autopilot/crates/vlm_client/tests/parser.rs

//! AZ-674 acceptance criteria.
//!
//! AC-1 — valid response parses successfully (round-trip through the
//! UDS fixture, verifying schema fields all survive).
//! AC-2 — schema-invalid response returns `status: SchemaInvalid` and
//! the schema-invalid counter increments.
//! AC-3 — model_version change logged once; subsequent identical
//! versions do NOT re-log (observed via the parser's `changes`
//! counter, which is incremented exactly once per change).
//! AC-4 — `VlmStatus` is exhaustive (compile-time check: this file
//! contains a `match` over every variant with no `_` arm).

#![cfg(feature = "vlm")]

use std::path::PathBuf;

use shared::contracts::VlmProvider;
use shared::models::vlm::{VlmLabel, VlmStatus};
use tempfile::tempdir;
use tokio::io::{AsyncReadExt, AsyncWriteExt};
use tokio::net::UnixListener;
use vlm_client::VlmClient;

async fn fixture_emitting(
    path: PathBuf,
    bodies: Vec<serde_json::Value>,
) -> tokio::task::JoinHandle<()> {
    let listener = UnixListener::bind(&path).unwrap();
    tokio::spawn(async move {
        for body in bodies {
            let (mut s, _) = listener.accept().await.unwrap();
            let mut lenbuf = [0u8; 4];
            if s.read_exact(&mut lenbuf).await.is_err() {
                return;
            }
            let len = u32::from_be_bytes(lenbuf) as usize;
            let mut req = vec![0u8; len];
            if s.read_exact(&mut req).await.is_err() {
                return;
            }
            let bytes = serde_json::to_vec(&body).unwrap();
            let _ = s.write_all(&(bytes.len() as u32).to_be_bytes()).await;
            let _ = s.write_all(&bytes).await;
            let _ = s.flush().await;
        }
    })
}

#[tokio::test]
async fn ac1_valid_response_parses_successfully() {
    // Arrange
    let dir = tempdir().unwrap();
    let path = dir.path().join("nanollm.sock");
    let body = serde_json::json!({
        "label": "confirmed_concealed_position",
        "confidence": 0.91,
        "evidence_spans": ["thicket"],
        "reason": "match",
        "status": "ok",
        "latency_ms": 42,
        "model_version": "VILA1.5-3B-int4"
    });
    let fixture = fixture_emitting(path.clone(), vec![body]).await;
    let client = VlmClient::open(&path).await.expect("connect");

    // Act
    let a = client
        .assess(b"\xff\xd8\xff".to_vec(), "describe".into())
        .await
        .unwrap();

    // Assert
    assert_eq!(a.status, VlmStatus::Ok);
    assert_eq!(a.label, VlmLabel::ConfirmedConcealedPosition);
    assert_eq!(a.model_version, "VILA1.5-3B-int4");
    assert_eq!(a.latency_ms, 42);
    assert_eq!(a.evidence_spans, vec!["thicket".to_string()]);

    // Parser counters reflect the success path.
    let parser = client.inner().unwrap().parser();
    assert_eq!(parser.schema_invalid_count(), 0);
    assert_eq!(parser.model_version_changes(), 1);

    fixture.await.unwrap();
}

#[tokio::test]
async fn ac2_schema_invalid_response_returns_schema_invalid_and_increments_counter() {
    // Arrange — fixture responds with valid JSON missing `model_version`.
    let dir = tempdir().unwrap();
    let path = dir.path().join("nanollm.sock");
    let bad_body = serde_json::json!({
        "label": "rejected",
        "confidence": 0.4,
        "status": "ok",
        "latency_ms": 5
        // model_version intentionally missing
    });
    let fixture = fixture_emitting(path.clone(), vec![bad_body]).await;
    let client = VlmClient::open(&path).await.expect("connect");

    // Act
    let a = client.assess(b"r".to_vec(), "p".into()).await.unwrap();

    // Assert
    assert_eq!(a.status, VlmStatus::SchemaInvalid);
    assert!(a.reason.starts_with("json:"), "got reason={}", a.reason);

    let parser = client.inner().unwrap().parser();
    assert_eq!(parser.schema_invalid_count(), 1);
    assert_eq!(parser.model_version_changes(), 0);

    fixture.await.unwrap();
}

/// AC-3 is exercised at the parser level — the model-version tracker
/// is a pure-state component that does not depend on the UDS layer.
/// The integration path is verified by AC-1 (one happy-path round
/// trip → parser sees one change event).
#[test]
fn ac3_model_version_change_logged_once_at_parser_level() {
    use vlm_client::AssessmentParser;

    // Arrange
    let parser = AssessmentParser::new();
    let mk = |v: &str| {
        serde_json::to_vec(&serde_json::json!({
            "label": "rejected",
            "confidence": 0.5,
            "status": "ok",
            "latency_ms": 1,
            "model_version": v
        }))
        .unwrap()
    };

    // Act — three responses: v1.0, v1.0 (no change), v1.1 (change).
    let _ = parser.parse(&mk("v1.0"));
    let _ = parser.parse(&mk("v1.0"));
    let _ = parser.parse(&mk("v1.1"));

    // Assert — exactly 2 change events: None→v1.0 and v1.0→v1.1.
    assert_eq!(parser.model_version_changes(), 2);
    assert_eq!(parser.current_model_version().as_deref(), Some("v1.1"));
}

/// Compile-time AC-4: this match must cover every `VlmStatus` variant
/// without a `_` arm. Adding a new variant breaks the build until
/// the consumer is updated.
#[test]
fn ac4_vlm_status_match_is_exhaustive() {
    // Arrange — synthesise one of each variant.
    let cases = [
        VlmStatus::Ok,
        VlmStatus::Inconclusive,
        VlmStatus::Timeout,
        VlmStatus::SchemaInvalid,
        VlmStatus::IpcError,
        VlmStatus::Disabled,
    ];

    // Act / Assert — every variant must produce a labelled outcome.
    for s in cases {
        let label: &'static str = match s {
            VlmStatus::Ok => "ok",
            VlmStatus::Inconclusive => "inconclusive",
            VlmStatus::Timeout => "timeout",
            VlmStatus::SchemaInvalid => "schema_invalid",
            VlmStatus::IpcError => "ipc_error",
            VlmStatus::Disabled => "disabled",
        };
        assert!(!label.is_empty());
    }
}

#[test]
fn schema_invalid_does_not_pollute_model_version_tracker() {
    use vlm_client::AssessmentParser;

    // Arrange — one valid body followed by one truncated/invalid body.
    // The tracker must not regress to None on the second call.
    let parser = AssessmentParser::new();
    let good = serde_json::to_vec(&serde_json::json!({
        "label": "rejected",
        "confidence": 0.5,
        "status": "ok",
        "latency_ms": 1,
        "model_version": "v1.0"
    }))
    .unwrap();
    let bad = good[..good.len() - 10].to_vec();

    // Act
    let r1 = parser.parse(&good);
    let r2 = parser.parse(&bad);

    // Assert
    assert_eq!(r1.status, VlmStatus::Ok);
    assert_eq!(r2.status, VlmStatus::SchemaInvalid);
    assert_eq!(parser.model_version_changes(), 1);
    assert_eq!(parser.current_model_version().as_deref(), Some("v1.0"));
    assert_eq!(parser.schema_invalid_count(), 1);
}