Configuration #936
Configuration #936
Conversation
…FrozenNamespace by Enum
# Conflicts: # model/common/pyproject.toml # uv.lock
… _nudge_max_coeff from config yml
fix Mixin import
# Conflicts: # model/atmosphere/diffusion/tests/diffusion/integration_tests/test_benchmark_diffusion.py # model/testing/src/icon4py/model/testing/definitions.py
# Conflicts: # model/atmosphere/diffusion/tests/diffusion/integration_tests/test_benchmark_diffusion.py # model/atmosphere/dycore/tests/dycore/integration_tests/test_benchmark_solve_nonhydro.py # model/driver/src/icon4py/model/driver/icon4py_configuration.py # model/testing/src/icon4py/model/testing/fixtures/datatest.py
# Conflicts: # model/atmosphere/diffusion/src/icon4py/model/atmosphere/diffusion/diffusion.py # uv.lock
|
Mandatory Tests Please make sure you run these tests via comment before you merge!
Optional Tests To run benchmarks you can use:
To run tests and benchmarks with the DaCe backend you can use:
To run test levels ignored by the default test suite (mostly simple datatest for static fields computations) you can use:
For more detailed information please look at CI in the EXCLAIM universe. |
| self._initialize_components(user_config) | ||
|
|
||
| # TODO (halungge): have this return a std dict to hide away omegaconf | ||
| def get( |
There was a problem hiding this comment.
Make this return a TypeDict..
OngChia
left a comment
OngChia
left a comment
There was a problem hiding this comment.
Thanks a lot for the configuration infrastructure!
It looks awesome. Currently, I have a few questions on the configuration manager.
| """ | ||
|
|
||
| # number of dynamics substeps | ||
| ndyn_substeps: int = dataclasses.field( |
There was a problem hiding this comment.
In contrast to DiffusionConfig, we do not need ndyn_substeps for NonHydrostaticConfig because stencils in SolveNonhydro need the runtime changeable ndyn_substeps_var.
|
|
||
| # from mo_nonhydrostatic_nml.f90 | ||
| ndyn_substeps: int = dataclasses.field( | ||
| default=common_config.resolve_or_else("model.ndyn_substeps", 5) |
There was a problem hiding this comment.
Does ndyn_substeps always have to be inherited from a outer config? Or I can do, let's say,
ndyn_substeps: int = dataclasses.field(
default=common_config.resolve_or_else("model.run.ndyn_substeps", 5)
)
I feel that it is more suitable for ndyn_substeps to sit in the run config, because this variable controls how many substeps the driver runs the dycore.
| default_reader = module.init_config() | ||
| default_reader.update(update) | ||
| else: | ||
| log.warning(f" module {module.__name__} has no `init_config` function") |
There was a problem hiding this comment.
Should we force config.py in all components has this init_config function?
Otherwise, it seems that users can have some wrong parameters in the full_model_config.yaml file and probably an error will be thrown when the granule is initialized.
| def get( | ||
| self, | ||
| is_default: bool = False, | ||
| ) -> oc.DictConfig: |
There was a problem hiding this comment.
I am thinking that whether to give users an option of returning a dictionary of config classes (run, dycore, diffusion, ...) for convenience?
| f"No default configuration and no user-configuration for component {name} defined in file {self._config_file}" | ||
| ) | ||
| else: | ||
| reader.update(module_config) |
There was a problem hiding this comment.
It may be true in the if statement above that users have developed a component without any config parameters.
However, I am wondering that should we just throw an error here in this case where users put a config for a particular component whereas there is in config.py.
Adds Configuration for ICON4Py in two parts:
commonprovides aConfigurationHandlerfor initialization of configuration for a@dataclassor a yaml file. This one can be used in each model component. There are two examples for this inmodel.atmosphere.diffusionandmodel.atmosphere.dycoreConfigurationMangerinmodel.driverwhich is supposed to be used in a model run. It reads a yaml file and initializes aConfigurationHandlerfor each model registered in the model config and makes the configuration available (as dictionary) to the model.