A Wadler–Lindig ✨pretty-printer✨ for Python

This library is for you if you need:

Something like the built-in pprint.pprint, but which consumes less horizontal space. For example in error messages.
If you have complicated custom types that you'd like to create pretty well-formatted reprs for. For example nested trees of dataclasses / PyTorch modules / etc.

Main features:

Absolutely tiny implementation (77 lines of code for the main Wadler–Lindig algorithm, 223 more for teaching it how to handle all Python types).
Simpler than the original algorithm by Wadler & Lindig (removes some dead code).
Supports multi-line unbroken text strings.
Supports ANSI escape codes and colours.
Zero dependencies.

Installation

pip install wadler_lindig

Documentation

Available at https://docs.kidger.site/wadler_lindig.

Example

import dataclasses
import numpy as np
import wadler_lindig as wl

@dataclasses.dataclass
class MyDataclass:
    x: list[str]
    y: np.ndarray

obj = MyDataclass(["lorem", "ipsum", "dolor sit amet"], np.zeros((2, 3)))

wl.pprint(obj, width=30, indent=4)
# MyDataclass(
#     x=[
#         'lorem',
#         'ipsum',
#         'dolor sit amet'
#     ],
#     y=f64[2,3](numpy)
# )

API at a glance

For day-to-day pretty-printing objects: pprint (to stdout), pformat (as a string), pdiff (between two objects).

For creating custom pretty-printed representations:

The core Wadler–Lindig document types: AbstractDoc, BreakDoc, ConcatDoc, GroupDoc, NestDoc, TextDoc.
pdoc will convert any Python object to a Wadler–Lindig document, with the __pdoc__ method called on custom types if it is available.
ansi_format (to add ANSI colour codes), ansi_strip (to remove ANSI codes).
Several common helpers for creating Wadler–Lindig documents: array_summary, bracketed, comma, join, named_objs.

FAQ

What is the difference to the built-in `pprint` library?

The main difference is that the Wadler–Lindig algorithm produces output like

MyDataclass(
  x=SomeNestedClass(
    y=[1, 2, 3]
  )
)

In contrast pprint produces output like

MyDataclass(x=SomeNestedClass(y=[1,
                                 2,
                                 3]))

which consumes a lot more horizontal space.

By default we print NumPy arrays / PyTorch tensors / etc. in a concise form e.g. f32[2,3](numpy) to denote a NumPy array with shape (2, 3) and dtype float32. (Set short_arrays=False to disable this.)

We provide support for customising the pretty-printed representations of your custom types. Typically this is done via:

import wadler_lindig as wl

class MyAmazingClass:
    def __pdoc__(self, **kwargs) -> wl.AbstractDoc:
        ...  # Create your pretty representation here!

    def __repr__(self):
        # Calls `__pdoc__` and then formats to a particular width.
        return wl.pformat(self, width=80)

In addition we support a wadler_lindig.pprint(..., custom=...) argument, if you don't own the type and so cannot add a __pdoc__ method.

What is the difference to `black` or `rust format`?

The above are formatters for your source code. This wadler_lindig library is intended as an alternative to the built-in pprint library, which pretty-format Python objects at runtime.