Crate si_scale

source ·
Expand description

Format value with units according to SI (système international d’unités).

Version requirement: rustc 1.50+

si-scale = "0.2"


This crate formats numbers using the SI Scales: from 1 y (yocto, i.e. 1e-24) to 1 Y (Yotta, i.e. 1e24).

It has the same purpose as the great human-repr, but strikes a different balance:

  • this crate yields more terse code at the call sites
  • it gives you more control over the output. As shown later in this page, you can extend it pretty easily to handle throughput, etc. (seriously, see below)
  • but it only operates on numbers, so it does not prevent you from using a function to print meters on a duration value (which human-repr does brilliantly).

Getting started

To use this crate, either use one of the few pre-defined helper functions, or build your own.

Basic example:

use si_scale::helpers::{seconds, seconds3};

let actual = format!("{}", seconds(1.3e-5));
let expected = "13 µs";
assert_eq!(actual, expected);

let actual = format!("{}", seconds3(1.3e-5));
let expected = "13.000 µs";
assert_eq!(actual, expected);

Pre-defined helper functions

The helper functions use the following naming convention:

  • the name indicates the units to use
  • a number suffix indicates the decimal digits for floating points
  • a _ suffix indicates the digits use “thousands grouping”

But that’s up to you to depart from that when writing your own functions.

Currently the helper functions are:

helper fninputoutput
number_()1.234567, 15151.234_567, 1_515
seconds()1.234567e-6, 16e-31.234567 µs, 16 ms
seconds3()1.234567e-6, 16e-31.235 µs, 16.000 ms
bytes()12345671.234567 MB
bytes_()12345671_234_567 B
bytes1()2.3 * 1e122.3 TB
bytes2()2.3 * 1e122.30 TB
bibytes()1024 * 1024 * 1.251.25 MiB
bibytes1()1024 * 1024 * 1.251.3 MiB
bibytes2()1024 * 1024 * 1.251.25 MiB

Custom helper functions - BYOU (bring your own unit)

To define your own format function, use the scale_fn!() macro. All pre-defined helper functions from this crate are defined using this macro.

helper fnmantissaprefix constraintbasegroupingsinputoutput
number_()"{}"UnitOnlyB1000_1.234567, 15151.234_567, 1_515
seconds()"{}"UnitAndBelowB1000none1.234567e-6, 16e-31.234567 µs, 16 ms
seconds3()"{:.3}"UnitAndBelowB1000none1.234567e-6, 16e-31.235 µs, 16.000 ms
bytes()"{}"UnitAndAboveB1000none12345671.234567 MB
bytes_()"{}"UnitOnlyB1000_12345671_234_567 B
bytes1()"{:.1}"UnitAndAboveB1000none2.3 * 1e122.3 TB
bytes2()"{:.2}"UnitAndAboveB1000none2.3 * 1e122.30 TB
bibytes()"{}"UnitAndAboveB1024none1024 * 1024 * 1.251.25 MiB
bibytes1()"{:.1}"UnitAndAboveB1024none1024 * 1024 * 1.251.3 MiB
bibytes2()"{:.2}"UnitAndAboveB1024none1024 * 1024 * 1.251.25 MiB

The additional table columns show the underlying controls.

The “mantissa” column

It is a format string which only acts on the mantissa after scaling. For instance, "{}" will display the value with all its digits or no digits if it is round, and "{:.1}" for instance will always display one decimal.

The “prefix constraint” column

In a nutshell, this allows values to be represented in unsurprising scales: for instance, you would never write 1.2 ksec, but always 1200 sec or 1.2e3 sec. In the same vein, you would never write 2 mB, but always 0.002 B or 2e-3 B.

So, here the term “unit” refers to the unit scale (1), and has nothing to do with units of measurements. It constrains the possible scales for a value:

  • UnitOnly means the provided value won’t be scaled: if you provide a value larger than 1000, say 1234, it will be printed as 1234.
  • UnitAndAbove means the provided value can only use higher scales, for instance 16 GB but never 4.3 µB.
  • UnitAndBelow means the provided value can only use lower scales, for instance 1.3 µsec but not 16 Gsec.

The “base” column

Base B1000 means 1k = 1000, the base B1024 means 1k = 1024. This is defined in an IEC document. If you set the base to B1024, the mantissa will be scaled appropriately, but in most cases, you will be using B1000.

The “groupings” column

Groupings refer to “thousands groupings”; the provided char will be used (for instance 1234 is displayed as 1_234), if none, the value is displayed 1234.

Example - how to define a helper for kibits/s

For instance, let’s define a formatting function for bits per sec which prints the mantissa with 2 decimals, and also uses base 1024 (where 1 ki = 1024). Note that although we define the function in a separate module, this is not a requirement.

mod unit_fmt {
    use si_scale::scale_fn;
    use si_scale::prelude::Value;

    // defines the `bits_per_sec()` function
              base: B1024,
              constraint: UnitAndAbove,
              mantissa_fmt: "{:.2}",
              groupings: '_',
              unit: "bit/s",
              doc: "Return a string with the value and its si-scaled unit of bit/s.");

use unit_fmt::bits_per_sec;

fn main() {
    let x = 2.1 * 1024 as f32;
    let actual = format!("throughput: {:>15}", bits_per_sec(x));
    let expected = "throughput:    2.10 kibit/s";
    assert_eq!(actual, expected);

    let x = 2;
    let actual = format!("throughput: {}", bits_per_sec(x));
    let expected = "throughput: 2.00 bit/s";
    assert_eq!(actual, expected);

You can omit the groupings argument of the macro to not separate thousands.

SI Scales - Developer doc

With base = 1000, 1k = 1000, 1M = 1_000_000, 1m = 0.001, 1µ = 0.000_001, etc.

min (incl.)max (excl.)magnitudeprefix

The base is usually 1000, but can also be 1024 (bibytes).

With base = 1024, 1ki = 1024, 1Mi = 1024 * 1024, etc.

API overview

The central representation is the Value type, which holds

  • the mantissa,
  • the SI unit prefix (such as “kilo”, “Mega”, etc),
  • and the base which represents the cases where “1 k” means 1000 (most common) and the cases where “1 k” means 1024 (for kiB, MiB, etc).

This crate provides 2 APIs: a low-level API, and a high-level API for convenience.

For the low-level API, the typical use case is

  • first parse a number into a Value. For doing this, you have to specify the base, and maybe some constraint on the SI scales. See Value::new() and Value::new_with()

  • then display the Value either by yourself formatting the mantissa and prefix (implements the fmt::Display trait), or using the provided Formatter.

For the high-level API, the typical use cases are

  1. parse and display a number using the provided functions such as bibytes(), bytes() or seconds(), they will choose for each number the most appropriate SI scale.

  2. In case you want the same control granularity as the low-level API (e.g. constraining the scale in some way, using some base, specific mantissa formatting), then you can build a custom function using the provided macro scale_fn!(). The existing functions such as bibytes(), bytes(), seconds() are all built using this same macro.

The high-level API

The seconds3() function parses a number into a Value and displays it using 3 decimals and the appropriate scale for seconds (UnitAndBelow), so that non-sensical scales such as kilo-seconds can’t be output. The seconds() function does the same but formats the mantissa with the default "{}", so no decimals are printed for integer mantissa.

use si_scale::helpers::{seconds, seconds3};

let actual = format!("result is {:>15}", seconds(1234.5678));
let expected = "result is     1234.5678 s";
assert_eq!(actual, expected);

let actual = format!("result is {:>10}", seconds3(12.3e-7));
let expected = "result is   1.230 µs";
assert_eq!(actual, expected);

The bytes() function parses a number into a Value using base 1000 and displays it using 1 decimal and the appropriate scale for bytes (UnitAndAbove), so that non-sensical scales such as milli-bytes may not appear.

use si_scale::helpers::{bytes, bytes1};

let actual = format!("result is {}", bytes1(12_345_678));
let expected = "result is 12.3 MB";
assert_eq!(actual, expected);

let actual = format!("result is {:>10}", bytes(16));
let expected = "result is       16 B";
assert_eq!(actual, expected);

let actual = format!("result is {}", bytes(0.12));
let expected = "result is 0.12 B";
assert_eq!(actual, expected);

The bibytes1() function parses a number into a Value using base 1024 and displays it using 1 decimal and the appropriate scale for bytes (UnitAndAbove), so that non-sensical scales such as milli-bytes may not appear.

use si_scale::helpers::{bibytes, bibytes1};

let actual = format!("result is {}", bibytes1(12_345_678));
let expected = "result is 11.8 MiB";
assert_eq!(actual, expected);

let actual = format!("result is {}", bibytes(16 * 1024));
let expected = "result is 16 kiB";
assert_eq!(actual, expected);

let actual = format!("result is {:>10}", bibytes1(16));
let expected = "result is     16.0 B";
assert_eq!(actual, expected);

let actual = format!("result is {}", bibytes(0.12));
let expected = "result is 0.12 B";
assert_eq!(actual, expected);

The low-level API

Creating a Value with Value::new()

The low-level function Value::new() converts any number convertible to f64 into a Value using base 1000. The Value struct implements From for common numbers and delegates to Value::new(), so they are equivalent in practice. Here are a few examples.

use std::convert::From;
use si_scale::prelude::*;

let actual = Value::from(0.123);
let expected = Value {
    mantissa: 123f64,
    prefix: Prefix::Milli,
    base: Base::B1000,
assert_eq!(actual, expected);
assert_eq!(Value::new(0.123), expected);

let actual: Value = 0.123.into();
assert_eq!(actual, expected);

let actual: Value = 1300i32.into();
let expected = Value {
    mantissa: 1.3f64,
    prefix: Prefix::Kilo,
    base: Base::B1000,
assert_eq!(actual, expected);

let actual: Vec<Value> = vec![0.123f64, -1.5e28]
    .iter().map(|n| n.into()).collect();
let expected = vec![
    Value {
        mantissa: 123f64,
        prefix: Prefix::Milli,
        base: Base::B1000,
    Value {
        mantissa: -1.5e4f64,
        prefix: Prefix::Yotta,
        base: Base::B1000,
assert_eq!(actual, expected);

As you can see in the last example, values which scale are outside of the SI prefixes are represented using the closest SI prefix.

Creating a Value with Value::new_with()

The low-level Value::new_with() operates similarly to Value::new() but also expects a base and a constraint on the scales you want to use. In comparison with the simple Value::new(), this allows base 1024 scaling (for kiB, MiB, etc) and preventing upper scales for seconds or lower scales for integral units such as bytes (e.g. avoid writing 1300 sec as 1.3 ks or 0.415 B as 415 mB).

use si_scale::prelude::*;

// Assume this is seconds, no kilo-seconds make sense.
let actual = Value::new_with(1234, Base::B1000, Constraint::UnitAndBelow);
let expected = Value {
    mantissa: 1234f64,
    prefix: Prefix::Unit,
    base: Base::B1000,
assert_eq!(actual, expected);

Don’t worry yet about the verbosity, the following parser helps with this.

Formatting values

In this example, the number x is converted into a value and displayed using the most appropriate SI prefix. The user chose to constrain the prefix to be anything lower than Unit (1) because kilo-seconds make no sense.

use si_scale::format_value;
use si_scale::{value::Value, base::Base, prefix::Constraint};

let x = 1234.5678;
let v = Value::new_with(x, Base::B1000, Constraint::UnitAndBelow);
let unit = "s";

let actual = format!(
    "result is {}{u}",
    format_value!(v, "{:.5}", groupings: '_'),
    u = unit
let expected = "result is 1_234.567_80 s";
assert_eq!(actual, expected);

Run code-coverage

Install the llvm-tools-preview component and grcov

rustup component add llvm-tools-preview
cargo install grcov

Install nightly

rustup toolchain install nightly

The following make invocation will switch to nigthly run the tests using Cargo, and output coverage HTML report in ./coverage/

make coverage

The coverage report is located in ./coverage/index.html


Licensed under either of

at your option.


Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.


  • Defines the Base struct and methods.
  • The format_value macro.
  • The helpers functions provide number parsing and correct SI formatting for various units. They are probably the most used functions in this crate.
  • The prefix is added to the unit string representation, such as µ in µs, and requires the mantissa to be scaled accordingly.
  • Holds first-class citizens of this crate, for convenience.
  • Represents a float value using its mantissa and unit Prefix in a base.


  • Formats a Value’s mantissa and unit prefix (but not the unit itself). Because it simply delegates to format_args!(), the output should be consumed by macros such as println!(), write!(), etc.
  • Three nearly identical variants: with the unit argument only, with unit and groupings arguments, with groupings argument only. If you happen to know how to factor this, please make a suggestion!


Type Aliases

  • Result type used by this crate.