uf.optional

Import optional dependencies without masking broken installs.

A bare except ImportError around import pandas also swallows the ImportError of a pandas that is installed but fails to import (a missing dependency of pandas, an incompatible binary). The feature then silently turns itself off. import_optional tolerates only the absence of the named package itself.

Functions

import_optional(module_name, *[, absent])

Return the imported module, or None if the optional package is not installed.

uf.optional.import_optional(module_name, *, absent=None)

Return the imported module, or None if the optional package is not installed.

Only a ModuleNotFoundError naming the package itself (the first segment of module_name, or one of absent) counts as “not installed”. Every other failure – a missing submodule, a missing dependency of the package, a broken import inside it – is a broken environment and propagates. A namespace package (a bare directory of that name on sys.path, with no __init__.py) is not the package either, so it counts as not installed – which also means a package that is legitimately a namespace package (google) can’t be imported with this.

Return type:

Optional[ModuleType]

>>> import_optional('json').__name__
'json'
>>> import_optional('_uf_no_such_package_') is None
True
>>> import_optional('json.no_such_submodule')
Traceback (most recent call last):
  ...
ModuleNotFoundError: No module named 'json.no_such_submodule'