Compatibility-breaking change in hydra 1.1.0dev7

[rsokl]

This is the culprit:

facebookresearch/hydra#1581

It looks like Hydra has support for callbacks now. I don't have very deep insight into how it works yet.

@jgbos I took a swing at fixing the compatibility-breaking change, but definitely take a look at their changes yourself. I am not confident that my "fix" is totally general / appropriate!

Note that run_job now also accepts an optional hydra_context argument. It looks like we need to pass it a context:

  tests/experimental/test_validation.py: 4 warnings
    /home/runner/work/hydra-zen/hydra-zen/.tox/py/lib/python3.9/site-packages/hydra/core/utils.py:94: UserWarning: 
    run_job's signature has changed in Hydra 1.1. Please pass in hydra_context.
    Support for the old style will be removed in Hydra 1.2.
    For more info, check https://github.com/facebookresearch/hydra/pull/1581.
      warnings.warn(

It also looks like hydra.initialize() should be passed a config_path in _load_config() (although it looks like it is valid to / we may want to just pass None)

  tests/experimental/test_implementations.py: 55 warnings
  tests/experimental/test_validation.py: 10 warnings
    /home/runner/work/hydra-zen/hydra-zen/.tox/py/lib/python3.9/site-packages/hydra_zen/experimental/_implementations.py:81: UserWarning: config_path is not specified in hydra.initialize().
    See https://hydra.cc/docs/next/upgrades/1.0_to_1.1/changes_to_hydra_main_config_path for more information.
      with initialize():

Some other things:

• I updated deprecated imports from hydra.experimental.
• I removed an interpolation utility that I wrote but that we never used / started failing a test.

[rsokl]

@jgbos it doesn't look like .on_run_end() necessarily accepts job_return; it looks like .on_job_end() does (which we don't have to worry about).

Is there a specific reason why you specified this in our case?

[jgbos]

i just copied that from one of hydra's test cases

[rsokl]

Hmm.. Well I'm thinking we may want to remove it in case it creates any issues with callbacks that don't expect it

[jgbos]

oh whoops, i misread that, I was thinking it was in the test. The code came from Hydra's code:

https://github.com/facebookresearch/hydra/blob/master/hydra/_internal/hydra.py#L107

Probably best to just stick with the signature in the abstract method.

[rsokl]

Ah! Hmm that might be worth adding back in then just so that we mirror them. Good catch / sleuthing!