neuralib.argp

Annotation-based argparse

author:

Ta-Shun Su

This module provide a way to integrate python argparse module into class attribute and annotation type, that allow options have type information, and allow parser combination easily.

class Argument and function argument

A simple option class that contains several options carried by their attributes.

>>> class ExampleOptions:
...     ANIMAL: str = argument('--ANIMAL')
...     EXP_DATE: str = argument('--EXP_DATE')
...     OUTPUT_DIR: str = argument()

In ExampleOptions, ANIMAL is an attribute with type annotation str. it has a class variable Argument (argument() return) which contains the arguments of ArgumentParser.add_argument. For now, we have an optional argument --ANIMAL which accept one argument. EXP_DATE as is another optional argument. And OUTPUT_DIR are a potional argument because it doesn’t have dashed options.

>>> opt = parse_args(ExampleOptions())
... print(opt.animal)

After class declared, you can use parse_args() to parse cli arguments. This function will create an ArgumentParser and find out all argument attributes. Then set the attributes from parsed result.

Commandline usage

In bash, you can call this option class

python -m module.path --ANIMAL name --EXP_DATE date output

print_help() does the similar things but print the help document to the stdout.

>>> print_help(opt)

Or use -h options

python -m module.path -h

Annotation type infering

In general, you can think argument() just a delegate funcion that passes the arguments to the ArgumentParser.add_argument. However, this function will try to infer missing arguments based on the annotation type when creating the ArgumentParser. For now, this module is not powerful to handle all possible case. there are support type (Please see complete_kwargs() for detailed):

  1. bool: infer parameter action to store_true, default to False.

  2. Literal[...] : infer parameter choices.

  3. Optional[T]: infer parameter type to T if it is callable. If it is Literal, apply 2.

  4. callable(...) with signature Callable[[str], T]: infer parameter type to T

  5. parameter dest always use the attribute name.

Additionally, Argument also provide a parameter group to reduce the complexity of create subgrouping parser.

Option class compose

Option class can be composed by inherition. Child option class can also change the value from parent’s argument. As well as disable it (by replacing a value)

>>> class MoreOptions(ExampleOptions):
...     # additional optional option
...     verbose: bool = argument('-v', '--verbose')
...     # change default value
...     ANIMAL: str = as_argument(ExampleOptions.animal).with_options(default='YW00')
...     # and disable an option
...     OUTPUT_DIR: str = 'output' # just replace with a value

Change options name is more complicate, because you might want to add more name, remove some name, or rename some name. with_options() allow you to do that:

>>> class ChangeExample(ExampleOptions):
...     # replace option name: --animal
...     ANIMAL: str = as_argument(ExampleOptions.animal).with_options('--animal')
...     # add more option name: -A, --ANIMAL
...     ANIMAL: str = as_argument(ExampleOptions.animal).with_options(..., '-A')
...     # rename option: --animal
...     ANIMAL: str = as_argument(ExampleOptions.animal).with_options({
...         '--ANIMAL': '--animal'
...     })

Utility function for option class

ExampleOptions doesn’t declated neither __str__ nor __repr__, so it is not convenient to debug. funciont as_dict() provide a way to take argument attribute’s value into a dictionary.

>>> as_dict(opt)
{'ANIMAL': ..., 'EXP_DATE': ..., 'OUTPUT_DIR': ...}

Option class is not restricted into only one use case. It works like a normal class.

class AbstractParser

This class provide a main like class that has more control on argparse.ArgumentParser creation.

>>> class ExampleParser(AbstractParser, ExampleOptions):
...     DESCRIPTION = 'Example parser'
...     def run(self):
...         ...
>>> if __name__ == '__main__':
...     ExampleParser().main()

Subcommands

This module isn’t fully support sub-command feature, but only provide a simple way for specific case:

>>> parse_command_args(
...     description='top level parser',
...     parsers=dict(ep=ExampleParser)
... )

In bash:

python -m module.path ep --ANIMAL name --EXP_DATE date output

Modules: