ferrilab / bitvec

A crate for managing memory bit by bit
https://myrrlyn.net/crates/bitvec
MIT License
1.23k stars 116 forks source link
bitset bitvector rust
# `bitvec` ## A Magnifying Glass for Memory [![Crate][crate_img]][crate_link] [![MSRV][msrv_img]][crate_link] [![License][license_img]][license_file] [![Documentation][docs_img]][docs_link] [![Crate Downloads][downloads_img]][crate_link]
  1. Summary
  2. Introduction
  3. Highlights
  4. Usage
  5. Examples
  6. User Stories
    1. Bit Collections
    2. Bit-Field Memory Access
    3. Transport Protocols
  7. Feature Flags
  8. Deeper Reading

Summary

bitvec provides a foundational API for bitfields in Rust. It specializes standard-library data structures (slices, arrays, and vectors of bool) to use one-bit-per-bool storage, similar to std::bitset<N> and std::vector<bool> in C++.

Additionally, it allows a memory region to be divided into arbitrary regions of integer storage, like binaries in Erlang.

If you need to view memory as bit-addressed instead of byte-addressed, then bitvec is the most capable, complete, and Rust-idiomatic crate for you.

Introduction

Computers do not operate on bits. The memory bus is byte-addressed, and processors operate on register words, which are typically four to eight bytes, or even wider. This means that when programmers wish to operate on individual bits within a byte of memory or a word of register, they have to do so manually, using shift and mask operations that are likely familiar to anyone who has done this before.

bitvec brings the capabilities of C++’s compact bool storage and Erlang’s decomposable bit-streams to Rust, in a package that fits in with your existing Rust idioms and in the most capable, performant, implementation possible. The bit-stream behavior provides the logic necessary for C-style structural bitfields, and syntax sugar for it can be found in deku.

bitvec enables you to write code for bit-addressed memory that is simple, easy, and fast. It compiles to the same, or even better, object code than you would get from writing shift/mask instructions manually. It leverages Rust’s powerful reference and type systems to create a system that seamlessly bridges single-bit addressing, precise control of in-memory layout, and Rust-native ownership and borrowing mechanisms.

Highlights

bitvec has a number of unique capabilities related to its place as a Rust library and as a bit-addressing system.

However, it does also have some small costs associated with its capabilities:

Usage

bitvec strives to follow the sequence APIs in the standard library. However, as most of its functionality is a reïmplementation that does not require the standard library to actually have the symbols present, doing so may not require an MSRV raise.

Now that bitvec is at 1.0, it will only raise MSRV in minor-edition releases. If you have a pinned Rust toolchain, you should depend on bitvec with a limiting minor-version constraint like "~1.0".

First, depend on it in your Cargo manifest:

[dependencies]
bitvec = "1"

Note: bitvec supports #![no_std] targets. If you do not have std, disable the default features, and explicitly restore any features that you do have:

[dependencies.bitvec]
version = "1"
default-features = false
features = ["atomic", "alloc"]

Once Cargo knows about it, bring its prelude into scope:

use bitvec::prelude::*;

You can read the prelude reëxports to see exactly which symbols are being imported. The prelude brings in many symbols, and while name collisions are not likely, you may wish to instead import the prelude module rather than its contents:

use bitvec::prelude as bv;

You should almost certainly use type aliases to make names for specific instantiations of bitvec type parameters, and use that rather than attempting to remain generic over an <T: BitStore, O: BitOrder> pair throughout your project.

Examples

use bitvec::prelude::*;

// All data-types have macro
// constructors.
let arr = bitarr![u32, Lsb0; 0; 80];
let bits = bits![u16, Msb0; 0; 40];

// Unsigned integers (scalar, array,
// and slice) can be borrowed.
let data = 0x2021u16;
let bits = data.view_bits::<Msb0>();
let data = [0xA5u8, 0x3C];
let bits = data.view_bits::<Lsb0>();

// Bit-slices can split anywhere.
let (head, rest) = bits.split_at(4);
assert_eq!(head, bits[.. 4]);
assert_eq!(rest, bits[4 ..]);

// And they are writable!
let mut data = [0u8; 2];
let bits = data.view_bits_mut::<Lsb0>();
// l and r each own one byte.
let (l, r) = bits.split_at_mut(8);

// but now a, b, c, and d own a nibble!
let ((a, b), (c, d)) = (
  l.split_at_mut(4),
  r.split_at_mut(4),
);

// and all four of them are writable.
a.set(0, true);
b.set(1, true);
c.set(2, true);
d.set(3, true);

assert!(bits[0]);  // a[0]
assert!(bits[5]);  // b[1]
assert!(bits[10]); // c[2]
assert!(bits[15]); // d[3]

// `BitSlice` is accessed by reference,
// which means it respects NLL styles.
assert_eq!(data, [0x21u8, 0x84]);

// Furthermore, bit-slices can store
// ordinary integers:
let eight = [0u8, 4, 8, 12, 16, 20, 24, 28];
//           a    b  c  d   e   f   g   h
let mut five = [0u8; 5];
for (slot, byte) in five
  .view_bits_mut::<Msb0>()
  .chunks_mut(5)
  .zip(eight.iter().copied())
{
  slot.store_be(byte);
  assert_eq!(slot.load_be::<u8>(), byte);
}

assert_eq!(five, [
  0b00000_001,
//  aaaaa bbb
  0b00_01000_0,
//  bb ccccc d
  0b1100_1000,
//  dddd eeee
  0b0_10100_11,
//  e fffff gg
  0b000_11100,
//  ggg hhhhh
]);

The BitSlice type is a view that alters the behavior of a borrowed memory region. It is never held directly, but only by references (created by borrowing integer memory) or the BitArray value type. In addition, the presence of a dynamic allocator enables the BitBox and BitVec buffer types, which can be used for more advanced buffer manipulation:

#[cfg(feature = "alloc")]
fn main() {

use bitvec::prelude::*;

let mut bv = bitvec![u8, Msb0;];
bv.push(false);
bv.push(true);
bv.extend([false; 4].iter());
bv.extend(&15u8.view_bits::<Lsb0>()[.. 4]);

assert_eq!(bv.as_raw_slice(), &[
  0b01_0000_11, 0b11_000000
//                   ^ dead
]);

}

While place expressions like bits[index] = value; are not available, bitvec instead provides a proxy structure that can be used as nearly an &mut bit reference:

use bitvec::prelude::*;

let bits = bits![mut 0];
// `bit` is not a reference, so
// it must be bound with `mut`.
let mut bit = bits.get_mut(0).unwrap();
assert!(!*bit);
*bit = true;
assert!(*bit);
// `bit` is not a reference,
// so NLL rules do not apply.
drop(bit);
assert!(bits[0]);

The bitvec data types implement a complete replacement for their standard-library counterparts, including all of the inherent methods, traits, and operator behaviors.

User Stories

Uses of bitvec generally fall into three major genres.

Bit Collections

At its most basic, bitvec provides sequence types analogous to the standard library’s bool collections. The default behavior is optimized for fast memory access and simple codegen, and can compact [bool] or Vec<bool> with minimal overhead.

While bitvec does not attempt to take advantage of SIMD or other vectorized instructions in its default work, its codegen should be a good candidate for autovectorization in LLVM. If explicit vectorization is important to you, please file an issue.

Example uses might be implementing a Sieve of Eratosthenes to store primes, or other collections that test a yes/no property of a number; or replacing Vec<Option<T>> with (BitVec, Vec<MaybeUninit<T>>).

To get started, you can perform basic text replacement on your project. Translate any existing types as follows:

and then follow any compiler errors that arise.

Bit-Field Memory Access

A single bit of information has very few uses. bitvec also enables you to store integers wider than a single bit, by selecting a bit-slice and using the BitField trait on it. You can store and retrieve both unsigned and signed integers, as long as the ordering type parameter is Lsb0 or Msb0.

If your bit-field storage buffers are never serialized for exchange between machines, then you can get away with using the default type parameters and unadorned load/store methods. While the in-memory layout of stored integers may be surprising if directly inspected, the overall behavior should be optimal for your target.

Remember: bitvec only provides array place expressions, using integer start and end points. You can use deku if you want C-style named structural fields with bit-field memory storage.

However, if you are de/serializing buffers for transport, then you fall into the third category.

Transport Protocols

Many protocols use sub-element fields in order to save space in transport; for example, TCP headers have single-bit and 4-bit fields in order to pack all the needed information into a desirable amount of space. In C or Erlang, these TCP protocol fields could be mapped by record fields in the language. In Rust, they can be mapped by indexing into a bit-slice.

When using bitvec to manage protocol buffers, you will need to select the exact type parameters that match your memory layout. For instance, TCP uses <u8, Msb0>, while IPv6 on a little-endian machine uses <u32, Lsb0>. Once you have done this, you can replace all of your (memory & mask) >> shift or memory |= (value & mask) << shift expressions with memory[start .. end].

As a direct example, the Itanium instruction set IA-64 uses very-long instruction words containing three 41-bit fields in a [u8; 16]. One IA-64 disassembler replaced its manual shift/mask implementation with bitvec range indexing, taking the bit numbers directly from the datasheet, and observed that their code was both easier to maintain and also had better performance as a result!

Feature Flags

bitvec has a few Cargo features that govern its API surface. The default feature set is:

[dependencies.bitvec]
version = "1"
features = [
  "alloc",
  "atomic",
  # "serde",
  "std",
]

Use default-features = false to disable all of them, then features = [] to restore the ones you need.

Deeper Reading

The API Documentation explores bitvec’s usage and implementation in great detail. In particular, you should read the documentation for the order, store, and field modules, as well as the BitSlice and BitArray types.

In addition, the user guide explores the philosophical and academic concepts behind bitvec’s construction, its goals, and the more intricate parts of its behavior.

While you should be able to get started with bitvec with only dropping it into your code and using the same habits you have with the standard library, both of these resources contain all of the information needed to understand what it does, how it works, and how it can be useful to you.