Concurrency

Every Rust developer who’s written lock-free code has written the same compare_exchange loop. Rust 1.95 finally gives atomics an update method that does it for you.

The old way

Atomically doubling a counter used to mean writing a retry loop yourself:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

use std::sync::atomic::{AtomicUsize, Ordering};

let counter = AtomicUsize::new(10);

loop {
    let current = counter.load(Ordering::Relaxed);
    let new_val = current * 2;
    match counter.compare_exchange(
        current, new_val,
        Ordering::SeqCst, Ordering::Relaxed,
    ) {
        Ok(_) => break,
        Err(_) => continue,
    }
}
// counter is now 20

It works, but it’s boilerplate — and easy to get wrong (use the wrong ordering, forget to retry, etc.).

The new way: update

1
2
3
4
5
6

use std::sync::atomic::{AtomicUsize, Ordering};

let counter = AtomicUsize::new(10);

counter.update(Ordering::SeqCst, Ordering::SeqCst, |x| x * 2);
// counter is now 20

One line. No loop. No chance of forgetting to retry on contention.

The method takes two orderings (one for the store on success, one for the load on failure) and a closure that transforms the current value. It handles the compare-and-swap retry loop internally.

It returns the previous value

Just like fetch_add and friends, update returns the value before the update:

1
2
3
4
5
6
7

use std::sync::atomic::{AtomicUsize, Ordering};

let counter = AtomicUsize::new(5);

let prev = counter.update(Ordering::SeqCst, Ordering::SeqCst, |x| x + 3);
assert_eq!(prev, 5);  // was 5
assert_eq!(counter.load(Ordering::SeqCst), 8);  // now 8

This makes it perfect for “fetch-and-modify” patterns where you need the old value.

Works on all atomic types

update isn’t just for AtomicUsize — it’s available on AtomicBool, AtomicIsize, AtomicUsize, and AtomicPtr too:

1
2
3
4
5

use std::sync::atomic::{AtomicBool, Ordering};

let flag = AtomicBool::new(false);
flag.update(Ordering::SeqCst, Ordering::SeqCst, |x| !x);
assert_eq!(flag.load(Ordering::SeqCst), true);

When to use update vs fetch_add

If your operation is a simple add, sub, or bitwise op, the specialized fetch_* methods are still better — they compile down to a single atomic instruction on most architectures.

Use update when your transformation is more complex: clamping, toggling state machines, applying arbitrary functions. Anywhere you’d previously hand-roll a CAS loop.

Summary

Available on stable since Rust 1.95.0 for AtomicBool, AtomicIsize, AtomicUsize, and AtomicPtr.

Multiple processes writing to the same file? That’s a recipe for corruption. Since Rust 1.89, File::lock gives you OS-backed file locking without external crates.

The problem

You have a CLI tool that appends to a shared log file. Two instances run at the same time, and suddenly your log entries are garbled — half a line from one process interleaved with another. Before 1.89, you’d reach for the fslock or file-lock crate. Now it’s built in.

Exclusive locking

File::lock() acquires an exclusive (write) lock. Only one handle can hold an exclusive lock at a time — all other attempts block until the lock is released:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17

use std::fs::File;
use std::io::{self, Write};

fn main() -> io::Result<()> {
    let mut file = File::options()
        .write(true)
        .create(true)
        .open("/tmp/rustbites_lock_demo.txt")?;

    // Blocks until the lock is acquired
    file.lock()?;

    writeln!(file, "safe write from process {}", std::process::id())?;

    // Lock is released when the file is closed (dropped)
    Ok(())
}

When the File is dropped, the lock is automatically released. No manual unlock() needed — though you can call file.unlock() explicitly if you want to release it early.

Shared (read) locking

Sometimes you want to allow multiple readers but block writers. That’s what lock_shared() is for:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

use std::fs::File;
use std::io::{self, Read};

fn main() -> io::Result<()> {
    let mut file = File::open("/tmp/rustbites_lock_demo.txt")?;

    // Multiple processes can hold a shared lock simultaneously
    file.lock_shared()?;

    let mut contents = String::new();
    file.read_to_string(&mut contents)?;
    println!("Read: {contents}");

    file.unlock()?; // explicit release
    Ok(())
}

Shared locks coexist with other shared locks, but block exclusive lock attempts. Classic reader-writer pattern, enforced at the OS level.

Non-blocking with try_lock

Don’t want to wait? try_lock() and try_lock_shared() return immediately instead of blocking:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

use std::fs::{self, File, TryLockError};

fn main() -> std::io::Result<()> {
    let file = File::options()
        .write(true)
        .create(true)
        .open("/tmp/rustbites_trylock.txt")?;

    match file.try_lock() {
        Ok(()) => println!("Lock acquired!"),
        Err(TryLockError::WouldBlock) => println!("File is busy, try later"),
        Err(TryLockError::Error(e)) => return Err(e),
    }

    Ok(())
}

If another process holds the lock, you get TryLockError::WouldBlock instead of hanging. Perfect for tools that should fail fast rather than block when another instance is already running.

Key details

• Advisory locks: these locks are advisory on most platforms — they don’t prevent other processes from reading/writing the file unless those processes also use locking
• Automatic release: locks are released when the File handle is dropped
• Cross-platform: works on Linux, macOS, and Windows (uses flock on Unix, LockFileEx on Windows)
• Stable since Rust 1.89

Need to share stack data with spawned threads? std::thread::scope lets you borrow local variables across threads — no Arc, no .clone().

The problem

With std::thread::spawn, you can’t borrow local data because the thread might outlive the data:

1
2
3
4
5
6
7

let data = vec![1, 2, 3];

// This won't compile — `data` might be dropped
// while the thread is still running
// std::thread::spawn(|| {
//     println!("{:?}", data);
// });

The classic workaround is wrapping everything in Arc:

1
2
3
4
5
6
7
8
9

use std::sync::Arc;

let data = Arc::new(vec![1, 2, 3]);
let data_clone = Arc::clone(&data);

let handle = std::thread::spawn(move || {
    println!("{:?}", data_clone);
});
handle.join().unwrap();

It works, but it’s noisy — especially when you just want to read some data in parallel.

The fix: std::thread::scope

Scoped threads guarantee that all spawned threads finish before the scope exits, so borrowing is safe:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17

let data = vec![1, 2, 3];
let mut results = vec![];

std::thread::scope(|s| {
    s.spawn(|| {
        // Borrowing `data` directly — no Arc needed
        println!("Thread sees: {:?}", data);
    });

    s.spawn(|| {
        let sum: i32 = data.iter().sum();
        println!("Sum: {sum}");
    });
});

// All threads have joined here — guaranteed
println!("Done! data is still ours: {:?}", data);

Mutable access works too

Since the scope enforces proper lifetimes, you can even have one thread mutably borrow something:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

let mut counts = [0u32; 3];

std::thread::scope(|s| {
    for (i, count) in counts.iter_mut().enumerate() {
        s.spawn(move || {
            *count = (i as u32 + 1) * 10;
        });
    }
});

assert_eq!(counts, [10, 20, 30]);

Each thread gets exclusive access to its own element — the borrow checker is happy, no Mutex required.

When to reach for scoped threads

Use std::thread::scope when you need parallel work on local data and don’t want the overhead or ceremony of Arc/Mutex. It’s perfect for fork-join parallelism: spin up threads, borrow what you need, collect results when they’re done.