Skip to content

Parsing deeply nested flow collection is quadratically slow #1

@dtolnay

Description

@dtolnay
// [dependencies]
// serde = "1"
// yaml_serde = "0.10"

fn main() {
    const N: usize = 10000;
    let y = r#"{"":["#.repeat(N) + &r"]}".repeat(N);
    let begin = std::time::Instant::now();
    let serde::de::IgnoredAny = yaml_serde::from_str(&y).unwrap();
    println!("{:?}", begin.elapsed());
}

At N=10k, cargo run --release takes about 1 second. At N=20k, about 4 seconds. At N=40k, 16 seconds. Each doubling of input size quadruples the runtime. This is O(n²) scaling, which is risky for many use cases because relatively small amounts of untrusted input can tie up large amount of CPU time.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions