Pretty-print dataclasses prettier

Question

Python Data Classes instances also include a string representation method, but its result isn't really sufficient for pretty printing purposes when classes have more than a few fields and/or longer field values.

Basically I'm looking for a way to customize the default dataclasses string representation routine or for a pretty-printer that understands data classes and prints them prettier.

So, it's just a small customization I have in mind: adding a line break after each field while indenting lines after the first one.

For example, instead of

x = InventoryItem('foo', 23)
print(x) # =>
InventoryItem(name='foo', unit_price=23, quantity_on_hand=0)

I want to get a string representation like this:

x = InventoryItem('foo', 23)
print(x) # =>
InventoryItem(
    name='foo',
    unit_price=23,
    quantity_on_hand=0
)

Or similar. Perhaps a pretty-printer could get even fancier, such as aligning the = assignment characters or something like that.

Of course, it should also work in a recursive fashion, e.g. fields that are also dataclasses should be indented more.

Answer 1

The pprint package supports pretty printing only since version 3.10 (NB: Python 3.10 was released in 2021).

Example:

[ins] In [1]: from dataclasses import dataclass
         ...:
         ...: @dataclass
         ...: class Point:
         ...:     x: int
         ...:     y: int
         ...:
         ...: @dataclass
         ...: class Coords:
         ...:     my_points: list
         ...:     my_dict: dict
         ...:
         ...: coords = Coords([Point(1, 2), Point(3, 4)], {'a': (1, 2), (1, 2): 'a'})

[ins] In [15]: pprint.pprint(coords, width=20)                                  
Coords(my_points=[Point(x=1,
                        y=2),
                  Point(x=3,
                        y=4)],
       my_dict={'a': (1,
                      2),
                (1, 2): 'a'})

---

When using Python 3.9 or older, there is the prettyprinter package that supports dataclasses and provides some nice pretty-printing features.

Example:

[ins] In [1]: from dataclasses import dataclass
         ...:
         ...: @dataclass
         ...: class Point:
         ...:     x: int
         ...:     y: int
         ...:
         ...: @dataclass
         ...: class Coords:
         ...:     my_points: list
         ...:     my_dict: dict
         ...:
         ...: coords = Coords([Point(1, 2), Point(3, 4)], {'a': (1, 2), (1, 2): 'a'})

[nav] In [2]: import prettyprinter as pp

[ins] In [3]: pp.pprint(coords)
Coords(my_points=[Point(x=1, y=2), Point(x=3, y=4)], my_dict={'a': (1, 2), (1, 2): 'a'})

The dataclasses support isn't enabled, by default, thus:

[nav] In [4]: pp.install_extras()
[ins] In [5]: pp.pprint(coords)
Coords(
    my_points=[Point(x=1, y=2), Point(x=3, y=4)],
    my_dict={'a': (1, 2), (1, 2): 'a'}
)

Or to force indenting of all fields:

[ins] In [6]: pp.pprint(coords, width=1)
Coords(
    my_points=[
        Point(
            x=1,
            y=2
        ),
        Point(
            x=3,
            y=4
        )
    ],
    my_dict={
        'a': (
            1,
            2
        ),
        (
            1,
            2
        ): 'a'
    }
)

Prettyprinter can even syntax-highlight! (cf. cpprint())

---

Considerations:

• prettyprinter isn't part of the python standard library
• default values aren't printed, at all and as of 2021 there is no way around this
• prettyprinter is pretty-printing very slowly, i.e. much slower than the standard pprint, e.g. for checking if a value is a default value, it's compared against a default-constructed value

Answer 2

Python 3.10+ Supports pretty printing dataclasses:

Python 3.10.0b2+ (heads/3.10:f807a4fad4, Sep  4 2021, 18:58:04) [GCC 11.1.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from dataclasses import dataclass
>>> @dataclass
... class Literal:
...     value: 'Any'
... 
>>> @dataclass
... class Binary:
...     left: 'Binary | Literal'
...     operator: str
...     right: 'Binary | Literal'
... 
>>> from pprint import pprint
>>> # magic happens here
>>> pprint(
... Binary(Binary(Literal(2), '*', Literal(100)), '+', Literal(50)))
Binary(left=Binary(left=Literal(value=2),
                   operator='*',
                   right=Literal(value=100)),
       operator='+',
       right=Literal(value=50))

Answer 3

We can use dataclasses.fields to recurse through nested dataclasses and pretty print them:

from collections.abc import Mapping, Iterable
from dataclasses import is_dataclass, fields

def pretty_print(obj, indent=4):
    """
    Pretty prints a (possibly deeply-nested) dataclass.
    Each new block will be indented by `indent` spaces (default is 4).
    """
    print(stringify(obj, indent))

def stringify(obj, indent=4, _indents=0):
    if isinstance(obj, str):
        return f"'{obj}'"

    if not is_dataclass(obj) and not isinstance(obj, (Mapping, Iterable)):
        return str(obj)

    this_indent = indent * _indents * ' '
    next_indent = indent * (_indents + 1) * ' '
    start, end = f'{type(obj).__name__}(', ')'  # dicts, lists, and tuples will re-assign this

    if is_dataclass(obj):
        body = '\n'.join(
            f'{next_indent}{field.name}='
            f'{stringify(getattr(obj, field.name), indent, _indents + 1)},' for field in fields(obj)
        )

    elif isinstance(obj, Mapping):
        if isinstance(obj, dict):
            start, end = '{}'

        body = '\n'.join(
            f'{next_indent}{stringify(key, indent, _indents + 1)}: '
            f'{stringify(value, indent, _indents + 1)},' for key, value in obj.items()
        )

    else:  # is Iterable
        if isinstance(obj, list):
            start, end = '[]'
        elif isinstance(obj, tuple):
            start = '('

        body = '\n'.join(
            f'{next_indent}{stringify(item, indent, _indents + 1)},' for item in obj
        )

    return f'{start}\n{body}\n{this_indent}{end}'

We can test it with a nested dataclass:

from dataclasses import dataclass

@dataclass
class Point:
    x: int
    y: int

@dataclass
class Coords:
    my_points: list
    my_dict: dict

coords = Coords([Point(1, 2), Point(3, 4)], {'a': (1, 2), (1, 2): 'a'})

pretty_print(coords)

# Coords(
#     my_points=[
#         Point(
#             x=1,
#             y=2,
#         ),
#         Point(
#             x=3,
#             y=4,
#         ),
#     ],
#     my_dict={
#         'a': (
#             1,
#             2,
#         ),
#         (
#             1,
#             2,
#         ): 'a',
#     },
# )

This should be general enough to cover most cases. Hope this helps!