In Python's format mini-language, what does the spec in f"{total:>12,.2f}" mean?

### The grammar Inside `{...}`, a colon separates the value from its **format spec**. The spec is a small, fixed-order mini-language: ``` [[fill]align][sign][z][#][0][width][grouping][.precision][type] ``` Every part is optional, but the order is not negotiable — `.2f>12` is not a rearranged `>12.2f`, it is an error. Reading `>12,.2f` left to right gives: align right (`>`) with the implicit space fill, minimum width 12, `,` as the thousands separator, two digits of precision, fixed-point presentation. ### Alignment and fill The alignment characters are `<` (left), `>` (right), `^` (centre) and `=`, which puts the padding *between the sign and the digits* and is only meaningful for numbers. Any single character placed before the alignment character becomes the fill: `f"{name:.<20}"` produces a dotted leader. The defaults are the part people get wrong. With a width but no explicit alignment, **numeric types right-align and everything else left-aligns**, which is why a column of mixed `str` and `int` values built from `{:10}` comes out ragged. A leading `0` before the width (`{:08.2f}`) is shorthand for zero-fill with `=` alignment, so the sign stays at the far left. **Width is a minimum, not a maximum.** A value wider than the field is printed in full and the row is pushed out of shape; if you need truncation you must use `.precision` on a string (`{:.10}` keeps ten characters) or slice the value yourself. ### Sign, grouping and precision `+` shows a sign on positive numbers too, `-` (the default) only on negatives, and a space puts a blank where the sign would be — useful for keeping columns aligned. `z` (added in 3.11) coerces a negative zero to positive zero, so `-0.001` at one decimal prints `0.0` rather than `-0.0`. Grouping is `,` for the conventional thousands separator or `_` for an underscore; with `b`, `o` and `x` the underscore groups four digits at a time, which is how you make a bitmask readable. `.precision` means different things per type: digits after the decimal point for `f` and `%`, total significant digits for `g` and `e`, and a maximum character count for `s`. ### The type character `d` for decimal integers, `f` for fixed-point, `e` for scientific, `g` for the general form that switches between them, `%` to multiply by 100 and append a percent sign, `b`/`o`/`x`/`X` for binary, octal and hex, `n` for a locale-aware number, and `s` for strings. `#` adds the base prefix, so `f"{255:#06x}"` gives `0x00ff` — note that the `0x` counts toward the width. ### The spec is not universal — it belongs to the type This is the mechanism worth stating in an interview: both f-strings and `str.format` take the text after the colon and hand it, unparsed, to the value's `__format__`. `object.__format__` accepts only an empty spec; the numeric and string types implement the mini-language above; a `datetime.date` interprets the whole spec as a strftime pattern, which is why `f"{today:%Y-%m-%d}"` works and looks nothing like the grammar. A class of your own joins in by defining `__format__`, and the spec text it receives is whatever you choose to support. That also explains a common error: `f"{value!r:.2f}"` raises `ValueError`, because `!r` converts the value to a `str` first and `.2f` is not a valid string spec. ### Nesting A replacement field may appear *inside* a spec, one level deep: `f"{total:>{width}.{places}f}"` computes the width and precision at runtime. This is how you align a report whose column widths are only known once the data is loaded, and it works in `str.format` too. ### Practical shape For money, `,.2f` with an explicit width is the workhorse; for percentages, `.1%` beats multiplying by 100 by hand; for keys and identifiers, `{:<24}` plus a fill character produces a readable listing. And when a column of numbers must line up, set the alignment explicitly rather than relying on the per-type default.