`no_std` — Rust Without the Standard Library

What you'll learn: How to write Rust for bare-metal and embedded targets using #![no_std] — the core and alloc crate split, panic handlers, and how this compares to embedded C without libc.

If you come from embedded C, you're already used to working without libc or with a minimal runtime. Rust has a first-class equivalent: the #![no_std] attribute.

What is `no_std`?

When you add #![no_std] to the crate root, the compiler removes the implicit extern crate std; and links only against core (and optionally alloc).

Layer	What it provides	Requires OS / heap?
`core`	Primitive types, `Option`, `Result`, `Iterator`, math, `slice`, `str`, atomics, `fmt`	No — runs on bare metal
`alloc`	`Vec`, `String`, `Box`, `Rc`, `Arc`, `BTreeMap`	Needs a global allocator, but no OS
`std`	`HashMap`, `fs`, `net`, `thread`, `io`, `env`, `process`	Yes — needs an OS

Rule of thumb for embedded devs: if your C project links against -lc and uses malloc, you can probably use core + alloc. If it runs on bare metal without malloc, stick with core only.

Declaring `no_std`

// src/lib.rs  (or src/main.rs for a binary with #![no_main])
#![no_std]

// You still get everything in `core`:
use core::fmt;
use core::result::Result;
use core::option::Option;

// If you have an allocator, opt in to heap types:
extern crate alloc;
use alloc::vec::Vec;
use alloc::string::String;

For a bare-metal binary you also need #![no_main] and a panic handler:

#![no_std]
#![no_main]

use core::panic::PanicInfo;

#[panic_handler]
fn panic(_info: &PanicInfo) -> ! {
    loop {} // hang on panic — replace with your board's reset/LED blink
}

// Entry point depends on your HAL / linker script

What you lose (and alternatives)

`std` feature	`no_std` alternative
`println!`	`core::write!` to a UART / `defmt`
`HashMap`	`heapless::FnvIndexMap` (fixed capacity) or `BTreeMap` (with `alloc`)
`Vec`	`heapless::Vec` (stack-allocated, fixed capacity)
`String`	`heapless::String` or `&str`
`std::io::Read/Write`	`embedded_io::Read/Write`
`thread::spawn`	Interrupt handlers, RTIC tasks
`std::time`	Hardware timer peripherals
`std::fs`	Flash / EEPROM drivers

Notable `no_std` crates for embedded

Crate	Purpose	Notes
`heapless`	Fixed-capacity `Vec`, `String`, `Queue`, `Map`	No allocator needed — all on the stack
`defmt`	Efficient logging over probe/ITM	Like `printf` but deferred formatting on the host
`embedded-hal`	Hardware abstraction traits (SPI, I²C, GPIO, UART)	Implement once, run on any MCU
`cortex-m`	ARM Cortex-M intrinsics & register access	Low-level, like CMSIS
`cortex-m-rt`	Runtime / startup code for Cortex-M	Replaces your `startup.s`
`rtic`	Real-Time Interrupt-driven Concurrency	Compile-time task scheduling, zero overhead
`embassy`	Async executor for embedded	`async/await` on bare metal
`postcard`	`no_std` serde serialization (binary)	Replaces `serde_json` when you can't afford strings
`thiserror`	Derive macro for `Error` trait	Works in `no_std` since v2; prefer over `anyhow`
`smoltcp`	`no_std` TCP/IP stack	When you need networking without an OS

C vs Rust: bare-metal comparison

A typical embedded C blinky:

// C — bare metal, vendor HAL
#include "stm32f4xx_hal.h"

void SysTick_Handler(void) {
    HAL_GPIO_TogglePin(GPIOA, GPIO_PIN_5);
}

int main(void) {
    HAL_Init();
    __HAL_RCC_GPIOA_CLK_ENABLE();
    GPIO_InitTypeDef gpio = { .Pin = GPIO_PIN_5, .Mode = GPIO_MODE_OUTPUT_PP };
    HAL_GPIO_Init(GPIOA, &gpio);
    HAL_SYSTICK_Config(HAL_RCC_GetHCLKFreq() / 1000);
    while (1) {}
}

The Rust equivalent (using embedded-hal + a board crate):

#![no_std]
#![no_main]

use cortex_m_rt::entry;
use panic_halt as _; // panic handler: infinite loop
use stm32f4xx_hal::{pac, prelude::*};

#[entry]
fn main() -> ! {
    let dp = pac::Peripherals::take().unwrap();
    let gpioa = dp.GPIOA.split();
    let mut led = gpioa.pa5.into_push_pull_output();

    let rcc = dp.RCC.constrain();
    let clocks = rcc.cfgr.freeze();
    let mut delay = dp.TIM2.delay_ms(&clocks);

    loop {
        led.toggle();
        delay.delay_ms(500u32);
    }
}

Key differences for C devs:

Peripherals::take() returns Option — ensures the singleton pattern at compile time (no double-init bugs)
.split() moves ownership of individual pins — no risk of two modules driving the same pin
All register access is type-checked — you can't accidentally write to a read-only register
The borrow checker prevents data races between main and interrupt handlers (with RTIC)

When to use `no_std` vs `std`

Loading diagram...

Exercise: `no_std` ring buffer

🔴 Challenge — combines generics, MaybeUninit, and #[cfg(test)] in a no_std context

In embedded systems you often need a fixed-size ring buffer (circular buffer) that never allocates. Implement one using only core (no alloc, no std).

Requirements:

Generic over element type T: Copy
Fixed capacity N (const generic)
push(&mut self, item: T) — overwrites oldest element when full
pop(&mut self) -> Option<T> — returns oldest element
len(&self) -> usize
is_empty(&self) -> bool
Must compile with #![no_std]

// Starter code
#![no_std]

use core::mem::MaybeUninit;

pub struct RingBuffer<T: Copy, const N: usize> {
    buf: [MaybeUninit<T>; N],
    head: usize,  // next write position
    tail: usize,  // next read position
    count: usize,
}

impl<T: Copy, const N: usize> RingBuffer<T, N> {
    pub const fn new() -> Self {
        todo!()
    }
    pub fn push(&mut self, item: T) {
        todo!()
    }
    pub fn pop(&mut self) -> Option<T> {
        todo!()
    }
    pub fn len(&self) -> usize {
        todo!()
    }
    pub fn is_empty(&self) -> bool {
        todo!()
    }
}

Solution

#![no_std]

use core::mem::MaybeUninit;

pub struct RingBuffer<T: Copy, const N: usize> {
    buf: [MaybeUninit<T>; N],
    head: usize,
    tail: usize,
    count: usize,
}

impl<T: Copy, const N: usize> RingBuffer<T, N> {
    pub const fn new() -> Self {
        Self {
            // SAFETY: MaybeUninit does not require initialization
            buf: unsafe { MaybeUninit::uninit().assume_init() },
            head: 0,
            tail: 0,
            count: 0,
        }
    }

    pub fn push(&mut self, item: T) {
        self.buf[self.head] = MaybeUninit::new(item);
        self.head = (self.head + 1) % N;
        if self.count == N {
            // Buffer is full — overwrite oldest, advance tail
            self.tail = (self.tail + 1) % N;
        } else {
            self.count += 1;
        }
    }

    pub fn pop(&mut self) -> Option<T> {
        if self.count == 0 {
            return None;
        }
        // SAFETY: We only read positions that were previously written via push()
        let item = unsafe { self.buf[self.tail].assume_init() };
        self.tail = (self.tail + 1) % N;
        self.count -= 1;
        Some(item)
    }

    pub fn len(&self) -> usize {
        self.count
    }

    pub fn is_empty(&self) -> bool {
        self.count == 0
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn basic_push_pop() {
        let mut rb = RingBuffer::<u32, 4>::new();
        assert!(rb.is_empty());

        rb.push(10);
        rb.push(20);
        rb.push(30);
        assert_eq!(rb.len(), 3);

        assert_eq!(rb.pop(), Some(10));
        assert_eq!(rb.pop(), Some(20));
        assert_eq!(rb.pop(), Some(30));
        assert_eq!(rb.pop(), None);
    }

    #[test]
    fn overwrite_on_full() {
        let mut rb = RingBuffer::<u8, 3>::new();
        rb.push(1);
        rb.push(2);
        rb.push(3);
        // Buffer full: [1, 2, 3]

        rb.push(4); // Overwrites 1 → [4, 2, 3], tail advances
        assert_eq!(rb.len(), 3);
        assert_eq!(rb.pop(), Some(2)); // oldest surviving
        assert_eq!(rb.pop(), Some(3));
        assert_eq!(rb.pop(), Some(4));
        assert_eq!(rb.pop(), None);
    }
}

Why this matters for embedded C devs:

MaybeUninit is Rust's equivalent of uninitialized memory — the compiler won't insert zero-fills, just like char buf[N]; in C
The unsafe blocks are minimal (2 lines) and each has a // SAFETY: comment
The const fn new() means you can create ring buffers in static variables without a runtime constructor
The tests run on your host with cargo test even though the code is no_std