-is during formatting |
| Format | Example |
|---|---|
YYYY-MM-DDThh:mm:ss |
2017-06-06T09:45:15 |
YYYYMMDDhhmmss |
20170606094515 |
YYYYMMDDhhmmss.SSSSSSZ |
20170606094515.123456Z |
MM/DD/YY hh:mm+HHMM |
06/06/17 09:45-0500 |
Dateformat is not trying to be a replacement for the builtin datetime module. datetime.datetime objects are used as the input/output to the parsing and formatting methods.
It is designed as a replacement for the datetime.datetime.strftime and datetime.datetime.strptime methods, providing:
- better timezone handling
- a simpler/more common syntax for specifying the date formats
- slightly faster parsing
dateutil.parser.parse's intent is to turn a string in an unknown format into a date. It does that by using a variety of heuristics to try to figure out the format the date has been expressed in.
This approach is highly useful, and very flexible, but suffers from a couple of drawbacks that dateformat doesn't have:
- There is ambiguity about what date will be produced from a given string, there are situations where that risk cannot be accepted, and it's important for the system to only accept a certain date format
- Because of all the work that dateutil is doing to work out the format used, it's fairly slow, at just under 10x slower than
strptime, this is very noticable over 10s - 100s thousands of dates.
arrow is the closest to the way dateformat works, the syntax for describing dates is very similar. Unfortunately, arrow constructs its parser every time a date is parsed, creating a significant overhead when parsing each date.
ciso8601 is really fast. Unfortunately both these libraries only handle a single date format, so are not useful in this situation.
the benchmark/ dir contains some simple scripts to show how the relative libraries perform at parsing and formatting dates.
Running on a 2016 macbook pro, on Python 3.6.3 gave the following results (best of 3 runs):
(Please note, the parse time chart y-axis has been clamped to 1s, but dateparser took 16s to complete)
