format() ignores leading 0 when `_` is used

[doko42]

Bug report

Bug description:

seen with 3.11 and 3.12, trunk not tested

>>> i = int.from_bytes(b'11112345', 'little')
>>> hex(i)
'0x3534333231313131'
>>> format(i, '064b')
'0011010100110100001100110011001000110001001100010011000100110001'
>>> format(i, '064_b')
'11_0101_0011_0100_0011_0011_0011_0010_0011_0001_0011_0001_0011_0001_0011_0001'

Expected output for the last line:

'0011_0101_0011_0100_0011_0011_0011_0010_0011_0001_0011_0001_0011_0001_0011_0001'

CPython versions tested on:

3.12

Operating systems tested on:

Linux

[mdickinson]

I think it's working as designed: the 64 there is a minimum width for the output rather than a digit count (or in this case bit count), so it takes into account the underscores. For formatting a number as 64-bit binary with underscores you'd want something like format(i, '079_b'), so that the width takes into account the expected 15 underscores.

[terryjreedy]

Mark is correct: "width is a decimal integer defining the minimum total field width, including any prefixes, separators, and other formatting characters." No bug.

>>> format(i, '078_b')  # 1 leading 0
'011_0101_0011_0100_0011_0011_0011_0010_0011_0001_0011_0001_0011_0001_0011_0001'
>>> format(i, '079_b')  # 2 leading 0s
'0011_0101_0011_0100_0011_0011_0011_0010_0011_0001_0011_0001_0011_0001_0011_0001'
>>> format(i, '080_b')  # leading '0_' makes 81 chars; leading '_' is not allowed
'0_0011_0101_0011_0100_0011_0011_0011_0010_0011_0001_0011_0001_0011_0001_0011_0001'