panic! vs Recoverable Errors

Panics abort unwinding (or abort strategy) for bugs or unrecoverable states. Result handles expected failures callers can respond to.

Recipe

fn divide(a: i32, b: i32) -> Result<i32, &'static str> {
    if b == 0 { Err("division by zero") } else { Ok(a / b) }
}
 
fn index_unchecked(v: &[i32], i: usize) -> i32 {
    v[i] // panics if out of bounds - programmer error if misused
}

When to reach for this: Result for I/O and validation; panic!/assert! for invariant violations and prototype unwrap.

Working Example

fn parse_config(s: &str) -> Result<u16, String> {
    s.trim().parse::<u16>()
        .map_err(|e| format!("invalid port: {e}"))
}
 
fn main() {
    match parse_config("8080") {
        Ok(p) => println!("{p}"),
        Err(e) => eprintln!("config error: {e}"),
    }
}

What this demonstrates:

• User misconfiguration -> Result
• Caller decides exit/retry
• No panic for expected bad input

Deep Dive

std::panic::catch_unwind rarely in app code - not for normal errors.

Gotchas

• unwrap in library - Panics caller's thread. Fix: Return Result.
• Panic for control flow - Expensive and unclear. Fix: Result or Option.
• Abort vs unwind - panic=abort in release changes unwinding. Fix: Know profile settings.
• Panic across FFI boundary - Undefined unless caught. Fix: Never panic across FFI.
• Indexing as default - v[i] panics. Fix: get(i) for user indices.

Alternatives

FAQs

Related

• Error Handling Basics
• The ? Operator
• Custom Error Types
• Panic & Unwrap Scenarios

Stack versions: This page was written for Rust 1.97.0 (edition 2024), Tokio 1.x, Axum 0.8, serde 1.0, sqlx 0.8, clap 4, and Polars 0.46+.