codebase-index 2.1.1__tar.gz → 2.1.3__tar.gz

codebase_index-2.1.3/.claude/skills/codebase-index/.skill_version

@@ -0,0 +1 @@
+2.1.3

{codebase_index-2.1.1 → codebase_index-2.1.3}/.claude-plugin/plugin.json

@@ -3,7 +3,7 @@
   "name": "codebase-index",
   "displayName": "Codebase Index",
   "description": "Give Claude a precise local map of your codebase: find implementations, trace behavior, and predict change impact with file-line evidence.",
-  "version": "2.1.1",
+  "version": "2.1.3",
   "author": {
     "name": "codebase-index contributors"
   },

codebase_index-2.1.3/.codex/skills/codebase-index/.skill_version

@@ -0,0 +1 @@
+2.1.3

codebase_index-2.1.3/.opencode/skills/codebase-index/.skill_version

@@ -0,0 +1 @@
+2.1.3

{codebase_index-2.1.1 → codebase_index-2.1.3}/CHANGELOG.md

@@ -6,6 +6,42 @@ All notable changes to this project are documented here. The format is based on
 
 ## [Unreleased]
 
+## [2.1.3] - 2026-11-02
+
+### Fixed
+
+- Removing files from a large index no longer takes hours: `chunks.symbol_id` and
+  `symbols.parent_id` are now indexed, so a deleted symbol no longer costs a full scan
+  of both tables. Existing indexes gain the two indexes the next time they are opened,
+  with no rebuild.
+- The `PreToolUse` guard now covers Claude Code's PowerShell tool:
+  - a `codebase-index` command run from PowerShell turns the guard off for the session,
+    and `Select-String` or `rg` code searches there are guarded like their Bash
+    equivalents. A `Select-String` that reads piped input is left alone, and here-string
+    bodies (`@'...'@`) are not scanned;
+  - `install-hooks` refreshes the matcher of an installed guard. Run it again after
+    upgrading; the plugin's hooks pick this up on their own.
+- The guard no longer intercepts searches that cannot be answered by the index: those
+  over `.git` internals, binaries (`.exe`, `.dll`, `.jar`, `.sqlite` and the like), or
+  absolute paths outside the indexed repository. Nor does it intercept anything in a
+  directory without an index.
+
+## [2.1.2] - 2026-09-25
+
+### Fixed
+
+- The `PreToolUse` guard no longer blocks work it should let through:
+  - the first `codebase-index` command of a session turns the guard off even when it
+    is the command that builds the index; before, a repository without an index kept
+    the guard on for the whole session. Session state now lives in the system temp
+    directory (`CBX_HOOK_STATE` overrides it) instead of inside the index;
+  - heredoc bodies (`cat > script.ps1 <<'EOF'`, `python - <<EOF`) are file content, not
+    commands, and are no longer scanned for `grep`/`Select-String`;
+  - a retry is matched by its search term, not by the whole tool call, so a rewritten
+    command or a new description for the same search goes through;
+  - a quoted alternation such as `grep "a\|b"` is read whole, and a search on its own
+    line of a multi-line command is recognised.
+
 ## [2.1.1] - 2026-09-25
 
 ### Added
@@ -886,7 +922,9 @@ Pooled over 305 queries (Python, Java, TypeScript), v1.8.0 → 1.9.0:
 - Hooks example + `watch` mode for keeping the index fresh without blocking the edit loop (M8).
 - `doctor`, `stats`, `clean` diagnostics/maintenance commands.
 
-[Unreleased]: https://github.com/denfry/codebase-index/compare/v2.1.1...HEAD
+[Unreleased]: https://github.com/denfry/codebase-index/compare/v2.1.3...HEAD
+[2.1.3]: https://github.com/denfry/codebase-index/compare/v2.1.2...v2.1.3
+[2.1.2]: https://github.com/denfry/codebase-index/compare/v2.1.1...v2.1.2
 [2.1.1]: https://github.com/denfry/codebase-index/compare/v2.1.0...v2.1.1
 [2.1.0]: https://github.com/denfry/codebase-index/compare/v2.0.0...v2.1.0
 [2.0.0]: https://github.com/denfry/codebase-index/compare/v1.9.0...v2.0.0

{codebase_index-2.1.1 → codebase_index-2.1.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codebase-index
-Version: 2.1.1
+Version: 2.1.3
 Summary: A precise local codebase map for AI coding agents: find, trace, and predict with file-line evidence.
 Project-URL: Homepage, https://github.com/denfry/codebase-index
 Project-URL: Documentation, https://github.com/denfry/codebase-index/tree/main/docs
@@ -69,6 +69,7 @@ Description-Content-Type: text/markdown
   <img src="https://img.shields.io/badge/MCP-stdio%20server-3fb950" alt="MCP stdio server">
   <img src="https://img.shields.io/badge/network-off%20by%20default-3fb950" alt="No network by default">
   <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-8b949e" alt="MIT license"></a>
+  <a href="https://discord.gg/mVZcNUAGZt"><img src="https://img.shields.io/badge/Discord-Join%20the%20community-5865F2?logo=discord&logoColor=white" alt="Discord"></a>
 </p>
 
 <p align="center">

{codebase_index-2.1.1 → codebase_index-2.1.3}/README.md

@@ -19,6 +19,7 @@
   <img src="https://img.shields.io/badge/MCP-stdio%20server-3fb950" alt="MCP stdio server">
   <img src="https://img.shields.io/badge/network-off%20by%20default-3fb950" alt="No network by default">
   <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-8b949e" alt="MIT license"></a>
+  <a href="https://discord.gg/mVZcNUAGZt"><img src="https://img.shields.io/badge/Discord-Join%20the%20community-5865F2?logo=discord&logoColor=white" alt="Discord"></a>
 </p>
 
 <p align="center">

{codebase_index-2.1.1 → codebase_index-2.1.3}/docs/SCHEMA.md

@@ -44,6 +44,7 @@ CREATE TABLE chunks (
     symbol_names  TEXT NOT NULL DEFAULT ''     -- denormalized symbol name, FTS-indexed (mirrored by triggers)
 );
 CREATE INDEX idx_chunks_file ON chunks(file_id);
+CREATE INDEX idx_chunks_symbol ON chunks(symbol_id);
 
 -- Symbols extracted via tree-sitter (definitions).
 CREATE TABLE symbols (
@@ -63,6 +64,7 @@ CREATE TABLE symbols (
 CREATE INDEX idx_symbols_name ON symbols(name);
 CREATE INDEX idx_symbols_file ON symbols(file_id);
 CREATE INDEX idx_symbols_kind ON symbols(kind);
+CREATE INDEX idx_symbols_parent ON symbols(parent_id);
 
 -- Graph edges: imports / calls / references / inheritance / dependencies.
 -- src is always a known symbol or file; dst may be unresolved (dst_id NULL, dst_name kept).

{codebase_index-2.1.1 → codebase_index-2.1.3}/docs/installer.md

@@ -110,7 +110,7 @@ pwsh ./install.ps1 -Target claude -InstallDir "D:\skills\codebase-index"
 **Pin to a branch or tag** (reproducibility and safety):
 
 ```sh
-sh install.sh --branch v2.1.1
+sh install.sh --branch v2.1.3
 ```
 
 ---

{codebase_index-2.1.1 → codebase_index-2.1.3}/hooks/hooks.json

@@ -17,7 +17,7 @@
     ],
     "PreToolUse": [
       {
-        "matcher": "Grep|Bash",
+        "matcher": "Grep|Bash|PowerShell",
         "hooks": [
           {
             "type": "command",

{codebase_index-2.1.1 → codebase_index-2.1.3}/requirements.lock

@@ -1,3 +1,3 @@
-codebase-index @ https://github.com/denfry/codebase-index/archive/refs/tags/v2.1.1.tar.gz
+codebase-index @ https://github.com/denfry/codebase-index/archive/refs/tags/v2.1.3.tar.gz
 tree-sitter==0.25.2
 tree-sitter-language-pack==1.8.1

{codebase_index-2.1.1 → codebase_index-2.1.3}/src/codebase_index/__init__.py

@@ -4,4 +4,4 @@ Public surface is intentionally small; the CLI (`codebase_index.cli:app`) is the
 See docs/ARCHITECTURE.md for the module map.
 """
 
-__version__ = "2.1.1"
+__version__ = "2.1.3"

{codebase_index-2.1.1 → codebase_index-2.1.3}/src/codebase_index/cli.py

@@ -1074,10 +1074,11 @@ def install_hooks(
 ) -> None:
     """Make Claude Code search through the index by default.
 
-    Adds a SessionStart note and a PreToolUse guard on Grep/Bash: the first code
-    search of a session, before any codebase-index command, is sent back with the
-    index command to run instead; repeating it lets it through. Both do nothing in
-    a directory without an index. Other hooks in the settings file are kept.
+    Adds a SessionStart note and a PreToolUse guard on Grep, Bash and PowerShell: the
+    first code search of a session, before any codebase-index command, is sent back
+    with the index command to run instead; repeating it lets it through. Both do
+    nothing in a directory without an index. Other hooks in the settings file are
+    kept. Run it again after an upgrade to refresh an installed guard.
     """
     from .config import find_root
     from . import scaffold
@@ -1092,7 +1093,7 @@ def install_hooks(
         verb = "removed from"
     else:
         changed = scaffold.install_guard_hooks(path)
-        verb = "added to"
+        verb = "added to or refreshed in"
     if changed:
         typer.echo(f"[install-hooks] {', '.join(changed)} {verb} {path}")
     else:

{codebase_index-2.1.1 → codebase_index-2.1.3}/src/codebase_index/hooks.py

@@ -4,19 +4,23 @@ Two events, both no-ops outside a repository that has an index:
 
 - ``session-start`` adds one short note to the session context: this repository
   is indexed, and which commands answer code questions.
-- ``guard`` (PreToolUse on Grep and Bash) intercepts a code search the first time
+- ``guard`` (PreToolUse on Grep, Bash and PowerShell) intercepts a code search the first time
   in a session, before any ``codebase-index`` command has run, and names the
-  index command to use instead. It is a nudge, not a wall: repeating the same
-  call lets it through, the first ``codebase-index`` command in the session turns
-  the guard off, and searches over docs, logs or config are never intercepted.
+  index command to use instead. It is a nudge, not a wall: searching for the
+  same term again lets it through, the first ``codebase-index`` command in the
+  session turns the guard off (even one that has yet to build the index), and
+  searches over docs, logs or config, or inside heredoc and here-string bodies,
+  are never intercepted.
 
 Measured on a 5.9k-file monorepo, an agent spent 19% fewer tokens with the index
 than with grep (tests/eval/results/2026-09-24-agent-pilot.md), but agents reach
-for Grep out of habit. The guard runs on every Bash call, so this module imports
+for Grep out of habit. The guard runs on every shell call, so this module imports
 only the standard library and does no index work beyond one small SQLite read.
 
 Entry point: ``codebase-index-hook <event>`` reads the hook JSON on stdin and
 prints the hook JSON reply (or nothing). ``CBX_GUARD=0`` disables the guard.
+Per-session state lives in the system temp directory (``CBX_HOOK_STATE``
+overrides it), so it exists before the repository has an index.
 """
 
 from __future__ import annotations
@@ -27,28 +31,48 @@ import os
 import re
 import sqlite3
 import sys
+import tempfile
 import time
 from pathlib import Path
 from typing import Any, Optional
 
 INDEX_REL = Path(".claude") / "cache" / "codebase-index" / "index.sqlite"
-_STATE_DIR = "hook-sessions"
+_STATE_DIR = "codebase-index-hooks"
 _STATE_TTL_S = 2 * 24 * 3600
+SHELL_TOOLS = ("Bash", "PowerShell")
 
 # A shell segment that searches file contents: the first command of the segment
-# (after `&&`, `||`, `;` or the start), so `ps aux | grep x` is not a repo search.
+# (after `&&`, `||`, `;`, a newline or the start), so `ps aux | grep x` is not a
+# repo search. Quoted arguments may hold `|`, as in grep "a\|b".
 _SEARCH_CMD = re.compile(
-    r"(?:^|&&|\|\||;)\s*(?:cd\s+\S+\s*&&\s*)?"
-    r"(?P<cmd>git\s+grep|grep|egrep|fgrep|rg|ag|ack|findstr|Select-String)\b(?P<args>[^|;&]*)",
+    r"(?:^|&&|\|\||;|\n)[ \t]*(?:cd\s+\S+\s*&&\s*)?"
+    r"(?P<cmd>git\s+grep|grep|egrep|fgrep|rg|ag|ack|findstr|Select-String)\b"
+    r"""(?P<args>(?:"[^"\n]*"|'[^'\n]*'|[^|;&\n"'])*)""",
     re.IGNORECASE,
 )
+# A heredoc body is file content (a script being written, stdin for python), not
+# commands; an unterminated one runs to the end of the command.
+_HEREDOC = re.compile(
+    r"""(?<!<)<<-?(?!<)[ \t]*(['"]?)([A-Za-z_][\w-]*)\1(?P<rest>[^\n]*)\n"""
+    r".*?(?:^[ \t]*\2[ \t]*$|\Z)",
+    re.DOTALL | re.MULTILINE,
+)
+# PowerShell's heredoc: the closing `'@` or `"@` starts its own line.
+_HERE_STRING = re.compile(r"""@(['"])[ \t]*\r?\n.*?^\1@""", re.DOTALL | re.MULTILINE)
 _INDEX_CMD = re.compile(r"(?:^|[\s;&|/\\\"'])(?:codebase-index|cbx)(?:\.exe|\.ps1)?\s+\w")
-# Searches over prose, logs and config are what grep is for; leave them alone.
+# Searches over prose, logs, config, binaries and git internals are what grep is for;
+# leave them alone.
 _NON_CODE = re.compile(
-    r"\.(?:md|mdx|rst|txt|log|ya?ml|json|toml|ini|cfg|conf|csv|xml|properties|lock|env|html?)\b"
-    r"|--include=\*?\.(?:md|txt|log|ya?ml|json|toml)",
+    r"\.(?:md|mdx|rst|txt|log|ya?ml|json|toml|ini|cfg|conf|csv|xml|properties|lock|env|html?"
+    r"|exe|dll|so|jar|class|bin|sqlite|db)\b"
+    r"|--include=\*?\.(?:md|txt|log|ya?ml|json|toml)"
+    r"|(?:^|[\s'\"/\\])\.git[\\/]",
     re.IGNORECASE,
 )
+# Shell words of a search's arguments: a quoted string or a bare word.
+_WORD = re.compile(r""""([^"]*)"|'([^']*)'|(\S+)""")
+# An absolute path: a drive (C:\, C:/), an MSYS drive or POSIX root (/c/, /), or home.
+_ABS_PATH = re.compile(r"^(?:[A-Za-z]:[\\/]|/|~(?:[\\/]|$)|\$HOME\b|\$\{HOME\})")
 _NON_CODE_TYPES = {"md", "markdown", "txt", "log", "yaml", "yml", "json", "toml", "xml",
                    "csv", "ini", "config"}
 
@@ -130,32 +154,38 @@ def guard(payload: dict) -> Optional[dict]:
         return None
     tool = payload.get("tool_name") or ""
     tool_input = payload.get("tool_input") or {}
-    if tool not in ("Grep", "Bash"):
+    if tool != "Grep" and tool not in SHELL_TOOLS:
         return None
-    root = find_index_root(_cwd(payload))
-    if root is None:
-        return None
-    session = str(payload.get("session_id") or "default")
-    state = _State(root, session)
+    state = _State(str(payload.get("session_id") or "default"))
 
-    if tool == "Bash":
-        command = str(tool_input.get("command") or "")
+    if tool in SHELL_TOOLS:
+        command = _HEREDOC.sub(lambda m: "<<" + m.group("rest"), str(tool_input.get("command") or ""))
+        if tool == "PowerShell":
+            command = _HERE_STRING.sub("''", command)
         if _INDEX_CMD.search(command):
+            # Checked before the index exists: the first search is what builds it.
             state.mark_used()
             return None
-        pattern = _shell_search_pattern(command)
+        root = find_index_root(_cwd(payload))
+        if root is None:
+            return None
+        pattern = _shell_search_pattern(command, root)
         if pattern is None:
             return None
     else:
-        if _grep_targets_non_code(tool_input):
+        root = find_index_root(_cwd(payload))
+        if root is None or _grep_targets_non_code(tool_input):
+            return None
+        target = str(tool_input.get("path") or "")
+        if target and _outside_repo([target], root):
             return None
         pattern = str(tool_input.get("pattern") or "")
 
     if state.used:
         return None
-    fingerprint = hashlib.sha1(
-        (tool + json.dumps(tool_input, sort_keys=True, default=str)).encode("utf-8")
-    ).hexdigest()
+    # Keyed by the search term, not the whole call: a retry rarely repeats the
+    # command byte for byte (a new description, another pipe, a regenerated script).
+    fingerprint = hashlib.sha1(_term(pattern).lower().encode("utf-8")).hexdigest()
     if state.was_denied(fingerprint):
         return None  # the retry after a nudge: the agent decided grep is right
     state.deny(fingerprint)
@@ -168,20 +198,63 @@ def guard(payload: dict) -> Optional[dict]:
     }
 
 
-def _shell_search_pattern(command: str) -> Optional[str]:
+def _shell_search_pattern(command: str, root: Path) -> Optional[str]:
     """The search term of a content search in `command`, or None if there is none."""
     for match in _SEARCH_CMD.finditer(command):
         args = match.group("args")
         if _NON_CODE.search(args):
             continue
-        quoted = re.search(r"""(['"])(.+?)\1""", args)
-        if quoted:
-            return quoted.group(2)
-        words = [w for w in args.split() if not w.startswith("-")]
-        return words[0] if words else ""
+        words = [next(g for g in m.groups() if g is not None) for m in _WORD.finditer(args)]
+        if match.group("cmd").lower() == "select-string":
+            pattern, paths = _select_string_terms(words)
+        else:
+            positional = [w for w in words if not w.startswith("-") and not w.isdigit()]
+            # The first positional word is the pattern, the rest are the paths searched.
+            paths = positional[1:]
+            quoted = re.search(r"""(['"])(.+?)\1""", args)
+            pattern = quoted.group(2) if quoted else (positional[0] if positional else "")
+        if not _outside_repo(paths, root):
+            return pattern
     return None
 
 
+def _select_string_terms(words: list[str]) -> tuple[str, list[str]]:
+    """The pattern and paths of a Select-String call, named or positional."""
+    pattern: list[str] = []
+    paths: list[str] = []
+    positional: list[str] = []
+    target = None
+    for word in words:
+        if target is not None:
+            target.extend(word.split(",") if target is paths else [word])
+            target = None
+        elif word.lower() in ("-path", "-literalpath", "-lp"):
+            target = paths
+        elif word.lower() == "-pattern":
+            target = pattern
+        elif not word.startswith("-") and not word.isdigit():
+            positional.append(word)
+    if not pattern and positional:
+        pattern.append(positional.pop(0))
+    return (pattern[0] if pattern else ""), paths or positional
+
+
+def _outside_repo(paths: list[str], root: Path) -> bool:
+    """True when every path is absolute and none lies inside the indexed repository."""
+    if not paths or not all(_ABS_PATH.match(p) for p in paths):
+        return False
+    base = os.path.normcase(os.path.abspath(root))
+    for raw in paths:
+        path = re.sub(r"^(?:~|\$HOME\b|\$\{HOME\})", lambda _m: str(Path.home()), raw)
+        msys = re.match(r"^/([a-zA-Z])(/.*)?$", path)
+        if os.name == "nt" and msys:
+            path = f"{msys.group(1)}:{msys.group(2) or '/'}"
+        full = os.path.normcase(os.path.abspath(path))
+        if full == base or full.startswith(base.rstrip(os.sep) + os.sep):
+            return False
+    return True
+
+
 def _grep_targets_non_code(tool_input: dict) -> bool:
     kind = str(tool_input.get("type") or "").lower()
     if kind in _NON_CODE_TYPES:
@@ -190,26 +263,31 @@ def _grep_targets_non_code(tool_input: dict) -> bool:
     return bool(_NON_CODE.search(target))
 
 
+def _term(pattern: str) -> str:
+    """A regex search pattern as plain query words."""
+    return " ".join(re.sub(r"""[\\^$()\[\]{}|*+?"']""", " ", pattern).split())[:80]
+
+
 def _reason(pattern: str) -> str:
-    term = re.sub(r"[\\^$()\[\]{}|*+?]", " ", pattern).strip() or "<what you are looking for>"
-    term = " ".join(term.split())[:80]
+    term = _term(pattern) or "<what you are looking for>"
     return (
         "codebase-index: this repository is indexed, and the index answers code "
         "searches with ranked, numbered lines in fewer tokens than grep. Run it first:\n"
         f'  codebase-index search "{term}" --compact     # where / how\n'
         '  codebase-index refs "Owner.member" --compact  # every call site, with callers\n'
         '  codebase-index symbol "Name"                 # a definition\n'
-        "If the index does not answer, repeat this exact call and it will run. Once any "
-        "codebase-index command has run in this session, searches are not intercepted."
+        "If the index does not answer, search for the same term again and it will run. "
+        "Once any codebase-index command has run in this session, searches are not "
+        "intercepted."
     )
 
 
 class _State:
     """Per-session guard state: whether the index was used, which calls were nudged."""
 
-    def __init__(self, root: Path, session: str) -> None:
+    def __init__(self, session: str) -> None:
         safe = re.sub(r"[^A-Za-z0-9_.-]", "_", session)[:80] or "default"
-        self.dir = root / INDEX_REL.parent / _STATE_DIR
+        self.dir = Path(os.environ.get("CBX_HOOK_STATE") or Path(tempfile.gettempdir()) / _STATE_DIR)
         self.path = self.dir / f"{safe}.json"
         self.data: dict[str, Any] = {"used": False, "denied": []}
         try:

{codebase_index-2.1.1 → codebase_index-2.1.3}/src/codebase_index/scaffold.py

@@ -292,7 +292,7 @@ GUARD_HOOKS: dict[str, dict] = {
                    "timeout": 5}],
     },
     "PreToolUse": {
-        "matcher": "Grep|Bash",
+        "matcher": "Grep|Bash|PowerShell",
         "hooks": [{"type": "command", "command": "codebase-index-hook guard", "timeout": 5}],
     },
 }
@@ -305,7 +305,9 @@ def _is_guard_entry(entry: dict) -> bool:
 def install_guard_hooks(settings_path: Path) -> list[str]:
     """Merge the index-first hooks into a Claude Code settings file.
 
-    Idempotent, and leaves every other hook in place. Returns the events added.
+    Idempotent, and leaves every other hook in place. An installed guard gets the
+    current matcher, so upgrading only takes running this again. Returns the events
+    added or refreshed.
     """
     settings: dict = {}
     if settings_path.exists():
@@ -314,9 +316,14 @@ def install_guard_hooks(settings_path: Path) -> list[str]:
     added = []
     for event, entry in GUARD_HOOKS.items():
         entries = hooks.setdefault(event, [])
-        if not any(_is_guard_entry(e) for e in entries):
+        installed = [e for e in entries if _is_guard_entry(e)]
+        if not installed:
             entries.append(json.loads(json.dumps(entry)))
             added.append(event)
+        elif "matcher" in entry and any(e.get("matcher") != entry["matcher"] for e in installed):
+            for existing in installed:
+                existing["matcher"] = entry["matcher"]
+            added.append(event)
     if added:
         settings_path.parent.mkdir(parents=True, exist_ok=True)
         settings_path.write_text(json.dumps(settings, indent=2) + "\n", encoding="utf-8")

{codebase_index-2.1.1 → codebase_index-2.1.3}/src/codebase_index/storage/schema.sql

@@ -35,6 +35,9 @@ CREATE TABLE IF NOT EXISTS symbols (
 CREATE INDEX IF NOT EXISTS idx_symbols_name ON symbols(name);
 CREATE INDEX IF NOT EXISTS idx_symbols_file ON symbols(file_id);
 CREATE INDEX IF NOT EXISTS idx_symbols_kind ON symbols(kind);
+-- Every deleted symbol looks up its children for ON DELETE SET NULL here and on
+-- chunks.symbol_id; without an index each lookup is a full table scan.
+CREATE INDEX IF NOT EXISTS idx_symbols_parent ON symbols(parent_id);
 
 CREATE TABLE IF NOT EXISTS chunks (
     id            INTEGER PRIMARY KEY,
@@ -52,6 +55,7 @@ CREATE TABLE IF NOT EXISTS chunks (
     symbol_names  TEXT NOT NULL DEFAULT ''
 );
 CREATE INDEX IF NOT EXISTS idx_chunks_file ON chunks(file_id);
+CREATE INDEX IF NOT EXISTS idx_chunks_symbol ON chunks(symbol_id);
 
 CREATE TABLE IF NOT EXISTS edges (
     id            INTEGER PRIMARY KEY,