Support for custom function registration

[jrycw]

Hello team:

I've been considering how we can help users integrate their custom functions into Great Tables more seamlessly. One possible approach is to introduce register() and unregister() methods (just a rough concept for now), like:

def register(*funcs):
    for func in funcs:
        if hasattr(func, "__name__"):
            func_name = func.__name__
            setattr(GT, func_name, func)


def unregister(*funcs):
    for func in funcs:
        if hasattr(func, "__name__"):
            func_name = func.__name__
            delattr(GT, func_name)

Here’s a simple example:

import polars as pl
from great_tables import GT
from great_tables.data import exibble

df = pl.from_pandas(exibble).select("num", "currency")

shape: (8, 2)
┌──────────┬──────────┐
│ num      ┆ currency │
│ ---      ┆ ---      │
│ f64      ┆ f64      │
╞══════════╪══════════╡
│ 0.1111   ┆ 49.95    │
│ 2.222    ┆ 17.95    │
│ 33.33    ┆ 1.39     │
│ 444.4    ┆ 65100.0  │
│ 5550.0   ┆ 1325.81  │
│ null     ┆ 13.255   │
│ 777000.0 ┆ null     │
│ 8.88e6   ┆ 0.44     │
└──────────┴──────────┘

def fmt_double_plus_n_as_integer(
    self: GT,
    columns: "SelectExpr" = None,
    rows: int | list[int] | None = None,
    plus_n: str | int | float = 1,
) -> GT:
    def fmt_fn(x: str | int | float | None, plus_n: str | int | float = plus_n) -> int:
        if x is None:
            return x
        return int(float(x) * 2 + float(plus_n))

    return self.fmt(fns=fmt_fn, columns=columns, rows=rows)

Currently, users can apply custom functions using GT.pipe(), like:

GT(df).pipe(fmt_double_plus_n_as_integer, columns=["num", "currency"])

However, this feels more like a one-off approach.

For users who heavily rely on custom functions—or for third-party package maintainers working with Great Tables—having register() and unregister() methods could offer greater flexibility and convenience.

register(fmt_double_plus_n_as_integer)

GT(df).fmt_double_plus_n_as_integer(columns=["num", "currency"])

One thing to note is that we may need to consider how to improve modern IDE support to enhance autocomplete functionality in this context.