Publish some benchmarks

[spacemanspiff2007]

When searching for different python time libraries there still countless articles popping up who suggest that especially pendulum has a huge speed advantage over arrow. While these might have been true, it's absolutely unclear how the situation has changed since then. It would be nice if there was a small benchmark section in the docs which would compare basic operations.

[anishnya]

@jadchaar, I think we should do this.

[jadchaar]

Agreed especially so we can compare before and after the Cython work we are planning. Could be cool to constantly update a wiki page (with benchmark graphs) with a CI workflow on every tag and every python version.

[systemcatch]

Using code from https://aboutsimon.com/blog/2016/08/04/datetime-vs-Arrow-vs-Pendulum-vs-Delorean-vs-udatetime.html and modifying it to run on Python 3.9 I was able to get the following results.

Note - I couldn't get udatetime to install and quite a bit has changed with pendulum making some of the comparisons difficult.

============ benchmark_parse
_datetime 9.640075228000114 [9.763600978999875, 9.668748288000188, 9.640075228000114, 9.760113094999724, 9.717294457999742]
_arrow 67.04907164500037 [67.04907164500037, 68.1874751600003, 67.57100422900021, 67.84225658600008, 67.89600008499974]
_pendulum 6.8031756150003275 [6.8477682889997595, 6.81757830000015, 6.8031756150003275, 6.8112750700001925, 6.854873013000088]
_delorean 79.36898372299993 [79.36898372299993, 79.53761111499989, 79.84778318200006, 80.14455868899995, 80.14512624700001]

============ benchmark_format
_datetime 3.239017418000003 [3.3066458970001804, 3.253738739000255, 3.239017418000003, 3.2690720719997444, 3.2413774930000727]
_arrow 1.4535876850000022 [1.455979157999991, 1.4539373530001285, 1.4550910959997054, 1.4535876850000022, 1.4555275489997257]
_pendulum 1.5255801739999697 [1.5255801739999697, 1.529671358000087, 1.5531356820001747, 1.526431623999997, 1.5340547209998476]
_delorean 3.875935237000249 [3.9699474889998783, 3.992204258000129, 3.959789434999948, 3.875935237000249, 3.8760139239998352]

============ benchmark_utcnow
_datetime 1.2329302449998067 [1.2587823889998617, 1.234189470000274, 1.2447910880000563, 1.2346941650002918, 1.2329302449998067]
_arrow 2.3831869979999283 [2.3915442749998874, 2.38454174699973, 2.3831869979999283, 2.410346625999864, 2.3863246780001646]
_pendulum 2.416928165000172 [2.418707472999813, 2.42483995099974, 2.416928165000172, 2.430766498000139, 2.4310383629999706]
_delorean 2.5732629419999284 [2.574094752000292, 2.574406983999779, 2.579042515999845, 2.57815893399993, 2.5732629419999284]

============ benchmark_now
_datetime 2.1431128250001166 [2.1431128250001166, 2.1466130099997827, 2.1494242970002233, 2.1619246580003164, 2.1500752520000788]
_arrow 32.88162481800009 [32.96679717000006, 32.95545565300017, 32.88708177999979, 32.88162481800009, 32.98677041999963]
_pendulum 3.2865496659997007 [3.33860700599962, 3.301104772999679, 3.291630373000771, 3.2939674620001824, 3.2865496659997007]
/home/chris/.pyenv/versions/bench_arrow/lib/python3.9/site-packages/delorean/dates.py:170: PytzUsageWarning: The normalize method is no longer necessary, as this time zone supports the fold attribute (PEP 495). For more details on migrating to a PEP 495-compliant implementation, see https://pytz-deprecation-shim.readthedocs.io/en/latest/migration.html
  dt = tz.normalize(dt)
_delorean 6.749951231000523 [6.775615546000154, 6.749951231000523, 6.754888565999863, 6.758163626000169, 6.768949623000481]

============ benchmark_fromtimestamp
_datetime 2.2006504309993034 [2.204797606000284, 2.2577747280010954, 2.2579442980004387, 2.2277943460012466, 2.2006504309993034]
_arrow 32.60892093700022 [33.383653486000185, 33.16171884899995, 32.60892093700022, 33.319603480000296, 33.178806079000424]
_pendulum 3.149250230000689 [3.2103926239997236, 3.1613073339995026, 3.149250230000689, 3.1933716659987113, 3.181643628000529]
_delorean 3.508655527999508 [3.508655527999508, 3.6782505570008652, 3.590653953999208, 3.670900126999186, 3.5859112749985798]

[jadchaar]

We should probably add descriptions to each benchmark and work on graphing the results.

[systemcatch]

We should probably add descriptions to each benchmark and work on graphing the results.

Absolutely, we also need to check this is actually doing fair comparisons as well. Looks like parse/now/fromtimestamp are where arrow is slower.