Refactor dependencies in pyproject.toml

[andersy005]

• Refactored dependencies in pyproject.toml. The original monolithic list has been split into optional dependencies under 'pangeo-forge' and 'catalog'.
• Added 'complete' directive under optional dependencies for encapsulating all dependencies.

I believe these changes will facilitate better dependecies management (for e.g. i would like to use the cataloging features without needing to install apache-beam, etc...)

[jbusecke]

Ah nice! Thanks @andersy005. Looks good to me!

[andersy005]

it turns out that the imports in

leap-data-management-utils/leap_data_management_utils/__init__.py

Lines 1 to 4 in cd86889

	from .cmip_transforms import CMIPBQInterface, IIDEntry, LogCMIPToBigQuery
	from .data_management_transforms import BQInterface, RegisterDatasetToCatalog
	__all__ = (

need to be updated for this to work. currently,

from leap_data_management_utils.catalog import main

goes through init.py and this triggers import of all dependencies (includding those i wanted to avoid (apache-beam, etc))

[jbusecke]

? I am not quite sure I understand.

[andersy005]

? I am not quite sure I understand.

when you execute the command pip install leap-data-management["catalog"], it installs both the base dependencies and the additional catalog dependencies. however, attempting to import the package with the following code:

from leap_data_management_utils import catalog

triggers the invocation of imports within the leap_data_management_utils.__init__ file, which currently references objects from subpackages, such as cmip_transforms.py and data_management_transforms.py.

[andersy005]

what is desired is the ability to install leap-data-management["catalog"] and subsequently import the catalog subpackage without initiating imports from any other subpackages. does this make sense? :0

[jbusecke]

Ahh thanks for clarifying. I think we can take those imports out of the init.py. It will break some of my stuff downstream, but should be super easy to fix.