Remove underscores in numeric literals

[jpy-git]

Is your feature request related to a problem? Please describe.

I've just learned that this

a = 1_2_3_4_5.6_7_8_9

is a valid alternative way of writing

a = 12345.6789

I think the intention is to let people break up large numbers if they want i.e. 1_000_000 instead of 1000000

Describe the solution you'd like

Should we decide on a consistent format for black and remove these underscores?

Describe alternatives you've considered

I've double checked and the AST is the same regardless of the presence of the underscores so we should be able to fix them without changing the AST

import ast
import astpretty

tree = ast.parse("a = 1_2_3_4_5.6_7_8_9")
astpretty.pprint(tree)

Module(
    body=[
        Assign(
            lineno=1,
            col_offset=0,
            end_lineno=1,
            end_col_offset=21,
            targets=[Name(lineno=1, col_offset=0, end_lineno=1, end_col_offset=1, id='a', ctx=Store())],
            value=Constant(lineno=1, col_offset=4, end_lineno=1, end_col_offset=21, value=12345.6789, kind=None),
            type_comment=None,
        ),
    ],
    type_ignores=[],
)

[JelleZijlstra]

We previously did this (inserting underscores every third digit only), but undid it because it was causing issues when we inserted underscores into numeric ids, making them harder to grep.

I suppose we could remove extra underscores in cases like your first example, but I'm not sure that's really worth the complexity.

[jpy-git]

I agree it would be a hassle to try and handle the every third digit logic.

My suggestion was actually just remove all underscores since that's consistent/simple.

Although, if it turns out using these underscores is a common occurrence then I'd be happy to leave this as out of scope for black 👍

[JelleZijlstra]

Well, I'd feel bad about making Black totally remove all usage of a language feature :)

The place where the underscores shine is if you have ONE_MILLION = 1_000_000, because 1000000 makes it really hard to verify the number of zeros is right. I wouldn't want Black to remove underscores in that case.

For reference: #696, #549, #467, #454, #452.

[jpy-git]

haha that's a fair point, in that case I understand why we leave it the users to handle these! Thanks 😄