Working with date-time formats can be pretty upsetting because of the variate of different formats people can come up with. date-times are used everywhere not just only logging or meta data in database entries and are pretty important. That’s why I encourage developers in using the ISO 8601 derived RFC3339 standard for their projects.
RFC3339 date-time: 2016-07-18T12:58:26.485897 +02:00
The RFC3339 specification offers the following advantages:
- Defined date, time, timezone, date-time format
- 4 digit year
- Fractional seconds
- Human readable
- No redundant information like weekday name
- Simple specification
- Machine readable
Having a date-time standard is nice, but using Python’s datetime library
to parse/format a RFC3339 date-time string or even create a datetime object
in UTC or local timezone can be painful and slowwwww. That’s why I decided to
implement a Python 2 library to deal with such tasks. The library is called
udatetime and available on github
or PyPI.
$ pip install udatetime
The goal of the library is to be fast and handy with RFC3339 date-time
formatted strings. The average performance increase of udatetime compared to
the equivalent datetime code is 76%. Due to the usage of Python2
CPython API and POSIX features the library is currently only supported on POSIX
systems and not Python3 or Pypy compatible. I’m working on cross-platform and
Pypy support. Support in working on the library is greatly appreciated.
Benchmark
The benchmark setup is the following.
fromdatetimeimportdatetimeimportudatetimeRFC3339_DATE='2016-07-18'RFC3339_TIME='12:58:26.485897+02:00'RFC3339_DATE_TIME=RFC3339_DATE+'T'+RFC3339_TIMERFC3339_DATE_TIME_DTLIB=RFC3339_DATE_TIME[:-6]# datetime can't parse timezones through strptimeDATE_TIME_FORMAT='%Y-%m-%dT%H:%M:%S.%f'DATETIME_OBJ=datetime.strptime(RFC3339_DATE_TIME_DTLIB,DATE_TIME_FORMAT)defbenchmark_parse():defdatetime_strptime():datetime.strptime(RFC3339_DATE_TIME_DTLIB,DATE_TIME_FORMAT)defudatetime_parse():udatetime.from_string(RFC3339_DATE_TIME)return(datetime_strptime,udatetime_parse)defbenchmark_format():defdatetime_strftime():DATETIME_OBJ.strftime(DATE_TIME_FORMAT)defudatetime_format():udatetime.to_string(DATETIME_OBJ)return(datetime_strftime,udatetime_format)defbenchmark_utcnow():defdatetime_utcnow():datetime.utcnow()defudatetime_utcnow():udatetime.utcnow()return(datetime_utcnow,udatetime_utcnow)defbenchmark_now():defdatetime_now():datetime.now()defudatetime_now():udatetime.now()return(datetime_now,udatetime_now)defbenchmark_utcnow_to_string():defdatetime_utcnow_to_string():datetime.utcnow().strftime(DATE_TIME_FORMAT)defudatetime_utcnow_to_string():udatetime.utcnow_to_string()return(datetime_utcnow_to_string,udatetime_utcnow_to_string)defbenchmark_now_to_string():defdatetime_now_to_string():datetime.now().strftime(DATE_TIME_FORMAT)defudatetime_now_to_string():udatetime.now_to_string()return(datetime_now_to_string,udatetime_now_to_string)If you like you can run the benchmark yourself by running the bench.py script
from the repository.
The results of 1 million executions and 3 repeats look like this.
benchmark_parse
datetime.strptime(RFC3339_DATE_TIME_DTLIB, DATE_TIME_FORMAT)
vs
udatetime.from_string(RFC3339_DATE_TIME)
benchmark_format
DATETIME_OBJ.strftime(DATE_TIME_FORMAT)
vs
udatetime.to_string(DATETIME_OBJ)
benchmark_now
datetime.now()
vs
udatetime.now()
benchmark_utcnow
datetime.utcnow()
vs
udatetime.utcnow()
benchmark_now_to_string
datetime.now().strftime(DATE_TIME_FORMAT)
vs
udatetime.now_to_string()
benchmark_utcnow_to_string
datetime.utcnow().strftime(DATE_TIME_FORMAT)
vs
udatetime.utcnow_to_string()
