Tap is a typed modernization of Python's argparse library.
Tap provides the following benefits:
- Static type checking
- Code completion
- Source code navigation (e.g. go to definition and go to implementation)
See this poster, which we presented at PyCon 2020, for a presentation of some of the relevant concepts we used to guide the development of Tap.
As of version 1.8.0, Tap includes tapify, which runs functions or initializes classes with arguments parsed from the command line. We show an example below.
# square.py
from tap import tapify
def square(num: float) -> float:
return num ** 2
if __name__ == '__main__':
print(f'The square of your number is {tapify(square)}.')Running python square.py --num 2 will print The square of your number is 4.0.. Please see tapify for more details.
Tap requires Python 3.9+
To install Tap from PyPI run:
pip install typed-argument-parser
To install Tap from source, run the following commands:
git clone https://github.com/swansonk14/typed-argument-parser.git
cd typed-argument-parser
pip install -e .
To develop this package, install development requirements (in a virtual environment):
python -m pip install -e ".[dev]"
Style:
To run tests, run:
pytest
To see this, let's look at an example:
"""main.py"""
from tap import Tap
class SimpleArgumentParser(Tap):
name: str # Your name
language: str = 'Python' # Programming language
package: str = 'Tap' # Package name
stars: int # Number of stars
max_stars: int = 5 # Maximum stars
args = SimpleArgumentParser().parse_args()
print(f'My name is {args.name} and I give the {args.language} package '
f'{args.package} {args.stars}/{args.max_stars} stars!')You use Tap the same way you use standard argparse.
>>> python main.py --name Jesse --stars 5
My name is Jesse and I give the Python package Tap 5/5 stars!
The equivalent argparse code is:
"""main.py"""
from argparse import ArgumentParser
parser = ArgumentParser()
parser.add_argument('--name', type=str, required=True,
help='Your name')
parser.add_argument('--language', type=str, default='Python',
help='Programming language')
parser.add_argument('--package', type=str, default='Tap',
help='Package name')
parser.add_argument('--stars', type=int, required=True,
help='Number of stars')
parser.add_argument('--max_stars', type=int, default=5,
help='Maximum stars')
args = parser.parse_args()
print(f'My name is {args.name} and I give the {args.language} package '
f'{args.package} {args.stars}/{args.max_stars} stars!')The advantages of being Python-native include being able to:
- Overwrite convenient built-in methods (e.g.
process_argsensures consistency among arguments) - Add custom methods
- Inherit from your own template classes
Now we are going to highlight some of our favorite features and give examples of how they work in practice.
Arguments are specified as class variables defined in a subclass of Tap. Variables defined as name: type are required arguments while variables defined as name: type = value are not required and default to the provided value.
class MyTap(Tap):
required_arg: str
default_arg: str = 'default value'Single line and/or multiline comments which appear after the argument are automatically parsed into the help string provided when running python main.py -h. The type and default values of arguments are also provided in the help string.
"""main.py"""
from tap import Tap
class MyTap(Tap):
x: float # What am I?
pi: float = 3.14 # I'm pi!
"""Pi is my favorite number!"""
args = MyTap().parse_args()Running python main.py -h results in the following:
>>> python main.py -h
usage: demo.py --x X [--pi PI] [-h]
optional arguments:
--x X (float, required) What am I?
--pi PI (float, default=3.14) I'm pi! Pi is my favorite number.
-h, --help show this help message and exit
To specify behavior beyond what can be specified using arguments as class variables, override the configure method.
configure provides access to advanced argument parsing features such as add_argument and add_subparser.
Since Tap is a wrapper around argparse, Tap provides all of the same functionality.
We detail these two functions below.
In the configure method, call self.add_argument just as you would use argparse's add_argument. For example,
from tap import Tap
class MyTap(Tap):
positional_argument: str
list_of_three_things: List[str]
argument_with_really_long_name: int
def configure(self):
self.add_argument('positional_argument')
self.add_argument('--list_of_three_things', nargs=3)
self.add_argument('-arg', '--argument_with_really_long_name')To add a subparser, override the configure method and call self.add_subparser. Optionally, to specify keyword arguments (e.g., help) to the subparser collection, call self.add_subparsers. For example,
class SubparserA(Tap):
bar: int # bar help
class SubparserB(Tap):
baz: Literal['X', 'Y', 'Z'] # baz help
class Args(Tap):
foo: bool = False # foo help
def configure(self):
self.add_subparsers(help='sub-command help')
self.add_subparser('a', SubparserA, help='a help')
self.add_subparser('b', SubparserB, help='b help')Tap automatically handles all the following types:
str, int, float, bool
Optional, Optional[str], Optional[int], Optional[float], Optional[bool]
List, List[str], List[int], List[float], List[bool]
Set, Set[str], Set[int], Set[float], Set[bool]
Tuple, Tuple[Type1, Type2, etc.], Tuple[Type, ...]
LiteralIf you're using Python 3.9+, then you can replace List with list, Set with set, and Tuple with tuple.
Tap also supports Union, but this requires additional specification (see Union section below).
Additionally, any type that can be instantiated with a string argument can be used. For example, in
from pathlib import Path
from tap import Tap
class Args(Tap):
path: Path
args = Args().parse_args()args.path is a Path instance containing the string passed in through the command line.
Each is automatically parsed to their respective types, just like argparse.
If an argument arg is specified as arg: bool or arg: bool = False, then adding the --arg flag to the command line will set arg to True. If arg is specified as arg: bool = True, then adding --arg sets arg to False.
Note that if the Tap instance is created with explicit_bool=True, then booleans can be specified on the command line as --arg True or --arg False rather than --arg. Additionally, booleans can be specified by prefixes of True and False with any capitalization as well as 1 or 0 (e.g. for True, --arg tRu, --arg T, --arg 1 all suffice).
These arguments are parsed in exactly the same way as str, int, float, and bool. Note bools can be specified using the same rules as above and that Optional is equivalent to Optional[str].
If an argument arg is a List, simply specify the values separated by spaces just as you would with regular argparse. For example, --arg 1 2 3 parses to arg = [1, 2, 3].
Identical to List but parsed into a set rather than a list.
Tuples can be used to specify a fixed number of arguments with specified types using the syntax Tuple[Type1, Type2, etc.] (e.g. Tuple[str, int, bool, str]). Tuples with a variable number of arguments are specified by Tuple[Type, ...] (e.g. Tuple[int, ...]). Note Tuple defaults to Tuple[str, ...].
Literal is analagous to argparse's choices, which specifies the values that an argument can take. For example, if arg can only be one of 'H', 1, False, or 1.0078 then you would specify that arg: Literal['H', 1, False, 1.0078]. For instance, --arg False assigns arg to False and --arg True throws error.
Union types must include the type keyword argument in add_argument in order to specify which type to use, as in the example below.
def to_number(string: str) -> Union[float, int]:
return float(string) if '.' in string else int(string)
class MyTap(Tap):
number: Union[float, int]
def configure(self):
self.add_argument('--number', type=to_number)In Python 3.10+, Union[Type1, Type2, etc.] can be replaced with Type1 | Type2 | etc., but the type keyword argument must still be provided in add_argument.
Tap can also support more complex types than the ones specified above. If the desired type is constructed with a single string as input, then the type can be specified directly without additional modifications. For example,
class Person:
def __init__(self, name: str) -> None:
self.name = name
class Args(Tap):
person: Person
args = Args().parse_args('--person Tapper'.split())
print(args.person.name) # TapperIf the desired type has a more complex constructor, then the type keyword argument must be provided in add_argument. For example,
class AgedPerson:
def __init__(self, name: str, age: int) -> None:
self.name = name
self.age = age
def to_aged_person(string: str) -> AgedPerson:
name, age = string.split(',')
return AgedPerson(name=name, age=int(age))
class Args(Tap):
aged_person: AgedPerson
def configure(self) -> None:
self.add_argument('--aged_person', type=to_aged_person)
args = Args().parse_args('--aged_person Tapper,27'.split())
print(f'{args.aged_person.name} is {args.aged_person.age}') # Tapper is 27With complex argument parsing, arguments often end up having interdependencies. This means that it may be necessary to disallow certain combinations of arguments or to modify some arguments based on other arguments.
To handle such cases, simply override process_args and add the required logic. process_args is automatically called when parse_args is called.
class MyTap(Tap):
package: str
is_cool: bool
stars: int
def process_args(self):
# Validate arguments
if self.is_cool and self.stars < 4:
raise ValueError('Cool packages cannot have fewer than 4 stars')
# Modify arguments
if self.package == 'Tap':
self.is_cool = True
self.stars = 5Similar to argparse's parse_known_args, Tap is capable of parsing only arguments that it is aware of without raising an error due to additional arguments. This can be done by calling parse_args with known_only=True. The remaining un-parsed arguments are then available by accessing the extra_args field of the Tap object.
class MyTap(Tap):
package: str
args = MyTap().parse_args(['--package', 'Tap', '--other_arg', 'value'], known_only=True)
print(args.extra_args) # ['--other_arg', 'value']It is sometimes useful to define a template Tap and then subclass it for different use cases. Since Tap is a native Python class, inheritance is built-in, making it easy to customize from a template Tap.
In the example below, StarsTap and AwardsTap inherit the arguments (package and is_cool) and the methods (process_args) from BaseTap.
class BaseTap(Tap):
package: str
is_cool: bool
def process_args(self):
if self.package == 'Tap':
self.is_cool = True
class StarsTap(BaseTap):
stars: int
class AwardsTap(BaseTap):
awards: List[str]Tap uses Python's pretty printer to print out arguments in an easy-to-read format.
"""main.py"""
from tap import Tap
from typing import List
class MyTap(Tap):
package: str
is_cool: bool = True
awards: List[str] = ['amazing', 'wow', 'incredible', 'awesome']
args = MyTap().parse_args()
print(args)Running python main.py --package Tap results in:
>>> python main.py
{'awards': ['amazing', 'wow', 'incredible', 'awesome'],
'is_cool': True,
'package': 'Tap'}
Tap makes reproducibility easy, especially when running code in a git repo.
Specifically, Tap has a method called get_reproducibility_info that returns a dictionary containing all the information necessary to replicate the settings under which the code was run. This dictionary includes:
- Python command
- The Python command that was used to run the program
- Ex.
python main.py --package Tap
- Time
- The time when the command was run
- Ex.
Thu Aug 15 00:09:13 2019
- Git root
- The root of the git repo containing the code that was run
- Ex.
/Users/swansonk14/typed-argument-parser
- Git url
- The url to the git repo, specifically pointing to the current git hash (i.e. the hash of HEAD in the local repo)
- Ex. https://github.com/swansonk14/typed-argument-parser/tree/446cf046631d6bdf7cab6daec93bf7a02ac00998
- Uncommitted changes
- Whether there are any uncommitted changes in the git repo (i.e. whether the code is different from the code at the above git hash)
- Ex.
TrueorFalse
Tap has methods as_dict and from_dict that convert Tap objects to and from dictionaries.
For example,
"""main.py"""
from tap import Tap
class Args(Tap):
package: str
is_cool: bool = True
stars: int = 5
args = Args().parse_args(["--package", "Tap"])
args_data = args.as_dict()
print(args_data) # {'package': 'Tap', 'is_cool': True, 'stars': 5}
args_data['stars'] = 2000
args = args.from_dict(args_data)
print(args.stars) # 2000 Note that as_dict does not include attributes set directly on an instance (e.g., arg is not included even after setting args.arg = "hi" in the code above because arg is not an attribute of the Args class).
Also note that from_dict ensures that all required arguments are set.
Tap has a method called save which saves all arguments, along with the reproducibility info, to a JSON file.
"""main.py"""
from tap import Tap
class MyTap(Tap):
package: str
is_cool: bool = True
stars: int = 5
args = MyTap().parse_args()
args.save('args.json')After running python main.py --package Tap, the file args.json will contain:
{
"is_cool": true,
"package": "Tap",
"reproducibility": {
"command_line": "python main.py --package Tap",
"git_has_uncommitted_changes": false,
"git_root": "/Users/swansonk14/typed-argument-parser",
"git_url": "https://github.com/swansonk14/typed-argument-parser/tree/446cf046631d6bdf7cab6daec93bf7a02ac00998",
"time": "Thu Aug 15 00:18:31 2019"
},
"stars": 5
}
Note: More complex types will be encoded in JSON as a pickle string.
âť—
⚠️
Never callargs.load('args.json')on untrusted files. Argument loading uses thepicklemodule to decode complex types automatically. Unpickling of untrusted data is a security risk and can lead to arbitrary code execution. See the warning in the pickle docs.
❗⚠️
Arguments can be loaded from a JSON file rather than parsed from the command line.
"""main.py"""
from tap import Tap
class MyTap(Tap):
package: str
is_cool: bool = True
stars: int = 5
args = MyTap()
args.load('args.json')Note: All required arguments (in this case package) must be present in the JSON file if not already set in the Tap object.
Arguments can be loaded from a Python dictionary rather than parsed from the command line.
"""main.py"""
from tap import Tap
class MyTap(Tap):
package: str
is_cool: bool = True
stars: int = 5
args = MyTap()
args.from_dict({
'package': 'Tap',
'stars': 20
})Note: As with load, all required arguments must be present in the dictionary if not already set in the Tap object. All values in the provided dictionary will overwrite values currently in the Tap object.
Configuration files can be loaded along with arguments with the optional flag config_files: List[str]. Arguments passed in from the command line overwrite arguments from the configuration files. Arguments in configuration files that appear later in the list overwrite the arguments in previous configuration files.
For example, if you have the config file my_config.txt
--arg1 1
--arg2 two
then you can write
from tap import Tap
class Args(Tap):
arg1: int
arg2: str
args = Args(config_files=['my_config.txt']).parse_args()Config files are parsed using shlex.split from the python standard library, which supports shell-style string quoting, as well as line-end comments starting with #.
For example, if you have the config file my_config_shlex.txt
--arg1 21 # Important arg value
# Multi-word quoted string
--arg2 "two three four"
then you can write
from tap import Tap
class Args(Tap):
arg1: int
arg2: str
args = Args(config_files=['my_config_shlex.txt']).parse_args()to get the resulting args = {'arg1': 21, 'arg2': 'two three four'}
The legacy parsing behavior of using standard string split can be re-enabled by passing legacy_config_parsing=True to parse_args.
tapify makes it possible to run functions or initialize objects via command line arguments. This is inspired by Google's Python Fire, but tapify also automatically casts command line arguments to the appropriate types based on the type hints. Under the hood, tapify implicitly creates a Tap object and uses it to parse the command line arguments, which it then uses to run the function or initialize the class. We show a few examples below.
# square_function.py
from tap import tapify
def square(num: float) -> float:
"""Square a number.
:param num: The number to square.
"""
return num ** 2
if __name__ == '__main__':
squared = tapify(square)
print(f'The square of your number is {squared}.')Running python square_function.py --num 5 prints The square of your number is 25.0..
# square_class.py
from tap import tapify
class Squarer:
def __init__(self, num: float) -> None:
"""Initialize the Squarer with a number to square.
:param num: The number to square.
"""
self.num = num
def get_square(self) -> float:
"""Get the square of the number."""
return self.num ** 2
if __name__ == '__main__':
squarer = tapify(Squarer)
print(f'The square of your number is {squarer.get_square()}.')Running python square_class.py --num 2 prints The square of your number is 4.0..
# square_dataclass.py
from dataclasses import dataclass
from tap import tapify
@dataclass
class Squarer:
"""Squarer with a number to square.
:param num: The number to square.
"""
num: float
def get_square(self) -> float:
"""Get the square of the number."""
return self.num ** 2
if __name__ == '__main__':
squarer = tapify(Squarer)
print(f'The square of your number is {squarer.get_square()}.')Running python square_dataclass.py --num -1 prints The square of your number is 1.0..
Argument descriptions
For dataclasses, the argument's description (which is displayed in the -h help message) can either be specified in the
class docstring or the field's description in metadata. If both are specified, the description from the docstring is
used. In the example below, the description is provided in metadata.
# square_dataclass.py
from dataclasses import dataclass, field
from tap import tapify
@dataclass
class Squarer:
"""Squarer with a number to square.
"""
num: float = field(metadata={"description": "The number to square."})
def get_square(self) -> float:
"""Get the square of the number."""
return self.num ** 2
if __name__ == '__main__':
squarer = tapify(Squarer)
print(f'The square of your number is {squarer.get_square()}.')Pydantic Models and
dataclasses can be tapifyd.
# square_pydantic.py
from pydantic import BaseModel, Field
from tap import tapify
class Squarer(BaseModel):
"""Squarer with a number to square.
"""
num: float = Field(description="The number to square.")
def get_square(self) -> float:
"""Get the square of the number."""
return self.num ** 2
if __name__ == '__main__':
squarer = tapify(Squarer)
print(f'The square of your number is {squarer.get_square()}.')Argument descriptions
For Pydantic v2 models and dataclasses, the argument's description (which is displayed in the -h help message) can
either be specified in the class docstring or the field's description. If both are specified, the description from the
docstring is used. In the example below, the description is provided in the docstring.
For Pydantic v1 models and dataclasses, the argument's description must be provided in the class docstring:
# square_pydantic.py
from pydantic import BaseModel
from tap import tapify
class Squarer(BaseModel):
"""Squarer with a number to square.
:param num: The number to square.
"""
num: float
def get_square(self) -> float:
"""Get the square of the number."""
return self.num ** 2
if __name__ == '__main__':
squarer = tapify(Squarer)
print(f'The square of your number is {squarer.get_square()}.')The help string on the command line is set based on the docstring for the function or class. For example, running python square_function.py -h will print:
usage: square_function.py [-h] --num NUM
Square a number.
options:
-h, --help show this help message and exit
--num NUM (float, required) The number to square.
Note that for classes, if there is a docstring in the __init__ method, then tapify sets the help string description to that docstring. Otherwise, it uses the docstring from the top of the class.
tapify can simultaneously use both arguments passed from the command line and arguments passed in explicitly in the tapify call. Arguments provided in the tapify call override function defaults, and arguments provided via the command line override both arguments provided in the tapify call and function defaults. We show an example below.
# add.py
from tap import tapify
def add(num_1: float, num_2: float = 0.0, num_3: float = 0.0) -> float:
"""Add numbers.
:param num_1: The first number.
:param num_2: The second number.
:param num_3: The third number.
"""
return num_1 + num_2 + num_3
if __name__ == '__main__':
added = tapify(add, num_2=2.2, num_3=4.1)
print(f'The sum of your numbers is {added}.')Running python add.py --num_1 1.0 --num_2 0.9 prints The sum of your numbers is 6.0.. (Note that add took num_1 = 1.0 and num_2 = 0.9 from the command line and num_3=4.1 from the tapify call due to the order of precedence.)
Calling tapify with known_only=True allows tapify to ignore additional arguments from the command line that are not needed for the function or class. If known_only=False (the default), then tapify will raise an error when additional arguments are provided. We show an example below where known_only=True might be useful for running multiple tapify calls.
# person.py
from tap import tapify
def print_name(name: str) -> None:
"""Print a person's name.
:param name: A person's name.
"""
print(f'My name is {name}.')
def print_age(age: int) -> None:
"""Print a person's age.
:param name: A person's age.
"""
print(f'My age is {age}.')
if __name__ == '__main__':
tapify(print_name, known_only=True)
tapify(print_age, known_only=True)Running python person.py --name Jesse --age 1 prints My name is Jesse. followed by My age is 1.. Without known_only=True, the tapify calls would raise an error due to the extra argument.
Tapify supports explicit specification of boolean arguments (see bool for more details). By default, explicit_bool=False and it can be set with tapify(..., explicit_bool=True).
to_tap_class turns a function or class into a Tap class. The returned class can be subclassed to add
special argument behavior. For example, you can override configure and
process_args.
If the object can be tapifyd, then it can be to_tap_classd, and vice-versa. to_tap_class provides full control
over argument parsing.
# main.py
"""
My script description
"""
from pydantic import BaseModel
from tap import to_tap_class
class Project(BaseModel):
package: str
is_cool: bool = True
stars: int = 5
if __name__ == "__main__":
ProjectTap = to_tap_class(Project)
tap = ProjectTap(description=__doc__) # from the top of this script
args = tap.parse_args()
project = Project(**args.as_dict())
print(f"Project instance: {project}")Running python main.py --package tap will print Project instance: package='tap' is_cool=True stars=5.
The general pattern is:
from tap import to_tap_class
class MyCustomTap(to_tap_class(my_class_or_function)):
# Special argument behavior, e.g., override configure and/or process_argsPlease see demo_data_model.py for an example of overriding configure and
process_args.
