sirchmunk 0.0.0__py3-none-any.whl → 0.0.1.post1__py3-none-any.whl

sirchmunk/retrieve/text_retriever.py

@@ -0,0 +1,1026 @@
+# Copyright (c) ModelScope Contributors. All rights reserved.
+import asyncio
+import json
+import math
+import re
+import subprocess
+from pathlib import Path
+from typing import Any, Dict, List, Literal, Optional, Union
+
+from loguru import logger
+
+from ..utils.constants import GREP_CONCURRENT_LIMIT, DEFAULT_WORK_PATH
+from ..utils.file_utils import StorageStructure
+from .base import BaseRetriever
+
+RGA_SEMAPHORE = asyncio.Semaphore(value=GREP_CONCURRENT_LIMIT)
+
+
+class GrepRetriever(BaseRetriever):
+    """A Python wrapper for ripgrep-all (rga), exposing its functionality via static methods.
+
+    All methods are static and return parsed results. JSON output is preferred where possible
+    for reliable parsing. Shell injection is mitigated by using `subprocess.run` with `shell=False`
+    and explicit argument lists.
+
+    For more information about ripgrep-all, please refer to `https://github.com/phiresky/ripgrep-all`
+    """
+
+    def __init__(self, work_path: Union[str, Path] = None, **kwargs):
+        super().__init__()
+
+        self.work_path: Path = Path(work_path or DEFAULT_WORK_PATH)
+        self.rga_cache: Path = (
+            self.work_path / StorageStructure.CACHE_DIR / StorageStructure.GREP_DIR
+        )
+        self.rga_cache.mkdir(parents=True, exist_ok=True)
+
+    async def retrieve(
+        self,
+        terms: Union[str, List[str]],
+        path: Union[str, Path, List[str], List[Path], None] = None,
+        logic: Literal["and", "or", "not"] = "or",
+        *,
+        case_sensitive: bool = False,
+        whole_word: bool = False,
+        literal: bool = False,
+        regex: bool = True,
+        max_depth: Optional[int] = None,
+        include: Optional[List[str]] = None,
+        exclude: Optional[List[str]] = None,
+        file_type: Optional[str] = None,
+        invert_match: bool = False,
+        count_only: bool = False,
+        line_number: bool = True,
+        with_filename: bool = True,
+        rank: bool = True,
+        rank_kwargs: Optional[Dict] = None,
+        rga_no_cache: bool = False,
+        rga_cache_max_blob_len: int = 10000000,
+        rga_cache_path: Optional[Union[str, Path]] = None,
+        timeout: float = 60.0,
+    ) -> List[Dict[str, Any]]:
+        """Search for terms in files using ripgrep-all, supporting AND/OR/NOT logic.
+
+        Args:
+            terms: Single pattern (str) or list of patterns (List[str]).
+            path: Single path (str/Path) or multiple paths (List[str]/List[Path]) to search in (defaults to current directory).
+            logic:
+                - "or" (default): match any term (A OR B OR C)
+                - "and": match all terms in same file (A AND B AND C)
+                - "not": match first term but NOT any of the rest (A AND NOT B AND NOT C)
+            case_sensitive: If True, enable case-sensitive search (`-s`).
+            whole_word: If True, match whole words only (`-w`).
+            literal: If True, treat patterns as literal strings (`-F`). Applies to all terms.
+            regex: If False, implies `literal=True`.
+            max_depth: Maximum directory depth to search (`--max-depth`).
+            include: List of glob patterns to include (`-g`).
+            exclude: List of glob patterns to exclude (`-g '!...'`).
+            file_type: Search only files of given type (`-t <type>`), e.g., 'py', 'md'.
+            invert_match: Invert match (`-v`). Note: conflicts with `logic="not"`; ignored in that case.
+            count_only: Only output match counts per file (`-c`).
+            line_number: Show line numbers (`-n`, default True).
+            with_filename: Show filenames (`-H`, default True).
+            rank: If True, rerank results by relevance score.
+            rank_kwargs: Additional kwargs for ranking (see `_rerank_results`).
+            rga_no_cache: If True, disable rga caching (`--rga-no-cache`).
+            rga_cache_max_blob_len: Max blob length for rga cache (`--rga-cache-max-blob-len`). Defaults to 10MB.
+            rga_cache_path: Custom path for rga cache (`--rga-cache-path`).
+                If None, then set the path to `/path/to/your_work_path/.cache/rga`
+            timeout: Maximum time in seconds to wait for the search to complete.
+
+        Returns:
+            List of match objects (from `rga --json`), or list of {'path': str, 'count': int} if `count_only=True`.
+            For "and"/"not", matches correspond to the **last term** (or first, if only one) in qualifying files/lines.
+        """
+        results: List[Dict[str, Any]] = []
+
+        # Normalize terms
+        if isinstance(terms, str):
+            terms = [terms]
+        if not terms:
+            return results
+
+        rga_cache_path = rga_cache_path or self.rga_cache
+        rga_cache_path = str(Path(rga_cache_path).resolve())
+
+        # Multi-term logic routing
+        if logic == "or":
+            results, retrieve_pattern = await self._retrieve_or(
+                terms=terms,
+                path=path,
+                case_sensitive=case_sensitive,
+                whole_word=whole_word,
+                literal=literal,
+                regex=regex,
+                max_depth=max_depth,
+                include=include,
+                exclude=exclude,
+                file_type=file_type,
+                invert_match=invert_match,
+                count_only=count_only,
+                line_number=line_number,
+                with_filename=with_filename,
+                rga_no_cache=rga_no_cache,
+                rga_cache_max_blob_len=rga_cache_max_blob_len,
+                rga_cache_path=rga_cache_path,
+                timeout=timeout,
+            )
+        elif logic == "and":
+            results = await self._retrieve_and(
+                terms=terms,
+                path=path,
+                case_sensitive=case_sensitive,
+                whole_word=whole_word,
+                literal=literal,
+                regex=regex,
+                max_depth=max_depth,
+                include=include,
+                exclude=exclude,
+                file_type=file_type,
+                count_only=count_only,
+                line_number=line_number,
+                with_filename=with_filename,
+                match_same_line=False,  # file-level AND (most useful)
+                rga_no_cache=rga_no_cache,
+                rga_cache_max_blob_len=rga_cache_max_blob_len,
+                rga_cache_path=rga_cache_path,
+                timeout=timeout,
+            )
+        elif logic == "not":
+            if len(terms) < 2:
+                raise ValueError(
+                    "logic='not' requires at least two terms: [positive, negative1, ...]"
+                )
+            results = await self._retrieve_not(
+                positive=terms[0],
+                negatives=terms[1:],
+                path=path,
+                case_sensitive=case_sensitive,
+                whole_word=whole_word,
+                literal=literal,
+                regex=regex,
+                max_depth=max_depth,
+                include=include,
+                exclude=exclude,
+                file_type=file_type,
+                count_only=count_only,
+                line_number=line_number,
+                with_filename=with_filename,
+                rga_no_cache=rga_no_cache,
+                rga_cache_max_blob_len=rga_cache_max_blob_len,
+                rga_cache_path=rga_cache_path,
+                timeout=timeout,
+            )
+        else:
+            raise ValueError(
+                f"Unsupported logic: {logic}. Choose from 'and', 'or', 'not'."
+            )
+
+        # ====== Reranking Post-Processing ======
+        if rank and not count_only and results:
+            rank_kwargs = rank_kwargs or {}
+
+            def _default_text_extractor(match: Dict) -> str:
+                try:
+                    return match["data"]["lines"]["text"]
+                except (KeyError, TypeError):
+                    return ""
+
+            score_opts = {
+                "case_sensitive": case_sensitive,
+                "whole_word": whole_word,
+                "length_norm": rank_kwargs.get("length_norm", "linear"),
+                "base_length": rank_kwargs.get("base_length", 100),
+                "exact_bonus": rank_kwargs.get("exact_bonus", 2.0),
+                "tf_weight": rank_kwargs.get("tf_weight", 1.0),
+                "term_weights": rank_kwargs.get("term_weights", None),
+            }
+
+            # Reconstruct file groups: [begin, match*, end]*
+            grouped: List[List[Dict]] = []
+            current_group: List[Dict] = []
+
+            for item in results:
+                item_type = item.get("type")
+                if item_type == "begin":
+                    # Start new group
+                    if current_group:
+                        grouped.append(current_group)
+                    current_group = [item]
+                elif item_type == "end":
+                    # Close current group
+                    current_group.append(item)
+                    grouped.append(current_group)
+                    current_group = []
+                elif item_type == "match":
+                    # Accumulate in current group
+                    if current_group:  # defensive: should always be inside begin/end
+                        current_group.append(item)
+                    else:
+                        # Orphan match? Append to new dummy group (should not happen)
+                        current_group = [
+                            {"type": "begin", "data": {"path": {"text": "<unknown>"}}},
+                            item,
+                        ]
+                else:
+                    # e.g., "summary" — append to current or new group
+                    if current_group:
+                        current_group.append(item)
+                    else:
+                        grouped.append([item])
+
+            # If unclosed group remains (e.g., no final 'end'), flush it
+            if current_group:
+                grouped.append(current_group)
+
+            # Process each group: rerank only the 'match' items inside
+            new_results: List[Dict] = []
+
+            for group in grouped:
+                if not group:
+                    continue
+
+                # Identify begin / match / end segments
+                match_items = [g for g in group if g.get("type") == "match"]
+
+                # Rerank match items
+                scored_matches = []
+                for m in match_items:
+                    text = _default_text_extractor(m)
+                    score = self._calculate_relevance_score(
+                        text=text, terms=terms, **score_opts
+                    )
+                    new_m = {**m, "score": score}
+                    scored_matches.append((score, new_m))
+
+                # Sort descending by score
+                scored_matches.sort(key=lambda x: x[0], reverse=True)
+                reranked_matches = [item for _, item in scored_matches]
+
+                # Rebuild group in correct order:
+                # [begin] + [other non-match items in original order] + [reranked matches] + [end]
+                # But preserve *relative order* of non-match items (e.g., context lines)
+                # Simpler: walk original group, replace match list with reranked one
+                rebuilt_group = []
+                match_iter = iter(reranked_matches)
+                for item in group:
+                    if item.get("type") == "match":
+                        # Pull next from reranked list (should be same length)
+                        try:
+                            rebuilt_group.append(next(match_iter))
+                        except StopIteration:
+                            pass  # fallback: skip (should not happen)
+                    else:
+                        rebuilt_group.append(item)
+
+                new_results.extend(rebuilt_group)
+
+            results = new_results
+
+        return results
+
+    @staticmethod
+    def _run_rga(
+        args: List[str], json_output: bool = True
+    ) -> subprocess.CompletedProcess:
+        """Run ripgrep-all with given arguments.
+
+        Args:
+            args: List of ripgrep-all CLI arguments.
+            json_output: If True, forces `--json` and parses stdout as JSON Lines.
+
+        Returns:
+            CompletedProcess object with parsed stdout (as list of dicts if json_output=True).
+        """
+        cmd = ["rga", "--no-config"]  # disable user config for reproducibility
+        if json_output:
+            cmd.append("--json")
+        cmd.extend(args)
+
+        try:
+            result = subprocess.run(
+                cmd,
+                stdout=subprocess.PIPE,
+                stderr=subprocess.PIPE,
+                text=True,
+                check=False,  # we handle non-zero exit codes manually
+            )
+
+            if result.returncode != 0:
+                if "ripgrep" in result.stderr.lower() or " rg " in result.stderr:
+                    raise RuntimeError(
+                        f"ripgrep-all depends on 'ripgrep' (rg), but it's missing: {result.stderr.strip()}"
+                    )
+                elif result.returncode > 1:
+                    raise RuntimeError(f"rga execution failed: {result.stderr.strip()}")
+
+            # Parse JSON Lines if requested
+            if json_output and result.returncode in (0, 1) and result.stdout.strip():
+                lines = result.stdout.strip().splitlines()
+                result.stdout = [json.loads(line) for line in lines if line]
+            return result
+        except FileNotFoundError:
+            raise RuntimeError(
+                "ripgrep-all ('rga') not found. Please install ripgrep-all first."
+            )
+        except json.JSONDecodeError as e:
+            raise RuntimeError(
+                f"Failed to parse ripgrep-all JSON output: {e}\nRaw output: {result.stdout}"
+            )
+
+    @staticmethod
+    async def _run_rga_async(
+            args: List[str], json_output: bool = True, timeout: float = 60.0
+    ) -> Dict[str, Any]:
+        cmd = ["rga", "--no-config"]
+        if json_output:
+            cmd.append("--json")
+        cmd.extend(args)
+
+        try:
+            await asyncio.wait_for(RGA_SEMAPHORE.acquire(), timeout=timeout)
+        except asyncio.TimeoutError:
+            raise RuntimeError(
+                f"rga search timed out while waiting for a queue slot ({timeout}s)."
+            )
+
+        try:
+            try:
+                process = await asyncio.create_subprocess_exec(
+                    *cmd, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE
+                )
+            except FileNotFoundError:
+                raise RuntimeError("ripgrep-all ('rga') not found. Please install it first.")
+
+            try:
+                stdout, stderr = await asyncio.wait_for(
+                    process.communicate(), timeout=timeout
+                )
+
+                stdout_str = stdout.decode().strip()
+                stderr_str = stderr.decode().strip()
+                returncode = process.returncode
+
+                if returncode != 0:
+                    if "ripgrep" in stderr_str.lower() or " rg " in stderr_str:
+                        raise RuntimeError(
+                            f"ripgrep-all depends on 'ripgrep' (rg), but it's missing or failed: {stderr_str}"
+                        )
+                    elif returncode > 1:
+                        raise RuntimeError(f"rga execution failed with code {returncode}: {stderr_str}")
+                # 4. Parse JSON Lines
+                parsed_stdout = stdout_str
+                if json_output and returncode in (0, 1) and stdout_str:
+                    try:
+                        parsed_stdout = [
+                            json.loads(line) for line in stdout_str.splitlines() if line
+                        ]
+                    except json.JSONDecodeError as e:
+                        raise RuntimeError(f"Failed to parse rga JSON output: {e}")
+
+                return {
+                    "returncode": returncode,
+                    "stdout": parsed_stdout,
+                    "stderr": stderr_str,
+                }
+            except asyncio.TimeoutError:
+                try:
+                    process.kill()
+                except ProcessLookupError:
+                    pass
+                raise RuntimeError(f"rga process execution timed out ({timeout}s).")
+
+        finally:
+            RGA_SEMAPHORE.release()
+
+    @staticmethod
+    async def _retrieve_single(**kwargs) -> List[Dict[str, Any]]:
+        """Wrapper for original single-pattern search (extracted for reuse)."""
+        pattern = kwargs.pop("pattern")
+        args = []
+
+        # Basic ripgrep-all args
+        regex = kwargs.get("regex", True)
+        literal = kwargs.get("literal", False)
+        case_sensitive = kwargs.get("case_sensitive", False)
+        whole_word = kwargs.get("whole_word", False)
+        invert_match = kwargs.get("invert_match", False)
+        count_only = kwargs.get("count_only", False)
+        line_number = kwargs.get("line_number", True)
+        with_filename = kwargs.get("with_filename", True)
+        max_depth = kwargs.get("max_depth")
+        include = kwargs.get("include")
+        exclude = kwargs.get("exclude")
+        file_type = kwargs.get("file_type")
+        path = kwargs.get("path")
+        timeout = kwargs.get("timeout", 60.0)
+
+        # Additional ripgrep-all args
+        rga_no_cache = kwargs.get("rga_no_cache", False)
+        rga_cache_max_blob_len = kwargs.get(
+            "rga_cache_max_blob_len", 10000000
+        )  # Default 10MB
+        rga_cache_path = kwargs.get("rga_cache_path")
+
+        # Build argument list
+        if not regex:
+            literal = True
+        if literal:
+            args.append("-F")
+        if case_sensitive:
+            args.append("-s")
+        else:
+            args.append("-i")
+        if whole_word:
+            args.append("-w")
+        if invert_match:
+            args.append("-v")
+        if count_only:
+            args.append("-c")
+        if not line_number:
+            args.append("--no-line-number")
+        if not with_filename:
+            args.append("--no-filename")
+        if max_depth is not None:
+            args.extend(["--max-depth", str(max_depth)])
+        if include:
+            for inc in include:
+                args.extend(["-g", inc])
+        if exclude:
+            for exc in exclude:
+                args.extend(["-g", f"!{exc}"])
+        if file_type:
+            args.extend(["-t", file_type])
+
+        if rga_no_cache:
+            args.append("--rga-no-cache")
+
+        args.extend([f"--rga-cache-max-blob-len={str(rga_cache_max_blob_len)}"])
+
+        if rga_cache_path:
+            args.extend([f"--rga-cache-path={str(rga_cache_path)}"])
+
+        args.append(pattern)
+
+        if path is not None:
+            if isinstance(path, (str, Path)):
+                args.append(str(path))
+            elif isinstance(path, list):
+                for p in path:
+                    args.append(str(p))
+            else:
+                raise TypeError(f"Unsupported type for 'path': {type(path)}")
+
+        # keys: returncode, stdout, stderr
+        result: Dict[str, Any] = await GrepRetriever._run_rga_async(
+            args=args,
+            json_output=not count_only,
+            timeout=timeout,
+        )
+
+        if result["returncode"] == 0:
+            if count_only:
+                counts = []
+                for line in result.stdout.strip().splitlines():
+                    if ":" in line:
+                        p, c = line.rsplit(":", 1)
+                        counts.append({"path": p, "count": int(c)})
+                return counts
+            else:
+                return result["stdout"]
+        elif result["returncode"] == 1:
+            return []
+        else:
+            raise RuntimeError(
+                f"ripgrep-all failed (exit {result['returncode']}): {result['stderr'].strip()}"
+            )
+
+    @staticmethod
+    async def _retrieve_or(
+        terms: List[str],
+        **kwargs,
+    ) -> (List[Dict[str, Any]], str):
+        """OR: Match any term — simply concatenate with | (ripgrep-all supports alternation)."""
+        # Escape terms if literal mode
+        literal = kwargs.get("literal", False)
+        if literal:
+            escaped_terms = [
+                term.replace("\\", "\\\\").replace("|", "\\|") for term in terms
+            ]
+            pattern = "|".join(escaped_terms)
+            kwargs["literal"] = True  # still use -F to avoid regex meta
+        else:
+            # Wrap each term in (?:...) to avoid precedence issues
+            pattern = "|".join(f"(?:{term})" for term in terms)
+            kwargs["literal"] = False
+
+        result = await GrepRetriever._retrieve_single(pattern=pattern, **kwargs)
+
+        return result, pattern
+
+    @staticmethod
+    async def _retrieve_and(
+        terms: List[str],
+        match_same_line: bool = False,
+        **kwargs,
+    ) -> List[Dict[str, Any]]:
+        """AND: All terms in same file (default) or same line."""
+        count_only = kwargs.get("count_only", False)
+
+        # Step 1: Get files containing first term
+        first_matches = await GrepRetriever._retrieve_single(pattern=terms[0], **kwargs)
+        if not first_matches:
+            return []
+
+        if match_same_line:
+            # Line-level AND: filter lines containing all terms
+            def line_contains_all(
+                line: str, others: List[str], case_sensitive: bool
+            ) -> bool:
+                if not case_sensitive:
+                    line = line.lower()
+                    others = [t.lower() for t in others]
+                return all(term in line for term in others)
+
+            case_sensitive = kwargs.get("case_sensitive", False)
+            others = terms[1:]
+            return [
+                m
+                for m in first_matches
+                if m["type"] == "match"
+                and line_contains_all(
+                    m["data"]["lines"]["text"], others, case_sensitive
+                )
+            ]
+        else:
+            # File-level AND (default)
+            files_with_first = {
+                m["data"]["path"]["text"] for m in first_matches if m["type"] == "match"
+            }
+            qualifying_files = set()
+
+            for f in files_with_first:
+                valid = True
+                for term in terms[1:]:
+                    res = await GrepRetriever._retrieve_single(
+                        pattern=term,
+                        path=f,
+                        count_only=count_only,
+                        **{
+                            k: v
+                            for k, v in kwargs.items()
+                            if k not in ["path", "count_only"]
+                        },
+                    )
+                    if not res:  # no match
+                        valid = False
+                        break
+                if valid:
+                    qualifying_files.add(f)
+
+            # Collect matches for last term (or first) in qualifying files
+            target_term = terms[-1]
+            all_matches = []
+            kwargs["pattern"] = target_term
+            for f in qualifying_files:
+                kwargs["path"] = f
+                matches = await GrepRetriever._retrieve_single(**kwargs)
+                all_matches.extend(matches)
+            return all_matches
+
+    @staticmethod
+    async def _retrieve_not(
+        positive: str,
+        negatives: List[str],
+        **kwargs,
+    ) -> List[Dict[str, Any]]:
+        """NOT: Match positive term, but exclude files/lines containing any negative term."""
+        # count_only = kwargs.get("count_only", False)
+
+        # Step 1: Get matches for positive term
+        pos_matches = await GrepRetriever._retrieve_single(pattern=positive, **kwargs)
+        if not pos_matches:
+            return []
+
+        # Decide: file-level NOT (default) vs line-level NOT
+        # We use file-level for efficiency (avoid per-line Python filtering on large outputs)
+        files_with_positive = {
+            m["data"]["path"]["text"] for m in pos_matches if m["type"] == "match"
+        }
+        excluded_files = set()
+
+        for f in files_with_positive:
+            for neg in negatives:
+                res = await GrepRetriever._retrieve_single(
+                    pattern=neg,
+                    path=f,
+                    count_only=True,
+                    **{
+                        k: v
+                        for k, v in kwargs.items()
+                        if k not in ["path", "count_only"]
+                    },
+                )
+                if res:  # found negative → exclude this file
+                    excluded_files.add(f)
+                    break
+
+        # Keep only matches from non-excluded files
+        kept_matches = [
+            m
+            for m in pos_matches
+            if m["type"] == "match" and m["data"]["path"]["text"] not in excluded_files
+        ]
+
+        return kept_matches
+
+    async def list_files(
+        self,
+        path: Optional[str] = None,
+        *,
+        max_depth: Optional[int] = None,
+        include: Optional[List[str]] = None,
+        exclude: Optional[List[str]] = None,
+        file_type: Optional[str] = None,
+        hidden: bool = False,
+        follow_symlinks: bool = False,
+    ) -> List[str]:
+        """List files that would be searched by ripgrep-all (like `rga --files`).
+
+        Args:
+            path: Path to list files in.
+            max_depth: Maximum directory depth.
+            include: Glob patterns to include.
+            exclude: Glob patterns to exclude.
+            file_type: Restrict to file type (e.g., 'py').
+            hidden: Include hidden files/dirs (`--hidden`).
+            follow_symlinks: Follow symbolic links (`--follow`).
+
+        Returns:
+            List of relative file paths (strings).
+        """
+        args = ["--files"]
+        if max_depth is not None:
+            args.extend(["--max-depth", str(max_depth)])
+        if include:
+            for inc in include:
+                args.extend(["-g", inc])
+        if exclude:
+            for exc in exclude:
+                args.extend(["-g", f"!{exc}"])
+        if file_type:
+            args.extend(["-t", file_type])
+        if hidden:
+            args.append("--hidden")
+        if follow_symlinks:
+            args.append("--follow")
+        if path:
+            args.append(path)
+
+        result: Dict[str, Any] = await GrepRetriever._run_rga_async(
+            args, json_output=False
+        )
+        if result["returncode"] not in (0, 1):
+            raise RuntimeError(
+                f"ripgrep-all --files failed: {result['stderr'].strip()}"
+            )
+
+        return result["stdout"].strip().splitlines() if result["stdout"].strip() else []
+
+    def file_types(self) -> Dict[str, List[str]]:
+        """List supported file types and their associated globs/extensions.
+
+        Returns:
+            Dict mapping type names (e.g., 'python') to list of globs (e.g., ['*.py', '*.pyi']).
+        """
+        result = subprocess.run(
+            ["rga", "--type-list"],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            text=True,
+            check=True,
+        )
+        types = {}
+        for line in result.stdout.strip().splitlines():
+            if ":" in line:
+                name, globs = line.split(":", 1)
+                types[name.strip()] = [g.strip() for g in globs.split(",") if g.strip()]
+        return types
+
+    async def replace(
+        self,
+        pattern: str,
+        replacement: str,
+        path: Optional[str] = None,
+        *,
+        dry_run: bool = False,
+        case_sensitive: bool = False,
+        literal: bool = False,
+        whole_word: bool = False,
+        max_depth: Optional[int] = None,
+        include: Optional[List[str]] = None,
+        exclude: Optional[List[str]] = None,
+    ) -> List[Dict[str, Any]]:
+        """Perform search-and-replace using ripgrep-all (via `--replace`).
+
+        Caution: This modifies files in-place if dry_run=False.
+
+        Args:
+            pattern: Regex pattern to search for.
+            replacement: Replacement string (supports $1, $2, etc.).
+            path: Path to operate on.
+            dry_run: If True, only show matches/replacements, don't modify files.
+            case_sensitive: Enable case-sensitive matching.
+            literal: Treat pattern as literal string.
+            whole_word: Match whole words only.
+            max_depth: Max search depth.
+            include/exclude: Globs to include/exclude.
+
+        Returns:
+            List of replacement events (from `rga --json` output).
+        """
+        args = ["--replace", replacement]
+        if dry_run:
+            args.append("--dry-run")
+        else:
+            args.append("--passthru")  # needed for in-place replace with --replace
+        if case_sensitive:
+            args.append("-s")
+        else:
+            args.append("-i")
+        if literal:
+            args.append("-F")
+        if whole_word:
+            args.append("-w")
+        if max_depth is not None:
+            args.extend(["--max-depth", str(max_depth)])
+        if include:
+            for inc in include:
+                args.extend(["-g", inc])
+        if exclude:
+            for exc in exclude:
+                args.extend(["-g", f"!{exc}"])
+
+        args.append(pattern)
+        if path:
+            args.append(path)
+
+        result = await GrepRetriever._run_rga_async(args)
+        if result["returncode"] not in (0, 1):
+            raise RuntimeError(
+                f"ripgrep-all replace failed: {result['stderr'].strip()}"
+            )
+
+        return result["stdout"]
+
+    def version(self) -> str:
+        """Get ripgrep-all version string.
+
+        Returns:
+            Version string.
+        """
+        result = subprocess.run(
+            ["rga", "--version"],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            text=True,
+            check=True,
+        )
+        return result.stdout.strip().split("\n")[0]
+
+    def supports_feature(self, feature: str) -> bool:
+        """
+        Check if ripgrep-all supports a given feature (e.g., 'pcre2', 'json').
+
+        Args:
+            feature: Feature name (e.g., 'json', 'pcre2', 'lz4').
+
+        Returns:
+            True if feature is available in this rga build.
+        """
+        result = subprocess.run(
+            ["rga", "--help"],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            text=True,
+            check=True,
+        )
+        return f"--{feature}" in result.stdout
+
+    @staticmethod
+    def _calculate_relevance_score(
+        text: str,
+        terms: List[str],
+        *,
+        case_sensitive: bool = False,
+        whole_word: bool = False,
+        length_norm: Literal["linear", "log", "none"] = "linear",
+        base_length: int = 100,
+        exact_bonus: float = 2.0,
+        tf_weight: float = 1.0,
+        term_weights: Optional[List[float]] = None,
+        tf_saturation: Literal["log", "sigmoid", "none"] = "sigmoid",
+        saturation_k: float = 1.0,
+        idf_simulate: bool = True,
+    ) -> float:
+        """
+        Compute a relevance score for a text w.r.t. a list of query terms.
+
+        Scoring formula (per term, then summed):
+            term_score = (TF_term ** tf_weight) * bonus_term * length_factor
+
+        Where:
+            - TF_term = number of matches for the term
+            - bonus_term = exact_bonus if at least one match is *isolated*, else 1.0
+            - length_factor = global penalty for long texts (shared across terms)
+
+        An *isolated* match means the term is surrounded by non-alphanumeric characters
+        or string boundaries (i.e., standalone token — higher relevance).
+
+        Args:
+            text: Text to score (e.g., matching line or surrounding context).
+            terms: List of query terms (e.g., ["TODO", "fix"]).
+            case_sensitive: Whether matching is case-sensitive.
+            whole_word: If True, only matches bounded by non-word chars are counted.
+            length_norm: Length penalty strategy ('linear' / 'log' / 'none').
+            base_length: Scaling for linear norm (default 100 chars).
+            exact_bonus: Bonus multiplier for isolated matches (e.g., 2.0).
+            tf_weight: Exponent for term frequency (e.g., 0.5 for sqrt(TF)).
+            term_weights: Optional weights for each term (default: uniform 1.0).
+                         Must have same length as `terms` if provided.
+            tf_saturation: How to saturate term frequency:
+                - 'log':    tf_adj = 1 + log(tf)               (smooth, BM25-like)
+                - 'sigmoid':tf_adj = tf / (tf + k)             (bounded [0,1))
+                - 'none':   tf_adj = tf                         (original)
+            saturation_k: Parameter for sigmoid (default 1.0); larger → slower saturation.
+            idf_simulate: If True, penalize short/common terms heuristically:
+                idf_factor = max(0.1, min(1.0, len(term) / 5))   # e.g., "a" → 0.2, "error" → 1.0
+
+        Returns:
+            Non-negative float relevance score (sum of term scores).
+        """
+        if not text or not terms:
+            return 0.0
+
+        if term_weights is not None:
+            if len(term_weights) != len(terms):
+                raise ValueError("term_weights must have same length as terms")
+        else:
+            term_weights = [1.0] * len(terms)
+
+        flags = 0 if case_sensitive else re.IGNORECASE
+
+        # Precompute length factor (shared)
+        n = len(text)
+        if length_norm == "linear":
+            length_factor = 1.0 / (n / base_length + 1)
+        elif length_norm == "log":
+            length_factor = 1.0 / (math.log(n + 1) + 1)
+        else:  # "none"
+            length_factor = 1.0
+
+        total_score = 0.0
+
+        for term, weight in zip(terms, term_weights):
+            if not term:
+                continue
+
+            if idf_simulate:
+                # Normalize: len=1→0.2, len=5+→1.0 (clamped)
+                idf_factor = max(0.2, min(1.0, len(term) / 5.0))
+            else:
+                idf_factor = 1.0
+
+            escaped = re.escape(term)
+            if whole_word:
+                regex = rf"(?<!\w){escaped}(?!\w)"
+            else:
+                regex = escaped
+
+            # Find matches
+            try:
+                matches_iter = re.finditer(regex, text, flags=flags)
+                match_positions: List[int] = [m.start() for m in matches_iter]
+                tf = len(match_positions)
+            except re.error as e:
+                logger.warning(f"Regex failed for term {term!r}: {e}", RuntimeWarning)
+                tf = 0
+
+            if tf == 0:
+                continue
+
+            # Isolation bonus
+            has_isolated_match = any(
+                (pos == 0 or not text[pos - 1].isalnum())
+                and (
+                    pos + len(term) == len(text) or not text[pos + len(term)].isalnum()
+                )
+                for pos in match_positions
+            )
+            bonus = exact_bonus if has_isolated_match else 1.0
+
+            if tf_saturation == "log":
+                tf_adj = 1.0 + math.log(tf)  # log(1)=0 → tf=1 → tf_adj=1.0
+            elif tf_saturation == "sigmoid":
+                tf_adj = tf / (tf + saturation_k)  # e.g., k=1: tf=1→0.5, tf=10→0.91
+            elif tf_saturation == "none":
+                tf_adj = float(tf)
+            else:
+                raise ValueError(f"Unknown tf_saturation: {tf_saturation}")
+
+            # Term score with saturation + IDF + bonus + weight
+            term_score = (tf_adj**tf_weight) * bonus * weight * idf_factor
+            total_score += term_score
+
+        score = total_score * length_factor
+        return max(0.0, score)
+
+    @staticmethod
+    def merge_results(
+        raw_results: List[Dict[str, Any]], limit: int = 50
+    ) -> List[Dict[str, Any]]:
+        """
+        Merge ripgrep-all --json output into a structured per-file result list.
+
+        This function:
+          - Groups 'match' entries by file path (bounded by 'begin' and 'end' events).
+          - For each file, collects 'match' items and sorts them by score (desc).
+          - Takes top-`limit` matches per file (default: 50).
+          - Combines all lines.text from selected matches into a list (in original match order).
+          - Returns a list of unified file results.
+
+        Args:
+            raw_results: List of parsed JSON objects from `rga --json` output.
+            limit: Maximum number of match items to keep per file (default: 50).
+
+        Returns:
+            A list of dictionaries, each representing one file with:
+              - "path": str
+              - "matches": List[Dict]  # top `limit` match items (sorted by score desc)
+              - "lines": List[str]     # lines.text from those matches, in match order
+              - "total_matches": int   # total matches found in this file (before limit)
+        """
+        if not raw_results:
+            return []
+
+        # State tracking
+        current_path: Optional[str] = None
+        file_matches: List[Dict[str, Any]] = []  # matches for current file
+        all_files: List[Dict[str, Any]] = []  # final result accumulator
+
+        for item in raw_results:
+            item_type = item.get("type")
+            data = item.get("data", {})
+
+            if item_type == "begin":
+                # Start a new file context
+                path_obj = data.get("path", {})
+                current_path = path_obj.get("text")
+                file_matches = []  # reset match buffer
+
+            elif item_type == "match" and current_path is not None:
+                # Accumulate match; retain full item for sorting & line extraction
+                # Note: 'score' is top-level in your example (not in 'data')
+                file_matches.append(item)
+
+            elif item_type == "end":
+                # Finalize current file
+                if current_path is not None:
+                    # Sort matches by score (descending); assume score exists
+                    file_matches.sort(key=lambda x: x.get("score", 0.0), reverse=True)
+
+                    total_count = len(file_matches)
+                    top_matches = file_matches[:limit]
+
+                    # Extract lines.text in match order (not sorted order if stable sort not guaranteed)
+                    # But since we sort, we preserve sorted order → lines follow score order
+                    lines = [
+                        match["data"]["lines"]["text"]
+                        for match in top_matches
+                        if "data" in match and "lines" in match["data"]
+                    ]
+
+                    all_files.append(
+                        {
+                            "path": current_path,
+                            "matches": top_matches,  # full match objects
+                            "lines": lines,  # list of line strings
+                            "total_matches": total_count,  # before limiting
+                            "total_score": 0.0,
+                        }
+                    )
+
+                # Reset
+                current_path = None
+                file_matches = []
+
+            # Ignore "summary" and unknown types
+
+        return all_files
+
+
+class TextRetriever(GrepRetriever):
+    """Alias for GrepRetriever for backward compatibility."""
+
+    pass