pyrollup 0.1.0

Description:

pyrollup 0.1.0

pyrollup
Simple mechanism to rollup API symbols to a Python module from its submodules
Background
Conventionally, a Python package's public API symbols (or at least a subset) are exported from the top-level package. These symbols are typically imported from various submodules in the module hierarchy. This leads to a rather tedious listing of symbols exported (via a module's __init__.py) at each level.
For example:
from .submodule_a import ClassA, ClassA2
from .submodule_b import ClassB
from .submodule_c import ClassC

__all__ = [
ClassA,
ClassA2
ClassB,
ClassC,
]

The definition of __all__ can be omitted, although some schools of thought encourage its use.
Problem statement
With the above approach, the burden of which symbols are considered public for a given submodule falls on the containing module importing them. In addition, __all__ must be maintained alongside symbol imports, potentially resulting in duplication.
This can become problematic for complex projects with many nested symbols imported at the top-level package. For example:
from .submodule_a import ClassA, ClassA2
from .submodule_a.submodule_a1 import ClassA1_1, ClassA1_2
from .submodule_a.submodule_a2 import ClassA2_1, ClassA2_2
from .submodule_a.submodule_a3 import ClassA3_1, ClassA3_2
from .submodule_b import ClassB
from .submodule_c import ClassC

__all__ = [
ClassA,
ClassA2,
ClassA1_1,
ClassA1_2,
ClassA2_1,
ClassA2_2,
ClassA3_1,
ClassA3_2,
ClassB,
ClassC,
]

In addition, submodules must similarly maintain their own imports and/or definition of __all__ if they export symbols from their submodules.
Proposed solution
Ideally, a module should "rollup" public symbols from its submodules, letting each submodule decide which symbols those are. Submodules should therefore optionally provide an allow-list and block-list to describe which public symbols should be propagated to the parent module.
In other words, a given module should have ownership of:

Its own public symbols (Python convention)

Defined by __all__


Which of those public symbols should be propagated to parent modules (functionality provided by pyrollup)

Defined by

__rollup__: allow-list, defaulting to __all__
__nrollup__: block-list, defaulting to []





Then, the example above can be modified as:
from pyrollup import rollup

# import submodules
from . import submodule_a, submodule_b, submodule_c

# import public symbols from each submodule
from .submodule_a import *
from .submodule_b import *
from .submodule_c import *

# export public symbols from each submodule, filtered by allow-list/block-list
__all__ = rollup(submodule_a, submodule_b, submodule_c)

This allows a project with a complex module hierarchy to flexibly propagate public symbols from wherever they are defined to the top-level package.
Downsides
Static analysis tools are unable to evaluate the value of __all__ since with this approach it is computed dynamically upon import. Therefore, code documentation generator tools using static analysis (e.g. autoapi or autodoc2) will fail to detect a module's public symbols.
Tools using the traditional approach of importing the package for which documentation is being generated (e.g. Sphinx's built-in autodoc) should have no problem with such dynamic imports.
It is possible to use a "hybrid" approach; the author has had some success with extending autodoc2 to work with dynamically-evaluated __all__.
TODO: document and generalize this solution