//! token counting without a model in the loop.
//! cl100k-ish heuristic: bytes / 3.6, corrected per language.

const CORRECTION: &[(&str, f32)] = &[
    ("rs", 1.08),
    ("ts", 1.02),
    ("py", 0.97),
    ("md", 0.88),
];

pub fn count(file: &RepoFile) -> usize {
    let base = file.bytes.len() as f32 / 3.6;
    let k = CORRECTION
        .iter()
        .find(|(ext, _)| *ext == file.ext())
        .map(|(_, k)| *k)
        .unwrap_or(1.0);
    (base * k).ceil() as usize
}