"""SQLite-backed vector store for CodeRAG retrieval demos."""

from __future__ import annotations

import hashlib
import json
import re
import sqlite3
from contextlib import closing
from dataclasses import dataclass
from pathlib import Path
from typing import Iterable

import numpy as np

from .records import FileIndex
from .synthetic_queries import build_synthetic_questions


DEFAULT_DIMENSIONS = 384
SCHEMA_VERSION = "coderag-sqlite-vector-db-v1"
TOKEN_RE = re.compile(r"[A-Za-z_][A-Za-z0-9_]*|\d+(?:\.\d+)?")


@dataclass(frozen=True)
class SearchHit:
    key: str
    file_path: str
    qualified_name: str
    symbol_type: str
    start_line: int
    end_line: int
    raw_code: str
    own_sketch: str
    propagated_text: str
    source_scores: dict[str, float]
    best_source: str
    best_score: float
    matched_text: str
    matched_scope: str | None
    matched_evidence: list[str]

    def to_dict(self) -> dict:
        return {
            "key": self.key,
            "file_path": self.file_path,
            "qualified_name": self.qualified_name,
            "symbol_type": self.symbol_type,
            "start_line": self.start_line,
            "end_line": self.end_line,
            "raw_code": self.raw_code,
            "own_sketch": self.own_sketch,
            "propagated_text": self.propagated_text,
            "source_scores": self.source_scores,
            "best_source": self.best_source,
            "best_score": self.best_score,
            "matched_text": self.matched_text,
            "matched_scope": self.matched_scope,
            "matched_evidence": self.matched_evidence,
        }


def _tokens(text: str) -> list[str]:
    words = [match.group(0).lower() for match in TOKEN_RE.finditer(text)]
    bigrams = [
        f"{left}_{right}"
        for left, right in zip(words, words[1:])
        if left != right
    ]
    return [*words, *bigrams]


def embed_text(text: str, dimensions: int = DEFAULT_DIMENSIONS) -> np.ndarray:
    """Create a deterministic normalized hashing vector for local search."""
    vector = np.zeros(dimensions, dtype=np.float32)
    for token in _tokens(text):
        digest = hashlib.blake2b(token.encode("utf-8"), digest_size=8).digest()
        bucket = int.from_bytes(digest[:4], "little") % dimensions
        sign = 1.0 if digest[4] & 1 else -1.0
        vector[bucket] += sign
    norm = float(np.linalg.norm(vector))
    if norm > 0:
        vector /= norm
    return vector


def _blob(vector: np.ndarray) -> bytes:
    return np.asarray(vector, dtype=np.float32).tobytes()


def _vector(blob: bytes) -> np.ndarray:
    return np.frombuffer(blob, dtype=np.float32)


def _record_key(record: dict, root_path: str | None = None) -> str:
    path = Path(record["file_path"])
    if root_path:
        try:
            file_path = path.resolve().relative_to(Path(root_path).resolve()).as_posix()
        except ValueError:
            file_path = path.as_posix()
    else:
        file_path = path.as_posix()
    return f"{file_path}::{record['qualified_name']}"


class CodeRagVectorDB:
    """Small vector database used by the CodeRAG React demo."""

    def __init__(self, path: str | Path, *, dimensions: int = DEFAULT_DIMENSIONS):
        self.path = Path(path)
        self.dimensions = dimensions

    def connect(self) -> sqlite3.Connection:
        self.path.parent.mkdir(parents=True, exist_ok=True)
        connection = sqlite3.connect(self.path)
        connection.row_factory = sqlite3.Row
        return connection

    def initialize(self) -> None:
        with closing(self.connect()) as connection, connection:
            connection.executescript(
                """
                CREATE TABLE IF NOT EXISTS metadata (
                    key TEXT PRIMARY KEY,
                    value TEXT NOT NULL
                );

                CREATE TABLE IF NOT EXISTS functions (
                    key TEXT PRIMARY KEY,
                    file_path TEXT NOT NULL,
                    qualified_name TEXT NOT NULL,
                    symbol_type TEXT NOT NULL,
                    start_line INTEGER NOT NULL DEFAULT 0,
                    end_line INTEGER NOT NULL DEFAULT 0,
                    raw_code TEXT NOT NULL DEFAULT '',
                    own_sketch TEXT NOT NULL DEFAULT '',
                    propagated_text TEXT NOT NULL DEFAULT ''
                );

                CREATE TABLE IF NOT EXISTS vectors (
                    id INTEGER PRIMARY KEY AUTOINCREMENT,
                    function_key TEXT NOT NULL,
                    source TEXT NOT NULL,
                    scope TEXT,
                    text TEXT NOT NULL,
                    evidence_json TEXT NOT NULL DEFAULT '[]',
                    vector BLOB NOT NULL,
                    FOREIGN KEY(function_key) REFERENCES functions(key)
                );

                CREATE INDEX IF NOT EXISTS idx_vectors_source
                    ON vectors(source);
                CREATE INDEX IF NOT EXISTS idx_vectors_function
                    ON vectors(function_key);
                """
            )
            connection.execute(
                """
                INSERT OR REPLACE INTO metadata(key, value)
                VALUES ('schema_version', ?), ('dimensions', ?)
                """,
                (SCHEMA_VERSION, str(self.dimensions)),
            )

    def clear(self) -> None:
        with closing(self.connect()) as connection, connection:
            connection.execute("DELETE FROM vectors")
            connection.execute("DELETE FROM functions")

    def build(
        self,
        *,
        repository_index_path: str | Path,
        synthetic_questions_path: str | Path | None = None,
    ) -> dict:
        self.initialize()
        self.clear()
        repository_index = json.loads(
            Path(repository_index_path).read_text(encoding="utf-8")
        )
        root_path = repository_index.get("root_path")
        function_count = 0
        vector_count = 0
        with closing(self.connect()) as connection, connection:
            for record in repository_index["functions"]:
                key = _record_key(record, root_path)
                self._upsert_function(connection, key, record)
                for source, field in (
                    ("raw_code", "raw_code"),
                    ("cross_file_propagated", "propagated_text"),
                ):
                    text = record.get(field, "")
                    if text.strip():
                        self._insert_vector(
                            connection,
                            function_key=key,
                            source=source,
                            text=text,
                            scope=None,
                            evidence=[],
                        )
                        vector_count += 1
                function_count += 1

            synthetic_count = 0
            if synthetic_questions_path is not None:
                synthetic_path = Path(synthetic_questions_path)
                if synthetic_path.exists():
                    synthetic = json.loads(
                        synthetic_path.read_text(encoding="utf-8")
                    )
                    for payload in synthetic["repositories"].values():
                        for key, questions in payload["functions"].items():
                            self._upsert_placeholder(connection, key)
                            for question in questions:
                                self._insert_vector(
                                    connection,
                                    function_key=key,
                                    source="semantic_synthetic",
                                    text=question["text"],
                                    scope=question.get("scope"),
                                    evidence=question.get("evidence", []),
                                )
                                vector_count += 1
                                synthetic_count += 1

            connection.execute(
                """
                INSERT OR REPLACE INTO metadata(key, value)
                VALUES
                    ('repository_index_path', ?),
                    ('synthetic_questions_path', ?),
                    ('function_count', ?),
                    ('vector_count', ?),
                    ('synthetic_question_count', ?)
                """,
                (
                    str(repository_index_path),
                    str(synthetic_questions_path or ""),
                    str(function_count),
                    str(vector_count),
                    str(synthetic_count),
                ),
            )
        return self.stats()

    def build_from_file_index(self, index: FileIndex) -> dict:
        """Replace the vector database with records from one pasted-code index."""
        self.initialize()
        self.clear()
        records_by_key = {
            f"{Path(record.file_path).as_posix()}::{record.qualified_name}": record
            for record in index.functions
        }
        graph = {}
        for key, record in records_by_key.items():
            graph[key] = [
                f"{Path(record.file_path).as_posix()}::{callee}"
                for callee in record.direct_callees
                if f"{Path(record.file_path).as_posix()}::{callee}" in records_by_key
            ]
        synthetic = build_synthetic_questions(records_by_key, graph)
        vector_count = 0
        with closing(self.connect()) as connection, connection:
            for key, record in records_by_key.items():
                self._upsert_function(connection, key, record.to_dict())
                for source, text in (
                    ("raw_code", record.raw_code),
                    ("cross_file_propagated", record.propagated_text),
                ):
                    if text.strip():
                        self._insert_vector(
                            connection,
                            function_key=key,
                            source=source,
                            text=text,
                            scope=None,
                            evidence=[],
                        )
                        vector_count += 1
                for question in synthetic.get(key, []):
                    self._insert_vector(
                        connection,
                        function_key=key,
                        source="semantic_synthetic",
                        text=question.text,
                        scope=question.scope,
                        evidence=list(question.evidence),
                    )
                    vector_count += 1

            connection.execute(
                """
                INSERT OR REPLACE INTO metadata(key, value)
                VALUES
                    ('repository_index_path', ?),
                    ('synthetic_questions_path', ?),
                    ('function_count', ?),
                    ('vector_count', ?),
                    ('synthetic_question_count', ?),
                    ('active_index_mode', ?),
                    ('active_file_path', ?)
                """,
                (
                    str(index.file_path),
                    "generated-from-pasted-code",
                    str(len(records_by_key)),
                    str(vector_count),
                    str(sum(len(items) for items in synthetic.values())),
                    "pasted_code",
                    str(index.file_path),
                ),
            )
        stats = self.stats()
        stats["indexing_order"] = index.indexing_order
        stats["call_edge_count"] = len(index.call_edges)
        return stats

    def stats(self) -> dict:
        self.initialize()
        with closing(self.connect()) as connection:
            metadata = {
                row["key"]: row["value"]
                for row in connection.execute("SELECT key, value FROM metadata")
            }
            sources = {
                row["source"]: row["count"]
                for row in connection.execute(
                    """
                    SELECT source, COUNT(*) AS count
                    FROM vectors
                    GROUP BY source
                    ORDER BY source
                    """
                )
            }
            function_count = connection.execute(
                "SELECT COUNT(*) AS count FROM functions"
            ).fetchone()["count"]
            vector_count = connection.execute(
                "SELECT COUNT(*) AS count FROM vectors"
            ).fetchone()["count"]
        return {
            "schema_version": metadata.get("schema_version", SCHEMA_VERSION),
            "database_path": str(self.path),
            "dimensions": int(metadata.get("dimensions", self.dimensions)),
            "function_count": function_count,
            "vector_count": vector_count,
            "sources": sources,
            "repository_index_path": metadata.get("repository_index_path", ""),
            "synthetic_questions_path": metadata.get(
                "synthetic_questions_path",
                "",
            ),
            "active_index_mode": metadata.get("active_index_mode", "repository"),
            "active_file_path": metadata.get("active_file_path", ""),
        }

    def search(
        self,
        query: str,
        *,
        top_k: int = 8,
        per_source: int = 5,
        sources: Iterable[str] | None = None,
    ) -> dict:
        self.initialize()
        requested_sources = list(sources or (
            "raw_code",
            "cross_file_propagated",
            "semantic_synthetic",
        ))
        query_vector = embed_text(query, self.dimensions)
        if not np.any(query_vector):
            return {
                "query": query,
                "hits": [],
                "candidate_count": 0,
                "per_source": per_source,
                "sources": requested_sources,
                "stats": self.stats(),
            }
        with closing(self.connect()) as connection:
            rows = connection.execute(
                """
                SELECT
                    vectors.id,
                    vectors.function_key,
                    vectors.source,
                    vectors.scope,
                    vectors.text,
                    vectors.evidence_json,
                    vectors.vector,
                    functions.file_path,
                    functions.qualified_name,
                    functions.symbol_type,
                    functions.start_line,
                    functions.end_line,
                    functions.raw_code,
                    functions.own_sketch,
                    functions.propagated_text
                FROM vectors
                JOIN functions ON functions.key = vectors.function_key
                WHERE vectors.source IN ({})
                """.format(",".join("?" for _ in requested_sources)),
                requested_sources,
            ).fetchall()

        ranked_by_source: dict[str, list[tuple[float, sqlite3.Row]]] = {
            source: [] for source in requested_sources
        }
        for row in rows:
            score = float(np.dot(_vector(row["vector"]), query_vector))
            ranked_by_source[row["source"]].append((score, row))

        candidate_rows: dict[str, dict] = {}
        for source, scored_rows in ranked_by_source.items():
            for score, row in sorted(
                scored_rows,
                key=lambda item: item[0],
                reverse=True,
            )[:per_source]:
                key = row["function_key"]
                candidate = candidate_rows.setdefault(
                    key,
                    {
                        "row": row,
                        "source_scores": {},
                        "best_score": score,
                        "best_source": source,
                        "matched_text": row["text"],
                        "matched_scope": row["scope"],
                        "matched_evidence": json.loads(row["evidence_json"]),
                    },
                )
                candidate["source_scores"][source] = round(score, 6)
                if score > candidate["best_score"]:
                    candidate["best_score"] = score
                    candidate["best_source"] = source
                    candidate["matched_text"] = row["text"]
                    candidate["matched_scope"] = row["scope"]
                    candidate["matched_evidence"] = json.loads(
                        row["evidence_json"]
                    )

        hits = []
        for key, candidate in candidate_rows.items():
            row = candidate["row"]
            hits.append(SearchHit(
                key=key,
                file_path=row["file_path"],
                qualified_name=row["qualified_name"],
                symbol_type=row["symbol_type"],
                start_line=row["start_line"],
                end_line=row["end_line"],
                raw_code=row["raw_code"],
                own_sketch=row["own_sketch"],
                propagated_text=row["propagated_text"],
                source_scores=candidate["source_scores"],
                best_source=candidate["best_source"],
                best_score=round(candidate["best_score"], 6),
                matched_text=candidate["matched_text"],
                matched_scope=candidate["matched_scope"],
                matched_evidence=candidate["matched_evidence"],
            ))
        hits.sort(key=lambda hit: hit.best_score, reverse=True)
        return {
            "query": query,
            "hits": [hit.to_dict() for hit in hits[:top_k]],
            "candidate_count": len(candidate_rows),
            "per_source": per_source,
            "sources": requested_sources,
            "stats": self.stats(),
        }

    def _upsert_function(
        self,
        connection: sqlite3.Connection,
        key: str,
        record: dict,
    ) -> None:
        connection.execute(
            """
            INSERT OR REPLACE INTO functions(
                key,
                file_path,
                qualified_name,
                symbol_type,
                start_line,
                end_line,
                raw_code,
                own_sketch,
                propagated_text
            )
            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
            """,
            (
                key,
                record.get("file_path", key.split("::", 1)[0]),
                record.get("qualified_name", key.split("::", 1)[-1]),
                record.get("symbol_type", "function"),
                int(record.get("start_line", 0) or 0),
                int(record.get("end_line", 0) or 0),
                record.get("raw_code", ""),
                record.get("own_sketch", ""),
                record.get("propagated_text", ""),
            ),
        )

    def _upsert_placeholder(
        self,
        connection: sqlite3.Connection,
        key: str,
    ) -> None:
        path, qualified = key.split("::", 1)
        connection.execute(
            """
            INSERT OR IGNORE INTO functions(
                key,
                file_path,
                qualified_name,
                symbol_type
            )
            VALUES (?, ?, ?, ?)
            """,
            (key, path, qualified, "function"),
        )

    def _insert_vector(
        self,
        connection: sqlite3.Connection,
        *,
        function_key: str,
        source: str,
        text: str,
        scope: str | None,
        evidence: list[str],
    ) -> None:
        connection.execute(
            """
            INSERT INTO vectors(
                function_key,
                source,
                scope,
                text,
                evidence_json,
                vector
            )
            VALUES (?, ?, ?, ?, ?, ?)
            """,
            (
                function_key,
                source,
                scope,
                text,
                json.dumps(evidence),
                _blob(embed_text(text, self.dimensions)),
            ),
        )