Travis: Add pypy3

[rht]

This addresses #882.

[codecov]

Codecov Report

Merging #883 (9453b67) into master (8a8e05c) will decrease coverage by 2.71%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #883      +/-   ##
==========================================
- Coverage   88.86%   86.14%   -2.72%     
==========================================
  Files          19       20       +1     
  Lines        1132     1083      -49     
  Branches      189      171      -18     
==========================================
- Hits         1006      933      -73     
- Misses         92      121      +29     
+ Partials       34       29       -5     

Impacted Files	Coverage Δ
mesa/batchrunner.py	73.74% <0.00%> (-16.31%)	⬇️
mesa/datacollection.py	97.46% <0.00%> (-0.24%)	⬇️
mesa/time.py	95.45% <0.00%> (-0.07%)	⬇️
mesa/model.py	100.00% <0.00%> (ø)
mesa/visualization/TextVisualization.py	0.00% <0.00%> (ø)
mesa/visualization/modules/TextVisualization.py	100.00% <0.00%> (ø)
mesa/visualization/__init__.py	100.00% <0.00%> (ø)
mesa/visualization/modules/HexGridVisualization.py	96.42% <0.00%> (+0.13%)	⬆️
...a/visualization/modules/CanvasGridVisualization.py	96.00% <0.00%> (+0.16%)	⬆️
mesa/visualization/ModularVisualization.py	76.66% <0.00%> (+0.19%)	⬆️
... and 6 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8a8e05c...319165a. Read the comment docs.

[rht]

I'm not sure why there is no Travis running here ...

[Corvince]

I'm not sure why there is no Travis running here ...

No idea why it doesn't show up. It did run though, but failed at installing pandas (https://travis-ci.org/github/projectmesa/mesa/jobs/718907972)

[rht]

It looks like it fails at pandas' dependency, ujson. I will keep this PR open and retry on future version of PyPy. Though PyPy developers claimed that pandas has been supported since pypy5.9 (https://doc.pypy.org/en/latest/release-v5.9.0.html#pypy2-7-and-pypy3-5-v5-9-dual-release). The current Travis PyPy is 7.1.1.

[Corvince]

Your link only mentions pypy2.7, but from what I could infer from googling around, yes, it should work with pandas. Maybe just an issue with travis? @rht could you (or someone else) try it locally?

[rht]

Tested on my laptop.
pypy3-7.3.1:

97 passed, 1 warning in 11.26s ================================================

With normal python 3.8

97 passed, 2 warnings in 11.63s ===============================================

[rht]

I probably need a simulation with a huge grid size in order to highlight PyPy's speedup.

[rht]

I ran the comparison on the bank_reserves example, because it doesn't have the web GUI.
The speedup is significant!
PyPy3-7.3.1

python batch_run.py  89.72s user 2.78s system 100% cpu 1:31.72 total

CPython 3.8

python batch_run.py  173.73s user 2.19s system 100% cpu 2:55.10 total

A 1.9x speedup!

[rht]

Here I removed 3.5 to make room for PyPy3. This is also justified by the fact that 3.5's EOL 2020-09-13.