Speedup parsing of datetime fieldtypes initialization by string

[yunzheng]

This change mainly removes the use of expensive regexes and exception handling, improving the speed significantly.

[codecov[bot]]

Codecov Report

Merging #87 (9b268b2) into main (f0a2608) will increase coverage by 0.04%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #87      +/-   ##
==========================================
+ Coverage   79.22%   79.27%   +0.04%     
==========================================
  Files          32       32              
  Lines        2932     2939       +7     
==========================================
+ Hits         2323     2330       +7     
  Misses        609      609              

Flag	Coverage Δ
unittests	79.27% <100.00%> (+0.04%)	:arrow_up:

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
flow/record/fieldtypes/__init__.py	91.53% <100.00%> (+0.13%)	:arrow_up:

:mega: We’re building smart automated test selection to slash your CI/CD build times. Learn more

[yunzheng]

Benchmark script

Benchmarked using the following script.

#!/bin/sh -x

# default datetime str without any TZ data
python3 -m timeit -n 100000  "from flow.record.fieldtypes import datetime" 'datetime("2023-09-04T13:33:37")'

# default isoformat(), this is how it's serialized in msgpack for tz aware datetime objects
python3 -m timeit -n 100000  "from flow.record.fieldtypes import datetime" 'datetime("2023-09-04T13:33:37+03:00")'

# RFC3339Nano, 2006-01-02T15:04:05.999999999, used by Docker
python3 -m timeit -n 100000  "from flow.record.fieldtypes import datetime" 'datetime("2023-09-04T13:33:37.123456789999999")'
python3 -m timeit -n 100000  "from flow.record.fieldtypes import datetime" 'datetime("2023-09-04T13:33:37.123456789999999-02:00")'
python3 -m timeit -n 100000  "from flow.record.fieldtypes import datetime" 'datetime("2023-09-04T13:33:37.123456789999999Z")'

# other variants, but less common
python3 -m timeit -n 100000  "from flow.record.fieldtypes import datetime" 'datetime("2023-09-04T13:33:37Z")'
python3 -m timeit -n 100000  "from flow.record.fieldtypes import datetime" 'datetime("2023-09-04T13:33:37.123456-02:00")'
python3 -m timeit -n 100000  "from flow.record.fieldtypes import datetime" 'datetime("2023-09-04T13:33:37.123456789+03:00")'

Benchmark results

Python 3.8

#	old	new
0	3.48 usec per loop	3.5 usec per loop	-0.02
1	15.9 usec per loop	2.51 usec per loop	13.39
2	8.79 usec per loop	4.35 usec per loop	4.44
3	31.5 usec per loop	3.5 usec per loop	28.0
4	27.6 usec per loop	3.2 usec per loop	24.4
5	13.4 usec per loop	2.57 usec per loop	10.83
6	18.3 usec per loop	3.47 usec per loop	14.83
7	32.8 usec per loop	3.29 usec per loop	29.51

Python 3.9

#	old	new
0	3.6 usec per loop	3.53 usec per loop	0.07
1	16.6 usec per loop	2.55 usec per loop	14.05
2	9.07 usec per loop	4.33 usec per loop	4.74
3	32.1 usec per loop	3.46 usec per loop	28.64
4	28.2 usec per loop	3.36 usec per loop	24.84
5	13.7 usec per loop	2.63 usec per loop	11.07
6	18.4 usec per loop	3.48 usec per loop	14.92
7	32.0 usec per loop	3.18 usec per loop	28.82

Python 3.10

#	old	new
0	3.11 usec per loop	3.14 usec per loop	-0.03
1	13.2 usec per loop	2.29 usec per loop	10.91
2	7.14 usec per loop	4.05 usec per loop	3.09
3	24.6 usec per loop	3.61 usec per loop	20.99
4	21.7 usec per loop	3.05 usec per loop	18.65
5	9.99 usec per loop	2.55 usec per loop	7.44
6	13.8 usec per loop	3.29 usec per loop	10.51
7	24.5 usec per loop	2.99 usec per loop	21.51

Python 3.11

note, Python3.11 was already fast pathed previously.

#	old	new
0	2.12 usec per loop	2.12 usec per loop	0.0
1	1.54 usec per loop	1.33 usec per loop	0.21
2	2.24 usec per loop	2.12 usec per loop	0.12
3	1.45 usec per loop	1.43 usec per loop	0.02
4	1.37 usec per loop	1.34 usec per loop	0.03
5	1.41 usec per loop	1.3 usec per loop	0.11
6	1.45 usec per loop	1.38 usec per loop	0.07
7	1.46 usec per loop	1.41 usec per loop	0.05