# ACB Adapter Bugs Discovered During Migration
**Phase 2.7 - Week 7 Day 6** (January 2025)
This document tracks bugs discovered in the ACB (Asynchronous Component Base) framework during the migration of session-buddy from direct DuckDB connections to ACB adapters.
## Bug #1: Graph Adapter SSL Property Shadowing
**Status:** ✅ RESOLVED (Fixed in ACB PR)
**Component:** `acb/adapters/graph/duckdb_pgq.py`
**Description:**
The `Graph` adapter class has both an `ssl_enabled` property from `BaseAdapter` and a local `self.ssl_enabled` attribute, causing property shadowing and `RecursionError`.
**Error:**
```python
RecursionError: maximum recursion depth exceeded
File: acb/adapters/graph/duckdb_pgq.py:41
Code: self.ssl_enabled = ssl_enabled # Calls property setter recursively
```
**Root Cause:**
```python
class Graph(BaseAdapter):
ssl_enabled: bool # Property from BaseAdapter
def __post_init__(self):
self.ssl_enabled = self.config.graph.get(..., False) # ERROR: recursion
```
**Resolution:**
Fixed in ACB repository. Property shadowing eliminated.
______________________________________________________________________
## Bug #2: Vector Search Array Dimension Type Cast
**Status:** ✅ RESOLVED (Fixed in ACB - session-buddy/Phase 2.7)
**Component:** `acb/adapters/vector/duckdb.py`
**Description:**
The vector search query builder uses `$1::FLOAT[]` without specifying array dimensions, causing DuckDB VSS extension functions to fail with type mismatch errors.
**Error:**
```
Binder Error: No function matches the given name and argument types
'array_distance(FLOAT[384], FLOAT[])'. You might need to add explicit type casts.
Candidate functions:
array_distance(FLOAT[ANY], FLOAT[ANY]) -> FLOAT
array_distance(DOUBLE[ANY], DOUBLE[ANY]) -> DOUBLE
```
**Location:** `acb/adapters/vector/duckdb.py` lines 155-179
**Problematic Code:**
```text
def _build_search_query(
self,
table_name: str,
select_fields: str,
filter_expr: dict[str, t.Any] | None,
limit: int,
) -> str:
"""Build the main search query with VSS."""
query = f"""
SELECT {safe_select_fields},
array_distance(vector, $1::FLOAT[]) as score # ❌ Missing dimension
FROM {safe_table_name}
"""
# ...
return query
```
**Fix Required:**
```text
def _build_search_query(
self,
table_name: str,
select_fields: str,
filter_expr: dict[str, t.Any] | None,
limit: int,
dimension: int,
) -> str: # Add dimension parameter
"""Build the main search query with VSS."""
query = f"""
SELECT {safe_select_fields},
array_cosine_similarity(vector, $1::FLOAT[{dimension}]) as score
FROM {safe_table_name}
"""
# ...
query += f" ORDER BY score DESC LIMIT {limit}" # DESC for similarity
return query
```
**Additional Issues:**
1. Uses `array_distance()` instead of `array_cosine_similarity()` - distance semantics are inverted (lower=better) vs similarity (higher=better)
1. Orders by ASC instead of DESC when using similarity metrics
1. Dimension information is available in `self.config.vector.default_dimension` but not passed to query builder
**Impact:**
- All vector searches fail and fall back to `_build_fallback_query()`
- Fallback query returns `score=0.0` for all results
- Semantic search completely non-functional
**Fix Applied** (session-buddy/Phase 2.7):
Modified `_build_search_query()` in `/Users/les/Projects/acb/acb/adapters/vector/duckdb.py`:
```python
def _build_search_query(
self,
table_name: str,
select_fields: str,
filter_expr: dict[str, t.Any] | None,
limit: int,
) -> str:
"""Build the main search query with VSS."""
safe_table_name = self._validate_table_name(table_name)
safe_select_fields = self._validate_select_fields(select_fields)
# Get dimension from config for proper type casting
dimension = self.config.vector.default_dimension
query = f"""
SELECT {safe_select_fields},
array_cosine_similarity(vector, $1::FLOAT[{dimension}]) as score
FROM {safe_table_name}
""" # Now includes {dimension} in cast and uses cosine_similarity
if filter_expr:
filter_conditions = self._build_filter_conditions(filter_expr)
if filter_conditions:
query += " WHERE " + " AND ".join(filter_conditions)
query += f" ORDER BY score DESC LIMIT {limit}" # DESC for similarity
return query
```
**Changes Made:**
1. Added `dimension = self.config.vector.default_dimension` to get dimension from config
1. Changed `$1::FLOAT[]` → `$1::FLOAT[{dimension}]` to include explicit array size
1. Changed `array_distance()` → `array_cosine_similarity()` for proper similarity semantics
1. Changed `ORDER BY score ASC` → `ORDER BY score DESC` (higher similarity = better)
**Verification:**
```bash
# Test that fixed ACB search() method returns proper scores
python3 -c "
import asyncio
from session_buddy.di import configure
from session_buddy.adapters.reflection_adapter import ReflectionDatabaseAdapter
configure(force=True)
async def test_fix():
async with ReflectionDatabaseAdapter() as db:
# Store test data
await db.store_conversation(
content='Python async patterns',
metadata={'project': 'test'}
)
# Search using ACB adapter (should work now)
results = await db.search_conversations(
query='Python async',
limit=5,
min_score=0.3
)
# Verify scores are in valid range [0.0, 1.0]
assert len(results) > 0, 'No results returned'
assert all(0.0 <= r['score'] <= 1.0 for r in results), 'Invalid scores'
print(f'✅ Fix verified: {len(results)} results with scores {[r[\"score\"] for r in results]}')
asyncio.run(test_fix())
"
# Output: ✅ Fix verified: 3 results with scores [0.8884, 0.8879, 0.8858]
```
______________________________________________________________________
## Summary
- **2 bugs discovered** during ACB adapter migration (Phase 2.7)
- **2 bugs resolved** ✅
- Graph adapter SSL property shadowing (fixed in ACB)
- Vector search array dimension casting (fixed in ACB - session-buddy/Phase 2.7)
**Impact on Migration:**
Both critical bugs have been resolved, allowing full ACB adapter integration:
- Vector adapter: ✅ Fully functional with semantic search
- Graph adapter: ✅ Ready for integration (SSL bug fixed)
**Phase 2 (Vector Migration) Status:**
- ACB adapter registration: ✅ Complete
- ReflectionDatabaseAdapter wrapper: ✅ Complete and tested
- All features working: Storage, search, statistics, filtering
**Next Steps:**
1. ✅ Complete Phase 2 vector migration
1. Begin Phase 3 graph adapter migration
1. Update memory_tools.py and search_tools.py to use new adapter
