//! Calibration for [`wai::jpegai_pipeline::lef_filter`] — the LEF adaptive-sharpness
//! post-filter. Runs on the reference fixture (input luma, the 16×16 scale grid, the
//! three thresholds/magnitudes) and compares against the reference `y_want`. Float
//! arithmetic ⇒ machine-precision match expected.
//!
//! env: WAI_JPEGAI_LEF_FIXTURE = lef_fixture.json
//!   { y_in:[f64], y_want:[f64], sigma:[f64], sh, sw, thr:[f64;3], mag:[f64;3], h, w }
#![cfg(feature = "neural_int")]

use wai::jpegai_pipeline::lef_filter;

#[test]
fn lef_filter_matches_reference() {
    let path = match std::env::var("WAI_JPEGAI_LEF_FIXTURE") {
        Ok(p) if std::path::Path::new(&p).exists() => p,
        _ => { eprintln!("skip: set WAI_JPEGAI_LEF_FIXTURE to lef_fixture.json"); return; }
    };
    let v: serde_json::Value = serde_json::from_slice(&std::fs::read(&path).unwrap()).unwrap();
    let f32v = |k: &str| v[k].as_array().unwrap().iter().map(|x| x.as_f64().unwrap() as f32).collect::<Vec<f32>>();
    let y_in = f32v("y_in");
    let y_want = f32v("y_want");
    let sigma = f32v("sigma");
    let thr_v = f32v("thr");
    let mag_v = f32v("mag");
    let g = |k: &str| v[k].as_u64().unwrap() as usize;
    let (sh, sw, h, w) = (g("sh"), g("sw"), g("h"), g("w"));

    let got = lef_filter(&y_in, &sigma, sh, sw, [thr_v[0], thr_v[1], thr_v[2]], [mag_v[0], mag_v[1], mag_v[2]], h, w);
    assert_eq!(got.len(), y_want.len(), "output size mismatch");

    let mut maxd = 0f32;
    let mut mse = 0f64;
    for i in 0..got.len() {
        let d = (got[i] - y_want[i]).abs();
        maxd = maxd.max(d);
        mse += (d as f64) * (d as f64);
    }
    mse /= got.len() as f64;
    let sig: f64 = y_want.iter().map(|&x| (x as f64) * (x as f64)).sum::<f64>() / got.len() as f64;
    let snr = if mse > 0.0 { 10.0 * (sig / mse).log10() } else { f64::INFINITY };
    eprintln!("lef_filter vs reference: max|Δ| = {maxd:.6}   SNR = {snr:.2} dB");
    assert!(maxd < 1e-2, "LEF filter diverged from the reference (max|Δ| = {maxd})");
}