omnibus-2024-08-02

[maddenp-noaa]

Synopsis

1. Simplify access to drivers' internal config. Previously, we had .config, .config_full, and ._driver_config properties. .config provided the block including the driver's own block and the platform block (i.e. {"driver-name": ..., "platform": ...}); .config_full provided the full input config, even content not relevant to the driver; ._driver_config provided the block under driver-name. This PR simplifies this: The private instance variable ._config_full now provides the {"driver-name": ..., "platform": ...} block, and .config_full provides a public read-only copy of that; and the private instance variable ._config now provides the block under driver-name, and .config provides a public read-only copy of that. So, .config replaces the old ._driver_config, and .config_full replaces the old .config. The driver no longer retains a copy of the complete input config, discarding the irrelevant parts of that input. Many references to these properties are updated to make use of this simpler arrangement.

2. Update uwtools.api.driver to explicitly import the Assets and Driver class families, instead of dynamically adding these to the module. I found in starting external coupled-driver development that running pylint on that driver code flagged e.g. from uwtools.api.driver import DriverCycleBased, reporting that uwtools.api.driver has no DriverCycleBased member. From a static analysis perspective, this is true: The class is added to the module only when it is imported and its top-level code executed, so pylint cannot know this. Without this change, all external driver developers would have to suppress complaints from static analyzers like pylint.

3. Refactor the key-path walking and error-checking code to a public function, and call it from the Assets constructor so that key-path walking there benefits from error checking and reporting.

4. Add public .cycle and .leadtime properties on the appropriate driver classes. I see no reason to hide these, and where external drivers need to access these values, using e.g. self.cycle is IMO less surprising than self._cycle.

5. Remove some .mkdir() calls where they are no longer necessary when they rely on downstream code that already makes necessary parent directories.

6. I recently realized that Path() accepts multiple arguments, similar to the legacy os.path.join, so I've simplified some calls like Path(a) / b to Path(a, b). It's only one character shorter, but IMO looks simpler and more familiar.

7. Fix some API docs.

See some inline comments, too.

Type

• Code maintenance (refactoring, etc. without behavior change)
• Documentation

Impact

• This is a non-breaking change (existing functionality continues to work as expected)

Checklist

• I have added myself and any co-authors to the PR's Assignees list.
• I have reviewed the documentation and have made any updates necessitated by this change.

[christinaholtNOAA]

LGTM. Just some minor suggestions.

[WeirAE]

Finally got through everything, LGTM!

[NaureenBharwaniNOAA]

LGTM!