[BUG] frame_context lowercases the whole document once per query word

Author: rohitdevsolCreated Aug 9, 2026Updated Aug 21, 2026

Labelsbug

Bug Description

In src/memvid/frame.rs (around 386) , the code lowercases the entire document again for every single word in the query

rust

.map(|needle| content.to_lowercase().matches(&needle).count())

Because content.to_lowercase() is inside the loop, a 30 word query(I tested with 10 first then 30) lowercases the whole document 30 times instead of once. That's a lot of wasted work and memory allocation on big documents.

The fix is one line - lowercase it once before the loop and reuse it:

let lowered = content.to_lowercase();

I benchmarked the pattern with cargo bench (30-word query, and a large document )

buggy (current): ~552 µs
fixed (lowercase once): ~496 µs
about 10% faster, with no downside

my test screenshot

I performed benchmarks using this code.

rust

use std::hint::black_box;
use criterion::{ criterion_group, criterion_main, Criterion };

fn count_buggy(content: &str, query: &str) -> usize {
    let mut total = 0;
    for word in query.split_whitespace() {
        total += content.to_lowercase().matches(&word.to_lowercase()).count();
    }
    total
}

fn count_fixed(content: &str, query: &str) -> usize {
    let lowered = content.to_lowercase();
    let mut total = 0;
    for word in query.split_whitespace() {
        total += lowered.matches(&word.to_lowercase()).count();
    }
    total
}

fn bench(c: &mut Criterion) {
    let content = "Memvid is fast and good. Other memory libraries are slow and noob. ".repeat(
        1000
    );
    let query =
        "memvid is good fast others are noob slow memvid is good fast \
                 others are noob slow memvid is good fast others are noob slow \
                 memvid is good fast";

    c.bench_function("buggy", |b| {
        b.iter(|| count_buggy(black_box(&content), black_box(query)))
    });
    c.bench_function("fixed", |b| {
        b.iter(|| count_fixed(black_box(&content), black_box(query)))
    });
}

criterion_group!(benches, bench);
criterion_main!(benches);

Source: memvid/memvid

View original on GitHub View discussion on GitHub