Proposal for fixed generation of benchmark data.

[trexfeathers]

Have a read of benchmarks/benchmarks/generate_data.py - the docstrings provide detail on my thoughts for this.

Summary: run generation scripts in an alternative environment that will therefore remain unchanged throughout the benchmark run.

[codecov[bot]]

Codecov Report

Merging #124 (5340ce6) into main (10820f2) will not change coverage. The diff coverage is n/a.

@@           Coverage Diff           @@
##             main     #124   +/-   ##
=======================================
  Coverage   98.85%   98.85%           
=======================================
  Files          14       14           
  Lines         699      699           
=======================================
  Hits          691      691           
  Misses          8        8           

---

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 10820f2...5340ce6. Read the comment docs.

[trexfeathers]

I've so far made minimal changes to the benchmarks themselves, but given this PR forces saving of sample data (rather than just creating a Python object), I'd be interested in @stephenworsley's thoughts on better optimisation via ASV's setup_cache.

[jamesp]

Take this as a suggestion, not a review. You could refactor into either a closure or a class that wraps the python executable rather than hard coding to a global constant.

e.g.

class PythonRunner:
  def __init__(self, python:Path):
     self.python=python

  def __call__(self, code, *args, **kwargs):
     ... code of run_elsewhere using self.python ...

# in the specific data gen code
run_elsewhere = PythonRunner(GEN_DATA_PYTHON)

or

def make_python_runner(python: Path):
  def run_elsewhere(code, *args, **kwargs): ...
  return run_elsewhere

run_elsewhere = make_python_runner(GEN_DATA_PYTHON)

You get the picture.

[trexfeathers]

You could refactor into either a closure or a class that wraps the python executable rather than hard coding to a global constant.

If this needed to be more generic, sure. But that's engineering for something that I don't expect to happen: https://github.com/SciTools-incubator/iris-esmf-regrid/pull/124#discussion_r742927858