fathomdb 0.3.1

#![allow(
    clippy::expect_used,
    clippy::missing_panics_doc,
    clippy::too_many_lines
)]

//! Phase 7a integration tests for the search FFI JSON roundtrip.
//!
//! These tests exercise the `execute_search_json` entry point that Python
//! and TypeScript SDKs will call into via pyo3 / napi wrappers. Each test
//! constructs the request as a JSON string, invokes the FFI, parses the
//! resulting `PySearchRows` JSON, and asserts field values. The FFI layer
//! is pure translation — compile plan → coordinator call → serialize — so
//! these tests double as a contract for the wire format consumed by the
//! higher-level SDKs in Packs 7b / 7c.

use fathomdb::search_ffi::{
    PyHitAttribution, PySearchHit, PySearchHitSource, PySearchMatchMode, PySearchRows,
    SearchFfiError, execute_search_json,
};
use fathomdb::{
    ChunkInsert, ChunkPolicy, Engine, EngineOptions, FtsPropertyPathSpec, NodeInsert, WriteRequest,
};
use tempfile::NamedTempFile;

fn open_engine() -> (NamedTempFile, Engine) {
    let db = NamedTempFile::new().expect("temporary db");
    let engine = Engine::open(EngineOptions::new(db.path())).expect("engine opens");
    (db, engine)
}

fn seed_budget_goals(engine: &Engine) {
    engine
        .register_fts_property_schema(
            "Goal",
            &["$.name".to_owned(), "$.description".to_owned()],
            None,
        )
        .expect("register property schema");

    engine
        .writer()
        .submit(WriteRequest {
            label: "seed-budget".to_owned(),
            nodes: vec![
                NodeInsert {
                    row_id: "budget-alpha-row".to_owned(),
                    logical_id: "budget-alpha".to_owned(),
                    kind: "Goal".to_owned(),
                    properties:
                        r#"{"name":"budget alpha goal","description":"quarterly budget rollup"}"#
                            .to_owned(),
                    source_ref: Some("seed".to_owned()),
                    upsert: false,
                    chunk_policy: ChunkPolicy::Preserve,
                    content_ref: None,
                },
                NodeInsert {
                    row_id: "budget-bravo-row".to_owned(),
                    logical_id: "budget-bravo".to_owned(),
                    kind: "Goal".to_owned(),
                    properties:
                        r#"{"name":"budget bravo goal","description":"annual budget summary"}"#
                            .to_owned(),
                    source_ref: Some("seed".to_owned()),
                    upsert: false,
                    chunk_policy: ChunkPolicy::Preserve,
                    content_ref: None,
                },
            ],
            node_retires: vec![],
            edges: vec![],
            edge_retires: vec![],
            chunks: vec![
                ChunkInsert {
                    id: "budget-alpha-chunk".to_owned(),
                    node_logical_id: "budget-alpha".to_owned(),
                    text_content: "alpha budget quarterly review notes".to_owned(),
                    byte_start: None,
                    byte_end: None,
                    content_hash: None,
                },
                ChunkInsert {
                    id: "budget-bravo-chunk".to_owned(),
                    node_logical_id: "budget-bravo".to_owned(),
                    text_content: "bravo budget annual summary notes".to_owned(),
                    byte_start: None,
                    byte_end: None,
                    content_hash: None,
                },
            ],
            runs: vec![],
            steps: vec![],
            actions: vec![],
            optional_backfills: vec![],
            vec_inserts: vec![],
            operational_writes: vec![],
        })
        .expect("seed goals");
}

/// Seed a budget-themed `Task` alongside the Goals planted by
/// [`seed_budget_goals`]. Both kinds share the word "budget" so filter
/// predicates are the only thing narrowing the result set.
fn seed_budget_task(engine: &Engine) {
    engine
        .register_fts_property_schema(
            "Task",
            &["$.name".to_owned(), "$.description".to_owned()],
            None,
        )
        .expect("register task schema");
    engine
        .writer()
        .submit(WriteRequest {
            label: "seed-budget-task".to_owned(),
            nodes: vec![NodeInsert {
                row_id: "budget-task-row".to_owned(),
                logical_id: "budget-task".to_owned(),
                kind: "Task".to_owned(),
                properties:
                    r#"{"name":"budget task","description":"reconcile quarterly budget figures"}"#
                        .to_owned(),
                source_ref: Some("seed".to_owned()),
                upsert: false,
                chunk_policy: ChunkPolicy::Preserve,
                content_ref: None,
            }],
            node_retires: vec![],
            edges: vec![],
            edge_retires: vec![],
            chunks: vec![ChunkInsert {
                id: "budget-task-chunk".to_owned(),
                node_logical_id: "budget-task".to_owned(),
                text_content: "task budget reconciliation notes".to_owned(),
                byte_start: None,
                byte_end: None,
                content_hash: None,
            }],
            runs: vec![],
            steps: vec![],
            actions: vec![],
            optional_backfills: vec![],
            vec_inserts: vec![],
            operational_writes: vec![],
        })
        .expect("seed task");
}

fn seed_recursive_note(engine: &Engine, logical_id: &str, props: &str) {
    engine
        .register_fts_property_schema_with_entries(
            "Note",
            &[FtsPropertyPathSpec::recursive("$.payload")],
            None,
            &[],
        )
        .expect("register recursive schema");
    engine
        .writer()
        .submit(WriteRequest {
            label: "seed-note".to_owned(),
            nodes: vec![NodeInsert {
                row_id: format!("{logical_id}-row"),
                logical_id: logical_id.to_owned(),
                kind: "Note".to_owned(),
                properties: props.to_owned(),
                source_ref: Some("seed".to_owned()),
                upsert: false,
                chunk_policy: ChunkPolicy::Preserve,
                content_ref: None,
            }],
            node_retires: vec![],
            edges: vec![],
            edge_retires: vec![],
            chunks: vec![],
            runs: vec![],
            steps: vec![],
            actions: vec![],
            optional_backfills: vec![],
            vec_inserts: vec![],
            operational_writes: vec![],
        })
        .expect("seed note");
}

#[test]
fn execute_search_json_rejects_malformed_json_as_parse_error() {
    // P7a-4 / P7.6a-2: malformed request JSON must surface as the
    // `Parse` variant, which the pyo3 mapper translates to `PyValueError`.
    let (_db, engine) = open_engine();
    let err = execute_search_json(&engine, "not json").expect_err("expected parse error");
    assert!(
        matches!(err, SearchFfiError::Parse(_)),
        "expected SearchFfiError::Parse, got {err:?}"
    );

    // Missing required field (`mode`) — also a Parse error, not an
    // engine or compile error.
    let missing_mode = r#"{
        "root_kind": "Goal",
        "strict_query": "budget",
        "limit": 10,
        "filters": []
    }"#;
    let err2 =
        execute_search_json(&engine, missing_mode).expect_err("expected missing-field parse error");
    assert!(
        matches!(err2, SearchFfiError::Parse(_)),
        "expected SearchFfiError::Parse, got {err2:?}"
    );
}

fn run(engine: &Engine, request_json: &str) -> PySearchRows {
    let response = execute_search_json(engine, request_json).expect("execute_search_json");
    serde_json::from_str::<PySearchRows>(&response).expect("parse PySearchRows")
}

#[test]
fn text_search_strict_hit_populates_all_fields() {
    let (_db, engine) = open_engine();
    seed_budget_goals(&engine);

    let request = r#"{
        "mode": "text_search",
        "root_kind": "Goal",
        "strict_query": "budget",
        "relaxed_query": null,
        "limit": 10,
        "filters": [{"type":"filter_kind_eq","kind":"Goal"}],
        "attribution_requested": false
    }"#;

    let rows = run(&engine, request);
    assert!(!rows.hits.is_empty());
    assert_eq!(rows.strict_hit_count, rows.hits.len());
    assert_eq!(rows.relaxed_hit_count, 0);
    assert!(!rows.fallback_used);
    assert!(!rows.was_degraded);

    let hit: &PySearchHit = &rows.hits[0];
    assert!(hit.score > 0.0, "score must be positive");
    assert!(matches!(hit.match_mode, Some(PySearchMatchMode::Strict)));
    assert!(matches!(
        hit.source,
        PySearchHitSource::Chunk | PySearchHitSource::Property
    ));
    assert!(hit.snippet.is_some());
    assert!(hit.written_at > 0);
    assert!(hit.projection_row_id.is_some());
    assert!(hit.attribution.is_none());
    // Node fields present.
    assert_eq!(hit.node.kind, "Goal");
    assert!(hit.node.logical_id.starts_with("budget-"));
}

#[test]
fn text_search_strict_miss_fires_relaxed_branch() {
    let (_db, engine) = open_engine();
    seed_budget_goals(&engine);

    // "budget quarterly nonexistentterm" — strict AND misses because of the
    // dead term; adaptive relaxation strips to "budget OR quarterly OR
    // nonexistentterm" and finds the seeded rows.
    let request = r#"{
        "mode": "text_search",
        "root_kind": "Goal",
        "strict_query": "budget quarterly zzznopeterm",
        "relaxed_query": null,
        "limit": 10,
        "filters": [{"type":"filter_kind_eq","kind":"Goal"}],
        "attribution_requested": false
    }"#;

    let rows = run(&engine, request);
    assert!(rows.fallback_used, "relaxed must fire on strict miss");
    assert!(!rows.hits.is_empty());
    assert_eq!(rows.strict_hit_count, 0);
    assert_eq!(rows.relaxed_hit_count, rows.hits.len());
    assert!(
        rows.hits
            .iter()
            .all(|h| matches!(h.match_mode, Some(PySearchMatchMode::Relaxed)))
    );
}

#[test]
fn text_search_with_attribution_populates_matched_paths() {
    let (_db, engine) = open_engine();
    seed_recursive_note(
        &engine,
        "note-attrib",
        r#"{"payload":{"body":"shipping quarterly docs"}}"#,
    );

    let request = r#"{
        "mode": "text_search",
        "root_kind": "Note",
        "strict_query": "shipping",
        "relaxed_query": null,
        "limit": 10,
        "filters": [],
        "attribution_requested": true
    }"#;

    let rows = run(&engine, request);
    assert!(!rows.hits.is_empty());
    let hit = &rows.hits[0];
    let att: &PyHitAttribution = hit
        .attribution
        .as_ref()
        .expect("attribution populated when requested");
    assert_eq!(att.matched_paths, vec!["$.payload.body".to_owned()]);
}

#[test]
fn fallback_search_two_shape_fires_relaxed_branch() {
    let (_db, engine) = open_engine();
    seed_budget_goals(&engine);
    // P7a-1/P7a-2: seed a second kind that ALSO matches "budget" so the
    // kind filter has work to do. Without the fallback FFI sentinel fix,
    // filters are silently dropped and this test would see hits of both
    // kinds.
    seed_budget_task(&engine);

    let request = r#"{
        "mode": "fallback_search",
        "root_kind": "",
        "strict_query": "zzznope1 zzznope2",
        "relaxed_query": "budget OR nothing",
        "limit": 10,
        "filters": [{"type":"filter_kind_eq","kind":"Goal"}],
        "attribution_requested": false
    }"#;

    // Unfiltered control — establishes that both kinds are in the index.
    let control_request = r#"{
        "mode": "fallback_search",
        "root_kind": "",
        "strict_query": "zzznope1 zzznope2",
        "relaxed_query": "budget OR nothing",
        "limit": 10,
        "filters": [],
        "attribution_requested": false
    }"#;
    let control = run(&engine, control_request);
    assert!(
        control.hits.iter().any(|h| h.node.kind == "Task"),
        "control (no filter) must include Task hits so the filter has work to do"
    );

    let rows = run(&engine, request);
    assert!(rows.fallback_used);
    assert!(!rows.hits.is_empty());
    assert_eq!(rows.strict_hit_count, 0);
    assert_eq!(rows.relaxed_hit_count, rows.hits.len());
    assert!(
        rows.hits
            .iter()
            .all(|h| matches!(h.match_mode, Some(PySearchMatchMode::Relaxed)))
    );
    // P7a-1: every hit must respect the kind filter.
    assert!(
        rows.hits.iter().all(|h| h.node.kind == "Goal"),
        "filter_kind_eq('Goal') must exclude Task rows (got kinds {:?})",
        rows.hits.iter().map(|h| &h.node.kind).collect::<Vec<_>>()
    );
    assert!(
        rows.hits.len() < control.hits.len(),
        "filtered result ({}) must be a strict subset of control ({})",
        rows.hits.len(),
        control.hits.len()
    );
    assert!(!rows.was_degraded);
}

#[test]
fn fallback_search_strict_only_matches_strict_only_text_search() {
    let (_db, engine) = open_engine();
    seed_budget_goals(&engine);
    // P7a-1/P7a-2: multi-kind seed so `filter_kind_eq('Goal')` does real
    // work. Without the fallback FFI filter-drop fix, the kind filter is
    // silently dropped and Task rows leak through.
    seed_budget_task(&engine);

    let fallback_strict = r#"{
        "mode": "fallback_search",
        "root_kind": "",
        "strict_query": "budget",
        "relaxed_query": null,
        "limit": 10,
        "filters": [{"type":"filter_kind_eq","kind":"Goal"}],
        "attribution_requested": false
    }"#;

    // Control: same fallback query with NO filter should return both kinds.
    let control_request = r#"{
        "mode": "fallback_search",
        "root_kind": "",
        "strict_query": "budget",
        "relaxed_query": null,
        "limit": 10,
        "filters": [],
        "attribution_requested": false
    }"#;
    let control = run(&engine, control_request);
    assert!(
        control.hits.iter().any(|h| h.node.kind == "Task"),
        "control (no filter) must include Task hits so the filter has work to do"
    );

    let rows_fb = run(&engine, fallback_strict);
    assert!(!rows_fb.hits.is_empty());
    assert!(!rows_fb.fallback_used);
    assert_eq!(rows_fb.relaxed_hit_count, 0);
    assert_eq!(rows_fb.strict_hit_count, rows_fb.hits.len());
    assert!(
        rows_fb
            .hits
            .iter()
            .all(|h| matches!(h.match_mode, Some(PySearchMatchMode::Strict)))
    );
    // P7a-1: every hit must respect the kind filter.
    assert!(
        rows_fb.hits.iter().all(|h| h.node.kind == "Goal"),
        "filter_kind_eq('Goal') must exclude Task rows (got {:?})",
        rows_fb
            .hits
            .iter()
            .map(|h| &h.node.kind)
            .collect::<Vec<_>>()
    );
    assert!(
        rows_fb.hits.len() < control.hits.len(),
        "filtered result ({}) must be a strict subset of control ({})",
        rows_fb.hits.len(),
        control.hits.len()
    );

    // P7a-5: cross-path parity — running the same strict-only query
    // through adaptive `text_search` should yield the same logical_ids in
    // the same order and the same top-level counts.
    let text_search_request = r#"{
        "mode": "text_search",
        "root_kind": "Goal",
        "strict_query": "budget",
        "relaxed_query": null,
        "limit": 10,
        "filters": [{"type":"filter_kind_eq","kind":"Goal"}],
        "attribution_requested": false
    }"#;
    let rows_ts = run(&engine, text_search_request);
    let fb_ids: Vec<&str> = rows_fb
        .hits
        .iter()
        .map(|h| h.node.logical_id.as_str())
        .collect();
    let ts_ids: Vec<&str> = rows_ts
        .hits
        .iter()
        .map(|h| h.node.logical_id.as_str())
        .collect();
    assert_eq!(
        fb_ids, ts_ids,
        "strict-only fallback and adaptive text_search must return the same hits"
    );
    assert_eq!(rows_fb.strict_hit_count, rows_ts.strict_hit_count);
    assert_eq!(rows_fb.relaxed_hit_count, rows_ts.relaxed_hit_count);
    assert_eq!(rows_fb.fallback_used, rows_ts.fallback_used);
    assert_eq!(rows_fb.was_degraded, rows_ts.was_degraded);
}

// ---------------------------------------------------------------------------
// Phase 13a: unified `search` FFI mode. Covers the Phase 12 SearchBuilder
// surface exposed through `execute_search_json`. Each case mirrors a Rust
// integration test in `text_search_surface.rs` but crosses the JSON
// boundary so the Python / TypeScript SDKs (Packs 13b / 13c) can rely on
// the wire shape.
// ---------------------------------------------------------------------------

#[test]
fn search_basic_populates_search_rows() {
    let (_db, engine) = open_engine();
    seed_budget_goals(&engine);

    let request = r#"{
        "mode": "search",
        "root_kind": "Goal",
        "strict_query": "budget",
        "relaxed_query": null,
        "limit": 10,
        "filters": [],
        "attribution_requested": false
    }"#;

    let rows = run(&engine, request);
    assert!(!rows.hits.is_empty());
    assert_eq!(rows.strict_hit_count, rows.hits.len());
    assert_eq!(rows.relaxed_hit_count, 0);
    assert_eq!(rows.vector_hit_count, 0, "v1 vector branch must stay empty");
    assert!(!rows.fallback_used);
    assert!(!rows.was_degraded);

    let hit: &PySearchHit = &rows.hits[0];
    assert!(hit.score > 0.0);
    assert!(matches!(hit.match_mode, Some(PySearchMatchMode::Strict)));
    assert_eq!(hit.node.kind, "Goal");
    assert!(hit.attribution.is_none());
}

#[test]
fn search_with_filter_kind_eq_is_fused() {
    // Mirrors `search_with_filter_kind_eq_fuses` in text_search_surface.rs.
    // Seeds a budget-themed Task alongside Goals so the kind filter has
    // real work to do; without filter fusion both kinds would come back.
    let (_db, engine) = open_engine();
    seed_budget_goals(&engine);
    seed_budget_task(&engine);

    let control_request = r#"{
        "mode": "search",
        "root_kind": "",
        "strict_query": "budget",
        "relaxed_query": null,
        "limit": 10,
        "filters": [],
        "attribution_requested": false
    }"#;
    let control = run(&engine, control_request);
    assert!(
        control.hits.iter().any(|h| h.node.kind == "Task"),
        "control (no kind filter) must include Task hits"
    );

    let filtered_request = r#"{
        "mode": "search",
        "root_kind": "",
        "strict_query": "budget",
        "relaxed_query": null,
        "limit": 10,
        "filters": [{"type":"filter_kind_eq","kind":"Goal"}],
        "attribution_requested": false
    }"#;
    let filtered = run(&engine, filtered_request);
    assert!(!filtered.hits.is_empty());
    assert!(
        filtered.hits.iter().all(|h| h.node.kind == "Goal"),
        "filter_kind_eq('Goal') must exclude Task rows"
    );
    assert!(filtered.hits.len() < control.hits.len());
}

#[test]
fn search_with_filter_json_text_eq_post_filter() {
    // JSON predicates run as residual post-filters. Assert the chain
    // crosses the FFI boundary without being dropped.
    let (_db, engine) = open_engine();
    seed_budget_goals(&engine);

    let request = r#"{
        "mode": "search",
        "root_kind": "Goal",
        "strict_query": "budget",
        "relaxed_query": null,
        "limit": 10,
        "filters": [{"type":"filter_json_text_eq","path":"$.name","value":"budget alpha goal"}],
        "attribution_requested": false
    }"#;

    let rows = run(&engine, request);
    assert!(!rows.hits.is_empty());
    assert!(
        rows.hits
            .iter()
            .all(|h| h.node.logical_id == "budget-alpha"),
        "json filter must restrict to budget-alpha, got {:?}",
        rows.hits
            .iter()
            .map(|h| &h.node.logical_id)
            .collect::<Vec<_>>()
    );
}

#[test]
fn search_with_attribution_on_recursive_schema() {
    let (_db, engine) = open_engine();
    seed_recursive_note(
        &engine,
        "note-search-attrib",
        r#"{"payload":{"body":"shipping quarterly docs"}}"#,
    );

    let request = r#"{
        "mode": "search",
        "root_kind": "Note",
        "strict_query": "shipping",
        "relaxed_query": null,
        "limit": 10,
        "filters": [],
        "attribution_requested": true
    }"#;

    let rows = run(&engine, request);
    assert!(!rows.hits.is_empty());
    let hit = &rows.hits[0];
    let att: &PyHitAttribution = hit
        .attribution
        .as_ref()
        .expect("attribution populated when requested");
    assert_eq!(att.matched_paths, vec!["$.payload.body".to_owned()]);
}

#[test]
fn search_empty_query_returns_empty_search_rows() {
    let (_db, engine) = open_engine();
    seed_budget_goals(&engine);

    let request = r#"{
        "mode": "search",
        "root_kind": "Goal",
        "strict_query": "",
        "relaxed_query": null,
        "limit": 10,
        "filters": [],
        "attribution_requested": false
    }"#;

    let rows = run(&engine, request);
    assert!(rows.hits.is_empty());
    assert_eq!(rows.strict_hit_count, 0);
    assert_eq!(rows.relaxed_hit_count, 0);
    assert_eq!(rows.vector_hit_count, 0);
    assert!(!rows.fallback_used);
}