Summary

Add support for formatting integers as hexadecimal with the fmt::Debug trait, including when they occur within larger types.

println!("{:02X?}", b"AZaz\0")
[41, 5A, 61, 7A, 00]

Motivation

Sometimes the bits that make up an integer are more meaningful than its purely numerical value. For example, an RGBA color encoded in u32 with 8 bits per channel is easier to understand when shown as 00CC44FF than 13387007.

The std::fmt::UpperHex and std::fmt::LowerHex traits provide hexadecimal formatting through {:X} and {:x} in formatting strings, but they’re only implemented for plain integer types and not other types like slices that might contain integers.

The std::fmt::Debug trait (used with {:?}) however is intended for formatting “in a programmer-facing, debugging context”. It can be derived, and doing so is recommended for most types.

This RFC proposes adding the missing combination of:

  • Output intended primarily for end-users (Display) v.s. for programmers (Debug)
  • Numbers shown in decimal v.s. hexadecimal

Guide-level explanation

In formatting strings like in the format! and println! macros, the formatting parameters x or X − to select lower-case or upper-case hexadecimal − can now be combined with ? which select the Debug trait.

For example, format!("{:X?}", [65280].first()) returns Some(FF00).

This can also be combined with other formatting parameters. For example, format!("{:02X?}", b"AZaz\0") zero-pads each byte to two hexadecimal digits and return [41, 5A, 61, 7A, 00].

An API returning Vec<u32> might be tested like this:

let return_value = foo(bar);
let expected = &[ /* ... */ ][..];
assert!(return_value == expected, "{:08X?} != {:08X?}", return_value, expected);

Reference-level explanation

Formatting strings

The syntax of formatting strings is specified with a grammar which at the moment is as follows:

format_string := <text> [ maybe-format <text> ] *
maybe-format := '{' '{' | '}' '}' | <format>
format := '{' [ argument ] [ ':' format_spec ] '}'
argument := integer | identifier

format_spec := [[fill]align][sign]['#']['0'][width]['.' precision][type]
fill := character
align := '<' | '^' | '>'
sign := '+' | '-'
width := count
precision := count | '*'
type := identifier | ''
count := parameter | integer
parameter := argument '$'

This RFC adds an optional radix immediately before type:

format_spec := [[fill]align][sign]['#']['0'][width]['.' precision][radix][type]
radix: 'x' | 'X'

Formatter API

Note that x and X are already valid types. They are only interpreted as a radix when the type is ?, since combining them with other types doesn’t make sense.

This radix is exposed indirectly in two additional methods of std::fmt::Formatter:

impl<'a> Formatter<'a> {
    // ...

    /// Based on the radix and type: 16, 10, 8, or 2.
    ///
    /// This is mostly useful in `Debug` impls,
    /// where the trait itself doesn’t imply a radix.
    fn number_radix(&self) -> u32

    /// true for `X` or `E`
    ///
    /// This is mostly useful in `Debug` impls,
    /// where the trait itself doesn’t imply a case.
    fn number_uppercase(&self) -> bool
}

Although the radix and type are separate in the formatting string grammar, they are intentionally conflated in this new API.

Debug impls

The Debug implementation for primitive integer types {u,i}{8,16,32,64,128,size} is modified to defer to LowerHex or UpperHex instead of Display, based on formatter.number_radix() and formatter.number_uppercase(). The alternate # flag is ignored, since it already has a separate meaning for Debug: the 0x prefix is not included.

As of Rust 1.22, impls using the Formatter::debug_* methods do not forward formatting parameters such as width when formatting keys/values/items. Doing so is important for this RFC to be useful. This is fixed by PR #46233.

Drawbacks

The hexadecimal flag in the the Debug trait is superficially redundant with the LowerHex and UpperHex traits. If these traits were not stable yet, we could have considered a more unified design.

Rationale and alternatives

Implementing LowerHex and UpperHex was proposed and rejected in PR #44751.

The status quo is that debugging or testing code that could be a one-liner requires manual Debug impls and/or concatenating the results of separate string formatting operations.

Unresolved questions

  • Should this be extended to octal and binary (as {:o?} and {:b?})? Other formatting types/traits too?
  • Details of the new Formatter API