Helptext generation

In addition to type annotations, tyro.cli() will also parse docstrings and comments. These are used to automatically generate helptext; see examples for how these end up being formatted.

General callables

For general callables, field helptext is extracted from the corresponding field docstring. Our examples use Google-style docstrings, but ReST, Numpydoc-style and Epydoc docstrings are supported as well. Under the hood, all of these options use docstring_parser.

def main(
    field1: str,
    field2: int = 3,
) -> None:
    """Function, whose arguments will be populated from a CLI interface.

    Args:
        field1: A string field.
        field2: A numeric field, with a default value.
    """
    print(field1, field2)

Dataclasses, TypedDict, NamedTuple

For types defined using class attributes, enumerating each argument list in the class docstring can be cumbersome.

If they are unavailable, tyro.cli() will generate helptext from docstrings and comments on attributes. These are parsed via source code inspection.

(1) Attribute docstrings

As per PEP 257.

@dataclasses.dataclass
class Args:
    field1: str
    """A string field."""
    field2: int = 3
    """A numeric field, with a default value."""

(2) Inline comments

Inline comments can be more succinct than true attribute docstrings.

@dataclasses.dataclass
class Args:
    field1: str  # A string field.
    field2: int = 3  # A numeric field, with a default value.

(3) Preceding comments

These can also be handy for semantically grouping fields, such as the two string fields below.

@dataclasses.dataclass
class Args:
    # String fields.
    field1: str
    field2: str

    # An integer field.
    # Multi-line comments are supported.
    field3: int