feat: update search data to Lucide v1.7.0, exclude search DB from wheel

- Rebase onto main (Lucide v1.7.0), regenerate descriptions for 9 new
  icons, skip 18 removed brand icons (chromium, github, twitter, etc.)
- Exclude lucide-search.db from wheel via wheel-exclude (363 KB vs 5.3 MB)
- Filter stale JSONL entries against icons DB during build
- Read version metadata from icons DB instead of JSONL records
- Add --icons-db flag to `lucide build-search` CLI

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: mmacpherson

pyproject.toml

@@ -149,3 +149,5 @@ source-include = [
   "tests/**",
   "src/lucide/data/*.db"
 ]
+# Search DB is downloaded on first use; keep it out of the wheel
+wheel-exclude = ["**/lucide-search.db"]

src/lucide/build_search.py

@@ -484,58 +484,24 @@ def _ensure_search_tables(conn: sqlite3.Connection) -> None:
     conn.commit()
 
 
-def build_search_db(
-    jsonl_path: pathlib.Path,
+def _write_search_db(  # noqa: PLR0913
     search_db_path: pathlib.Path,
+    ordered_names: list[str],
+    records: dict[str, DescriptionRecord],
+    embeddings: list,
     clusters_path: pathlib.Path,
     *,
+    version: str | None = None,
     verbose: bool = False,
 ) -> None:
-    """Build the SQLite search database from descriptions, embeddings, and clusters.
-
-    Reads descriptions from *jsonl_path*, computes embeddings with fastembed,
-    loads cluster assignments from *clusters_path*, and writes everything to
-    *search_db_path*.  The DB is rebuilt from scratch each time.
-
-    Args:
-        jsonl_path: Input JSONL file with VLM descriptions.
-        search_db_path: Output SQLite database path.
-        clusters_path: JSON file with cluster assignments.
-        verbose: Verbose logging.
-    """
+    """Write descriptions, embeddings, clusters, and metadata to SQLite."""
     import numpy as np  # noqa: PLC0415
-    from fastembed import TextEmbedding  # noqa: PLC0415
-
-    records = load_descriptions_jsonl(jsonl_path)
-    if not records:
-        logger.warning("No descriptions found in %s", jsonl_path)
-        return
-
-    ordered_names = sorted(records.keys())
-    logger.info("Building search DB from %d descriptions", len(ordered_names))
-
-    # Build embedding input: name + tags + categories + description
-    def _embedding_text(rec: DescriptionRecord) -> str:
-        parts = [rec["name"].replace("-", " ")]
-        if rec["tags"]:
-            parts.append(f"Tags: {', '.join(rec['tags'])}")
-        if rec["categories"]:
-            parts.append(f"Categories: {', '.join(rec['categories'])}")
-        parts.append(rec["description"])
-        return f"{EMBEDDING_DOCUMENT_PREFIX}{'. '.join(parts)}"
-
-    texts = [_embedding_text(records[n]) for n in ordered_names]
-    embedder = TextEmbedding(model_name=DEFAULT_EMBEDDING_MODEL)
-    embeddings = list(embedder.embed(texts))
-    logger.info("Computed %d embeddings", len(embeddings))
 
-    # Write SQLite
     search_db_path.parent.mkdir(parents=True, exist_ok=True)
     conn = sqlite3.connect(search_db_path)
     try:
         _ensure_search_tables(conn)
 
-        # Clear and repopulate
         conn.execute("DELETE FROM icon_descriptions")
         conn.execute("DELETE FROM icon_embeddings")
 
@@ -568,11 +534,12 @@ def _embedding_text(rec: DescriptionRecord) -> str:
                     (icon_name, int(cid), theme),
                 )
 
-        # Metadata — use first record's lucide_version as representative
+        # Metadata
         first = records[ordered_names[0]]
+        resolved_version = version or first.get("lucide_version", "unknown")
         now = datetime.now(tz=timezone.utc).isoformat()
         for key, value in [
-            ("version", first.get("lucide_version", "unknown")),
+            ("version", resolved_version),
             ("embedding_model", DEFAULT_EMBEDDING_MODEL),
             ("embedding_dim", str(DEFAULT_EMBEDDING_DIM)),
             ("description_model", DEFAULT_VLM_MODEL),
@@ -593,6 +560,91 @@ def _embedding_text(rec: DescriptionRecord) -> str:
         logger.info("Search DB written: %s (%.0f KB)", search_db_path, size_kb)
 
 
+def _read_icons_db_info(
+    icons_db_path: pathlib.Path,
+) -> tuple[set[str], str | None]:
+    """Read icon names and version from the icons database."""
+    conn = sqlite3.connect(icons_db_path)
+    try:
+        names = {r[0] for r in conn.execute("SELECT name FROM icons").fetchall()}
+        row = conn.execute(
+            "SELECT value FROM metadata WHERE key = 'version'"
+        ).fetchone()
+        return names, row[0] if row else None
+    finally:
+        conn.close()
+
+
+def build_search_db(
+    jsonl_path: pathlib.Path,
+    search_db_path: pathlib.Path,
+    clusters_path: pathlib.Path,
+    *,
+    icons_db_path: pathlib.Path | None = None,
+    verbose: bool = False,
+) -> None:
+    """Build the SQLite search database from descriptions, embeddings, and clusters.
+
+    Reads descriptions from *jsonl_path*, computes embeddings with fastembed,
+    loads cluster assignments from *clusters_path*, and writes everything to
+    *search_db_path*.  The DB is rebuilt from scratch each time.
+
+    When *icons_db_path* is provided, only icons present in that database are
+    included and the version metadata is read from it.
+
+    Args:
+        jsonl_path: Input JSONL file with VLM descriptions.
+        search_db_path: Output SQLite database path.
+        clusters_path: JSON file with cluster assignments.
+        icons_db_path: Optional icons database for filtering and version.
+        verbose: Verbose logging.
+    """
+    from fastembed import TextEmbedding  # noqa: PLC0415
+
+    records = load_descriptions_jsonl(jsonl_path)
+    if not records:
+        logger.warning("No descriptions found in %s", jsonl_path)
+        return
+
+    # Filter to only icons present in the icons DB when provided
+    icons_version: str | None = None
+    if icons_db_path is not None:
+        valid_names, icons_version = _read_icons_db_info(icons_db_path)
+        skipped = sorted(set(records.keys()) - valid_names)
+        if skipped:
+            logger.info("Skipping %d stale descriptions: %s", len(skipped), skipped)
+        ordered_names = sorted(n for n in records if n in valid_names)
+    else:
+        ordered_names = sorted(records.keys())
+
+    logger.info("Building search DB from %d descriptions", len(ordered_names))
+
+    # Build embedding input: name + tags + categories + description
+    def _embedding_text(rec: DescriptionRecord) -> str:
+        parts = [rec["name"].replace("-", " ")]
+        if rec["tags"]:
+            parts.append(f"Tags: {', '.join(rec['tags'])}")
+        if rec["categories"]:
+            parts.append(f"Categories: {', '.join(rec['categories'])}")
+        parts.append(rec["description"])
+        return f"{EMBEDDING_DOCUMENT_PREFIX}{'. '.join(parts)}"
+
+    texts = [_embedding_text(records[n]) for n in ordered_names]
+    embedder = TextEmbedding(model_name=DEFAULT_EMBEDDING_MODEL)
+    embeddings = list(embedder.embed(texts))
+    logger.info("Computed %d embeddings", len(embeddings))
+
+    _write_search_db(
+        search_db_path,
+        ordered_names,
+        records,
+        embeddings,
+        clusters_path,
+        version=icons_version,
+        verbose=verbose,
+    )
+
+
 # ---------------------------------------------------------------------------
 # Combined pipeline
 # ---------------------------------------------------------------------------
@@ -638,6 +690,7 @@ def build_search_data(  # noqa: PLR0913
         jsonl_path,
         search_db_path,
         clusters_path,
+        icons_db_path=main_db_path,
         verbose=verbose,
     )
 

src/lucide/cli.py

@@ -490,9 +490,15 @@ def _cmd_build_search(args: argparse.Namespace) -> int:
         args.output or descriptions_file.parent / "lucide-search.db"
     )
 
+    icons_db = pathlib.Path(args.icons_db) if args.icons_db else None
+
     try:
         build_search_db(
-            descriptions_file, search_db, clusters_file, verbose=args.verbose
+            descriptions_file,
+            search_db,
+            clusters_file,
+            icons_db_path=icons_db,
+            verbose=args.verbose,
         )
         print(f"Search DB built → {search_db}")
     except Exception as e:
@@ -786,6 +792,11 @@ def main() -> int:
         help="Input JSON file with cluster assignments",
         required=True,
     )
+    build_parser.add_argument(
+        "--icons-db",
+        help="Icons database for filtering and version metadata",
+        default=None,
+    )
     build_parser.add_argument(
         "-o",
         "--output",

src/lucide/data/gemini-icon-descriptions.jsonl

@@ -1701,3 +1701,12 @@
 {"name": "zodiac-virgo", "description": "This icon depicts the astrological symbol for Virgo. It represents virginity, maidenhood, and the harvest. It also evokes concepts of precision, analysis, and earthiness, often associated with this star sign and horoscope.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "0.577.0", "tags": ["virgin", "maiden", "harvest", "precision", "earth", "analysis", "astrology", "star sign", "horoscope", "constellation", "celestial"], "categories": ["social", "emoji"], "timestamp": "2026-03-26T04:45:28.261407+00:00"}
 {"name": "zoom-in", "description": "This icon visually represents a magnifying glass with a plus symbol inside its lens. It symbolizes the action of increasing detail, focusing on specifics, or making something larger to improve visibility and understanding. It conveys an idea of enhancement and closer examination.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "0.577.0", "tags": ["magnifying glass", "plus"], "categories": ["accessibility", "layout", "design", "text", "photography"], "timestamp": "2026-03-26T04:45:27.741436+00:00"}
 {"name": "zoom-out", "description": "This icon depicts a magnifying glass with a minus sign inside, symbolizing the action of zooming out. It represents a reduction in scale, offering a broader perspective or a step back from fine details, often associated with exploration and comprehension.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "0.577.0", "tags": ["magnifying glass", "plus"], "categories": ["accessibility", "layout", "design", "text", "photography"], "timestamp": "2026-03-26T04:45:27.741498+00:00"}
+{"name": "map-pin-search", "description": "This icon visually represents a map pin with a magnifying glass superimposed over it. It signifies finding or locating specific points on a map, searching for destinations, or pinpointing a particular place with precision.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "1.7.0", "tags": ["location", "navigation", "travel", "waypoint", "marker", "drop"], "categories": ["text", "navigation", "travel", "account"], "timestamp": "2026-04-02T04:35:47.179475+00:00"}
+{"name": "beef-off", "description": "This icon visually depicts a stylized ear with sound waves emanating from it, crossed by a diagonal line. It represents the concept of muting, silencing, or turning off sound. It can signify a desire for quiet or a state of no audio.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "1.7.0", "tags": ["food", "dish", "restaurant", "course", "meal", "meat", "bbq", "steak", "vegetarian"], "categories": ["food-beverage"], "timestamp": "2026-04-02T04:35:47.199205+00:00"}
+{"name": "shield-cog-corner", "description": "This icon visually combines a shield and a cogwheel, suggesting robust security settings and advanced protection. It represents strong defense, vigilant system management, and the controlled application of safety measures for robust, secure operations.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "1.7.0", "tags": ["cybersecurity", "secure", "safety", "protection", "guardian", "armored", "armoured", "defense", "defence", "defender", "block", "threat", "prevention", "antivirus", "vigilance", "vigilant", "detection", "scan", "find", "strength", "strong", "tough", "invincible", "invincibility", "invulnerable", "undamaged", "audit", "admin", "verification", "crest", "shieldcog", "bravery", "knight", "foot soldier", "infantry", "trooper", "pawn", "battle", "war", "military", "army", "cadet", "scout"], "categories": ["account", "security", "development", "gaming", "shapes"], "timestamp": "2026-04-02T04:35:47.248178+00:00"}
+{"name": "radio-off", "description": "This icon visually represents a crossed-out signal radiating outwards, indicating the absence of a radio broadcast or signal. It conveys concepts of disconnection, muting, or the deactivation of live transmission and frequency.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "1.7.0", "tags": ["signal", "broadcast", "connectivity", "live", "frequency"], "categories": ["devices", "multimedia", "social"], "timestamp": "2026-04-02T04:35:47.251911+00:00"}
+{"name": "road", "description": "This icon depicts a stylized road or highway, with a dashed line dividing lanes. It represents travel, journey, direction, and progress, evoking concepts of navigation, routes, and transportation.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "1.7.0", "tags": ["road", "street", "highway", "route", "path", "transport", "traffic", "drive", "map"], "categories": ["transportation"], "timestamp": "2026-04-02T04:35:47.267447+00:00"}
+{"name": "shield-cog", "description": "This icon visually depicts a shield with a gear inside, symbolizing robust protection and mechanical or operational security. It represents concepts of defense, system integrity, administrative control, and the strength to withstand threats through vigilant oversight and proactive management.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "1.7.0", "tags": ["cybersecurity", "secure", "safety", "protection", "guardian", "armored", "armoured", "defense", "defence", "defender", "block", "threat", "prevention", "antivirus", "vigilance", "vigilant", "detection", "scan", "find", "strength", "strong", "tough", "invincible", "invincibility", "invulnerable", "undamaged", "audit", "admin", "verification", "crest", "bravery", "knight", "foot soldier", "infantry", "trooper", "pawn", "battle", "war", "military", "army", "cadet", "scout"], "categories": ["account", "security", "development", "gaming", "shapes"], "timestamp": "2026-04-02T04:35:47.351638+00:00"}
+{"name": "sport-shoe", "description": "This icon shows a stylized athletic shoe in motion, with a distinct sole and laces detail. It represents activities like running, training, and general fitness, evoking concepts of movement, speed, and active exercise.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "1.7.0", "tags": ["footwear", "sports", "running", "athletic", "shoe", "sneaker", "training", "exercise", "fitness"], "categories": ["sports"], "timestamp": "2026-04-02T04:35:47.456484+00:00"}
+{"name": "line-style", "description": "This icon visually represents a set of horizontal dashed lines, varying in length, with a solid line below them. It symbolizes different line styles, strokes, and borders, suggesting options for design or outlining elements. It conveys a sense of structure, pattern, and customizable aesthetics.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "1.7.0", "tags": ["line", "stroke", "style", "dashed", "border"], "categories": ["design", "tools"], "timestamp": "2026-04-02T04:35:47.466965+00:00"}
+{"name": "cctv-off", "description": "This icon shows a stylized camera lens crossed out by a diagonal line, indicating deactivation. It represents the absence of surveillance or recording, signifying a security system being offline, inactive, or disabled, suggesting a state of being unmonitored.", "model": "gemini-2.5-flash-lite", "prompt_template_hash": "7ae15a902cba", "lucide_version": "1.7.0", "tags": ["camera", "surveillance", "recording", "film", "videotape", "crime", "watching"], "categories": ["security", "devices", "communication", "connectivity", "photography"], "timestamp": "2026-04-02T04:35:47.629215+00:00"}