copilot changes again

Author: qchapp

src/ai_agent/agent/tools/repo_info_tool.py

@@ -121,7 +121,17 @@ async def tool_repo_summary(input: RepoSummaryInput) -> RepoSummaryOutput:
         shared = await inflight
         return shared.model_copy(deep=True)
 
-    result = await _fetch_repo_summary(effective_url)
+    try:
+        result = await _fetch_repo_summary(effective_url)
+    except BaseException as exc:
+        await _REPO_INFO_LOCK.acquire()
+        try:
+            if not inflight.done():
+                inflight.set_exception(exc)
+            _REPO_INFO_INFLIGHT.pop(cache_key, None)
+        finally:
+            _REPO_INFO_LOCK.release()
+        raise
 
     await _REPO_INFO_LOCK.acquire()
     try:

src/ai_agent/agent/tools/search_alternative_tool.py

@@ -4,7 +4,7 @@
 from pydantic import BaseModel, Field
 
 from ai_agent.generator.schema import CandidateDoc
-from .utils import get_catalog_docs, get_known_names, get_pipeline
+from .utils import get_known_names, get_pipeline
 from .query_utils import append_format_tokens, normalize_formats, sanitize_retrieval_query
 
 

src/ai_agent/agent/tools/search_tool.py

@@ -4,7 +4,7 @@
 from pydantic import BaseModel, Field
 
 from ai_agent.generator.schema import CandidateDoc
-from .utils import get_catalog_docs, get_known_names, get_pipeline
+from .utils import get_known_names, get_pipeline
 from .query_utils import (
     append_format_tokens,
     normalize_formats,

src/ai_agent/catalog/sync.py

@@ -12,6 +12,7 @@
 import logging
 
 from ai_agent.utils.full_processing import full_processing
+from ai_agent.utils.config import get_retrieval_config
 
 import hashlib
 from ai_agent.retriever.software_doc import SoftwareDoc
@@ -21,6 +22,20 @@
 log = logging.getLogger("ai_agent.catalog.sync")
 
 
+def _build_embedder() -> LocalBGEEmbedder:
+    """Build a LocalBGEEmbedder from the current retrieval config."""
+    retrieval_cfg = get_retrieval_config()
+    embed_cfg = retrieval_cfg.get("embedder", {}) if isinstance(retrieval_cfg, dict) else {}
+    return LocalBGEEmbedder(
+        model_name=embed_cfg.get("model_name", "Qwen/Qwen3-Embedding-8B"),
+        device=embed_cfg.get("device"),
+        backend=embed_cfg.get("backend", "remote"),
+        base_url=embed_cfg.get("base_url", "https://inference-rcp.epfl.ch/v1"),
+        api_key_env=embed_cfg.get("api_key_env", "EPFL_API_KEY_EMBEDDER"),
+        timeout_s=float(embed_cfg.get("timeout_s", 20.0)),
+    )
+
+
 def _index_artifacts_present(index_dir: Path) -> bool:
     """Return True when minimal FAISS artifacts exist."""
     return (index_dir / "index.faiss").exists() and (index_dir / "meta.json").exists()
@@ -398,7 +413,7 @@ def sync_once(
         faiss_rebuilt = False
         faiss_delta: Dict[str, int] = {"added": 0, "updated": 0, "removed": 0}
         try:
-            embedder = LocalBGEEmbedder()
+            embedder = _build_embedder()
             VectorIndex.load(index_dir, embedder)
             log.info(
                 "Catalog unchanged (semantic sha1=%s); keeping FAISS index", digest[:12]
@@ -408,7 +423,7 @@ def sync_once(
                 "Catalog unchanged but FAISS index is missing/incompatible; rebuilding index (%s)",
                 e,
             )
-            embedder = LocalBGEEmbedder()
+            embedder = _build_embedder()
             idx = VectorIndex(embedder)
             items = [
                 IndexItem(id=d.name, doc=d) for d in docs if getattr(d, "name", None)
@@ -461,7 +476,7 @@ def sync_once(
             log.info("  changed (sample): %s", chg_s)
         log.info("Full diff written to %s", diff_path)
 
-    embedder = LocalBGEEmbedder()
+    embedder = _build_embedder()
     try:
         idx = VectorIndex.load(index_dir, embedder)
     except Exception as e:

src/ai_agent/utils/image_meta.py

@@ -1,5 +1,6 @@
 # utils/image_meta.py
 from __future__ import annotations
+from collections import OrderedDict
 from pathlib import Path
 from typing import Optional, List
 import threading
@@ -15,8 +16,7 @@
 # Avoids re-reading large files (e.g. TIFF stacks) on every retrieval call.
 # ---------------------------------------------------------------------------
 _META_CACHE_MAX = int(os.getenv("IMAGE_META_CACHE_MAX", "128"))
-_meta_cache: dict[tuple, str] = {}  # key -> result string
-_meta_cache_order: list[tuple] = []  # insertion-order for simple LRU eviction
+_meta_cache: OrderedDict[tuple, str] = OrderedDict()  # key -> result string (LRU order)
 _meta_cache_lock = threading.Lock()
 
 
@@ -31,19 +31,21 @@ def _meta_cache_key(p: Path) -> tuple:
 
 def _meta_cache_get(key: tuple) -> Optional[str]:
     with _meta_cache_lock:
-        return _meta_cache.get(key)
+        value = _meta_cache.get(key)
+        if value is not None:
+            _meta_cache.move_to_end(key)
+        return value
 
 
 def _meta_cache_set(key: tuple, value: str) -> None:
     with _meta_cache_lock:
         if key in _meta_cache:
+            _meta_cache.move_to_end(key)
             return
         _meta_cache[key] = value
-        _meta_cache_order.append(key)
-        # Evict oldest entries when over capacity
-        while len(_meta_cache_order) > _META_CACHE_MAX:
-            oldest = _meta_cache_order.pop(0)
-            _meta_cache.pop(oldest, None)
+        # Evict least-recently-used entries when over capacity
+        while len(_meta_cache) > _META_CACHE_MAX:
+            _meta_cache.popitem(last=False)
 
 # ---- small helpers -----------------------------------------------------------