The docs say:

"The fill character can be any character other than ‘{‘ or ‘}’."

http://docs.python.org/dev/library/string.html#format-specification-mini-language

But:

>>> "{0:\x01>8.2f}".format(12)
'\x01\x01\x0112.00'

whereas:

>>> "{0:\x00>8.2f}".format(12)
'   12.00'

Unsurprisingly (libmpdec is a C library) this also does not work in _decimal. I could add a special case in _decimal.c at the cost of
two additional if statements for all regular use cases.

Is padding with NUL a legitimate use case? IOW, is the slowdown justified?

Is padding with NUL a legitimate use case?

I don't see a good reason to disallow it, and it seems like a fairly plausible need. Numpy, for example, pads strings will NUL bytes when placing a short string in long fixed-width field.

Mark Dickinson <report@bugs.python.org> wrote:

Numpy, for example, pads strings will NUL bytes when placing a short
string in long fixed-width field.

I was hoping to escape the work, but that's quite convincing. ;)

Changing libmpdec doesn't look very appealing, so probably I'll use
"{" as a placeholder for NUL and then rewrite the result.

I worked on a patch that allows NUL padding for longs, floats and complex numbers. It seems NUL was being used as a sentinel in the format string parsing when no padding character is given. It was then replaced with a space in each call to fill_padding if it was NUL.

This looks like a duplicate of bpo-12546.