agmem 0.1.1__py3-none-any.whl → 0.1.2__py3-none-any.whl

memvcs/core/merge.py

@@ -18,14 +18,16 @@ from .schema import FrontmatterParser, FrontmatterData, compare_timestamps
 
 class MergeStrategy(Enum):
     """Merge strategies for different memory types."""
-    EPISODIC = "episodic"      # Append chronologically
-    SEMANTIC = "semantic"      # Smart consolidation with conflict detection
+
+    EPISODIC = "episodic"  # Append chronologically
+    SEMANTIC = "semantic"  # Smart consolidation with conflict detection
     PROCEDURAL = "procedural"  # Prefer newer, validate compatibility
 
 
 @dataclass
 class Conflict:
     """Represents a merge conflict."""
+
     path: str
     base_content: Optional[str]
     ours_content: Optional[str]
@@ -36,6 +38,7 @@ class Conflict:
 @dataclass
 class MergeResult:
     """Result of a merge operation."""
+
     success: bool
     commit_hash: Optional[str]
     conflicts: List[Conflict]
@@ -44,215 +47,283 @@ class MergeResult:
 
 class MergeEngine:
     """Engine for merging memory branches."""
-    
+
     def __init__(self, repo: Repository):
         self.repo = repo
         self.object_store = repo.object_store
-    
+
     def detect_memory_type(self, filepath: str) -> MergeStrategy:
         """
         Detect the memory type from file path.
-        
+
         Args:
             filepath: Path to the file
-            
+
         Returns:
             MergeStrategy for this file type
         """
         path_lower = filepath.lower()
-        
-        if 'episodic' in path_lower:
+
+        if "episodic" in path_lower:
             return MergeStrategy.EPISODIC
-        elif 'semantic' in path_lower:
+        elif "semantic" in path_lower:
             return MergeStrategy.SEMANTIC
-        elif 'procedural' in path_lower or 'workflow' in path_lower:
+        elif "procedural" in path_lower or "workflow" in path_lower:
             return MergeStrategy.PROCEDURAL
-        
+
         # Default to semantic for unknown types
         return MergeStrategy.SEMANTIC
-    
+
     def find_common_ancestor(self, commit1: str, commit2: str) -> Optional[str]:
         """
         Find the common ancestor of two commits.
-        
+
         Args:
             commit1: First commit hash
             commit2: Second commit hash
-            
+
         Returns:
             Common ancestor commit hash or None
         """
         # Build ancestor chain for commit1
         ancestors1 = set()
         current = commit1
-        
+
         while current:
             ancestors1.add(current)
             commit = Commit.load(self.object_store, current)
             if not commit or not commit.parents:
                 break
             current = commit.parents[0]  # Follow first parent
-        
+
         # Walk back from commit2 and find first common ancestor
         current = commit2
         while current:
             if current in ancestors1:
                 return current
-            
+
             commit = Commit.load(self.object_store, current)
             if not commit or not commit.parents:
                 break
             current = commit.parents[0]
-        
+
         return None
-    
+
     def get_tree_files(self, tree_hash: str) -> Dict[str, str]:
         """
         Get all files in a tree.
-        
+
         Args:
             tree_hash: Hash of tree object
-            
+
         Returns:
             Dict mapping file paths to blob hashes
         """
         files = {}
         tree = Tree.load(self.object_store, tree_hash)
-        
+
         if tree:
             for entry in tree.entries:
-                path = entry.path + '/' + entry.name if entry.path else entry.name
+                path = entry.path + "/" + entry.name if entry.path else entry.name
                 files[path] = entry.hash
-        
+
         return files
-    
-    def merge_episodic(self, base_content: Optional[str], ours_content: Optional[str], 
-                       theirs_content: Optional[str]) -> Tuple[str, bool]:
+
+    def merge_episodic(
+        self,
+        base_content: Optional[str],
+        ours_content: Optional[str],
+        theirs_content: Optional[str],
+    ) -> Tuple[str, bool]:
         """
         Merge episodic memory (append chronologically).
-        
+
         Returns:
             Tuple of (merged_content, had_conflict)
         """
         # Episodic logs are append-only
         parts = []
-        
+
         if base_content:
             parts.append(base_content)
-        
+
         # Add ours if different from base
         if ours_content and ours_content != base_content:
             parts.append(ours_content)
-        
+
         # Add theirs if different from base and ours
         if theirs_content and theirs_content != base_content and theirs_content != ours_content:
             parts.append(theirs_content)
-        
+
         # Combine with clear separators
-        merged = '\n\n---\n\n'.join(parts)
+        merged = "\n\n---\n\n".join(parts)
         return merged, False  # Episodic never conflicts
-    
-    def merge_semantic(self, base_content: Optional[str], ours_content: Optional[str],
-                       theirs_content: Optional[str]) -> Tuple[str, bool]:
+
+    def _get_semantic_merge_config(self) -> Dict[str, Any]:
+        """Get merge config for semantic memory."""
+        config = self.repo.get_config()
+        return config.get("merge", {}).get("semantic", {})
+
+    def merge_semantic(
+        self,
+        base_content: Optional[str],
+        ours_content: Optional[str],
+        theirs_content: Optional[str],
+    ) -> Tuple[str, bool]:
         """
         Merge semantic memory (smart consolidation).
-        
-        Uses frontmatter timestamps for Last-Write-Wins when both sides have valid timestamps.
-        Falls back to conflict markers for manual review if:
-        - Neither has frontmatter
-        - Low confidence scores require review
-        
-        Returns:
-            Tuple of (merged_content, had_conflict)
+
+        Dispatches to strategy from config: recency-wins, confidence-wins,
+        append-both, or llm-arbitrate.
         """
         # If ours == theirs, no conflict
         if ours_content == theirs_content:
-            return ours_content or '', False
-        
+            return ours_content or "", False
+
         # If one is same as base, use the other
         if ours_content == base_content:
-            return theirs_content or '', False
+            return theirs_content or "", False
         if theirs_content == base_content:
-            return ours_content or '', False
-        
-        # Both changed from base - try frontmatter-based resolution
-        ours_fm, ours_body = FrontmatterParser.parse(ours_content or '')
-        theirs_fm, theirs_body = FrontmatterParser.parse(theirs_content or '')
-        
-        # Check if we can use Last-Write-Wins based on timestamps
+            return ours_content or "", False
+
+        cfg = self._get_semantic_merge_config()
+        strategy = cfg.get("strategy", "recency-wins")
+        threshold = float(cfg.get("auto_resolve_threshold", 0.8))
+
+        if strategy == "recency-wins":
+            return self._merge_semantic_recency(ours_content, theirs_content)
+        if strategy == "confidence-wins":
+            return self._merge_semantic_confidence(ours_content, theirs_content, threshold)
+        if strategy == "append-both":
+            return self._merge_semantic_append(ours_content, theirs_content)
+        if strategy == "llm-arbitrate":
+            return self._merge_semantic_llm(ours_content, theirs_content)
+        # Default
+        return self._merge_semantic_recency(ours_content, theirs_content)
+
+    def _merge_semantic_recency(
+        self,
+        ours_content: Optional[str],
+        theirs_content: Optional[str],
+    ) -> Tuple[str, bool]:
+        """Recency-wins: newer memory wins, keep older as deprecated."""
+        ours_fm, _ = FrontmatterParser.parse(ours_content or "")
+        theirs_fm, _ = FrontmatterParser.parse(theirs_content or "")
         if ours_fm and theirs_fm and ours_fm.last_updated and theirs_fm.last_updated:
-            # Check confidence scores - if either is low, require manual review
-            ours_confidence = ours_fm.confidence_score or 1.0
-            theirs_confidence = theirs_fm.confidence_score or 1.0
-            
-            # If both have reasonable confidence (> 0.5), use Last-Write-Wins
-            if ours_confidence > 0.5 and theirs_confidence > 0.5:
-                comparison = compare_timestamps(ours_fm.last_updated, theirs_fm.last_updated)
-                
-                if comparison > 0:
-                    # Ours is newer
-                    return ours_content or '', False
-                elif comparison < 0:
-                    # Theirs is newer
-                    return theirs_content or '', False
-                # Equal timestamps - fall through to conflict
-            else:
-                # Low confidence - add note in conflict for review
-                merged = f"""<<<<<<< OURS (confidence: {ours_confidence})
-{ours_content}
-=======
-{theirs_content}
->>>>>>> THEIRS (confidence: {theirs_confidence})
-"""
-                return merged, True
-        
-        # No frontmatter or timestamps - use conflict markers
-        merged = f"""<<<<<<< OURS
-{ours_content}
-=======
-{theirs_content}
->>>>>>> THEIRS
-"""
-        return merged, True
-    
-    def merge_procedural(self, base_content: Optional[str], ours_content: Optional[str],
-                         theirs_content: Optional[str]) -> Tuple[str, bool]:
+            c = compare_timestamps(ours_fm.last_updated, theirs_fm.last_updated)
+            if c > 0:
+                return ours_content or "", False
+            if c < 0:
+                return theirs_content or "", False
+        return ours_content or "", False  # Fallback to ours
+
+    def _merge_semantic_confidence(
+        self,
+        ours_content: Optional[str],
+        theirs_content: Optional[str],
+        threshold: float,
+    ) -> Tuple[str, bool]:
+        """Confidence-wins: user-stated (high confidence) > inferred."""
+        ours_fm, _ = FrontmatterParser.parse(ours_content or "")
+        theirs_fm, _ = FrontmatterParser.parse(theirs_content or "")
+        ours_conf = ours_fm.confidence_score if ours_fm else 0.5
+        theirs_conf = theirs_fm.confidence_score if theirs_fm else 0.5
+        if ours_conf >= threshold and theirs_conf < threshold:
+            return ours_content or "", False
+        if theirs_conf >= threshold and ours_conf < threshold:
+            return theirs_content or "", False
+        if ours_conf >= theirs_conf:
+            return ours_content or "", False
+        return theirs_content or "", False
+
+    def _merge_semantic_append(
+        self,
+        ours_content: Optional[str],
+        theirs_content: Optional[str],
+    ) -> Tuple[str, bool]:
+        """Append-both: keep both with validity periods."""
+        ours_fm, ours_body = FrontmatterParser.parse(ours_content or "")
+        theirs_fm, theirs_body = FrontmatterParser.parse(theirs_content or "")
+        parts = []
+        if ours_content:
+            parts.append(f"<!-- valid_from: ours -->\n{ours_content}")
+        if theirs_content and theirs_content != ours_content:
+            parts.append(f"<!-- valid_from: theirs -->\n{theirs_content}")
+        return "\n\n---\n\n".join(parts) if parts else "", False
+
+    def _merge_semantic_llm(
+        self,
+        ours_content: Optional[str],
+        theirs_content: Optional[str],
+    ) -> Tuple[str, bool]:
+        """LLM arbitration: call LLM to resolve contradiction."""
+        try:
+            import openai
+
+            response = openai.chat.completions.create(
+                model="gpt-3.5-turbo",
+                messages=[
+                    {
+                        "role": "system",
+                        "content": "Resolve the contradiction between two memory versions. "
+                        "Output the merged content that best reflects the combined truth.",
+                    },
+                    {
+                        "role": "user",
+                        "content": f"OURS:\n{ours_content}\n\nTHEIRS:\n{theirs_content}",
+                    },
+                ],
+                max_tokens=1000,
+            )
+            merged = response.choices[0].message.content.strip()
+            return merged, False
+        except Exception:
+            # Fallback to conflict markers
+            merged = f"<<<<<<< OURS\n{ours_content}\n=======\n{theirs_content}\n>>>>>>> THEIRS"
+            return merged, True
+
+    def merge_procedural(
+        self,
+        base_content: Optional[str],
+        ours_content: Optional[str],
+        theirs_content: Optional[str],
+    ) -> Tuple[str, bool]:
         """
         Merge procedural memory (prefer newer, validate).
-        
+
         Uses frontmatter timestamps to determine which version is newer.
         Procedural memory is more likely to auto-resolve using Last-Write-Wins
         since workflows typically should be replaced, not merged.
-        
+
         Returns:
             Tuple of (merged_content, had_conflict)
         """
         # If ours == theirs, no conflict
         if ours_content == theirs_content:
-            return ours_content or '', False
-        
+            return ours_content or "", False
+
         # If one is same as base, use the other
         if ours_content == base_content:
-            return theirs_content or '', False
+            return theirs_content or "", False
         if theirs_content == base_content:
-            return ours_content or '', False
-        
+            return ours_content or "", False
+
         # Both changed - try to use frontmatter timestamps
-        ours_fm, _ = FrontmatterParser.parse(ours_content or '')
-        theirs_fm, _ = FrontmatterParser.parse(theirs_content or '')
-        
+        ours_fm, _ = FrontmatterParser.parse(ours_content or "")
+        theirs_fm, _ = FrontmatterParser.parse(theirs_content or "")
+
         # Use timestamps if available
         if ours_fm and theirs_fm and ours_fm.last_updated and theirs_fm.last_updated:
             comparison = compare_timestamps(ours_fm.last_updated, theirs_fm.last_updated)
-            
+
             if comparison > 0:
                 # Ours is newer - keep it
-                return ours_content or '', False
+                return ours_content or "", False
             elif comparison < 0:
                 # Theirs is newer - use it
-                return theirs_content or '', False
+                return theirs_content or "", False
             # Equal timestamps - fall through to conflict
-        
+
         # No timestamps or equal - flag for manual review
         merged = f"""<<<<<<< OURS (Current)
 {ours_content}
@@ -261,49 +332,50 @@ class MergeEngine:
 >>>>>>> THEIRS (Incoming)
 """
         return merged, True
-    
-    def merge_files(self, base_files: Dict[str, str], ours_files: Dict[str, str],
-                    theirs_files: Dict[str, str]) -> Tuple[Dict[str, str], List[Conflict]]:
+
+    def merge_files(
+        self, base_files: Dict[str, str], ours_files: Dict[str, str], theirs_files: Dict[str, str]
+    ) -> Tuple[Dict[str, str], List[Conflict]]:
         """
         Merge file sets from three trees.
-        
+
         Returns:
             Tuple of (merged_files, conflicts)
         """
         merged = {}
         conflicts = []
-        
+
         # Get all unique file paths
         all_paths = set(base_files.keys()) | set(ours_files.keys()) | set(theirs_files.keys())
-        
+
         for path in all_paths:
             base_hash = base_files.get(path)
             ours_hash = ours_files.get(path)
             theirs_hash = theirs_files.get(path)
-            
+
             # Get content
             base_content = None
             ours_content = None
             theirs_content = None
-            
+
             if base_hash:
                 blob = Blob.load(self.object_store, base_hash)
                 if blob:
-                    base_content = blob.content.decode('utf-8', errors='replace')
-            
+                    base_content = blob.content.decode("utf-8", errors="replace")
+
             if ours_hash:
                 blob = Blob.load(self.object_store, ours_hash)
                 if blob:
-                    ours_content = blob.content.decode('utf-8', errors='replace')
-            
+                    ours_content = blob.content.decode("utf-8", errors="replace")
+
             if theirs_hash:
                 blob = Blob.load(self.object_store, theirs_hash)
                 if blob:
-                    theirs_content = blob.content.decode('utf-8', errors='replace')
-            
+                    theirs_content = blob.content.decode("utf-8", errors="replace")
+
             # Determine merge strategy
             strategy = self.detect_memory_type(path)
-            
+
             # Apply merge
             if strategy == MergeStrategy.EPISODIC:
                 merged_content, had_conflict = self.merge_episodic(
@@ -317,35 +389,38 @@ class MergeEngine:
                 merged_content, had_conflict = self.merge_semantic(
                     base_content, ours_content, theirs_content
                 )
-            
+
             # Store merged content
             if merged_content is not None:
-                blob = Blob(content=merged_content.encode('utf-8'))
+                blob = Blob(content=merged_content.encode("utf-8"))
                 merged_hash = blob.store(self.object_store)
                 merged[path] = merged_hash
-            
+
             # Record conflict if any
             if had_conflict:
-                conflicts.append(Conflict(
-                    path=path,
-                    base_content=base_content,
-                    ours_content=ours_content,
-                    theirs_content=theirs_content,
-                    message=f"{strategy.value} merge conflict in {path}"
-                ))
-        
+                conflicts.append(
+                    Conflict(
+                        path=path,
+                        base_content=base_content,
+                        ours_content=ours_content,
+                        theirs_content=theirs_content,
+                        message=f"{strategy.value} merge conflict in {path}",
+                    )
+                )
+
         return merged, conflicts
-    
-    def merge(self, source_branch: str, target_branch: Optional[str] = None,
-              message: Optional[str] = None) -> MergeResult:
+
+    def merge(
+        self, source_branch: str, target_branch: Optional[str] = None, message: Optional[str] = None
+    ) -> MergeResult:
         """
         Merge source branch into target branch (or current branch).
-        
+
         Args:
             source_branch: Branch to merge from
             target_branch: Branch to merge into (None for current)
             message: Merge commit message
-            
+
         Returns:
             MergeResult with success status and conflicts
         """
@@ -356,9 +431,9 @@ class MergeEngine:
                 success=False,
                 commit_hash=None,
                 conflicts=[],
-                message=f"Source branch not found: {source_branch}"
+                message=f"Source branch not found: {source_branch}",
             )
-        
+
         if target_branch:
             target_commit_hash = self.repo.resolve_ref(target_branch)
             if not target_commit_hash:
@@ -366,109 +441,111 @@ class MergeEngine:
                     success=False,
                     commit_hash=None,
                     conflicts=[],
-                    message=f"Target branch not found: {target_branch}"
+                    message=f"Target branch not found: {target_branch}",
                 )
         else:
             head = self.repo.refs.get_head()
-            if head['type'] == 'branch':
-                target_commit_hash = self.repo.refs.get_branch_commit(head['value'])
+            if head["type"] == "branch":
+                target_commit_hash = self.repo.refs.get_branch_commit(head["value"])
             else:
-                target_commit_hash = head['value']
-        
+                target_commit_hash = head["value"]
+
         # Find common ancestor
         ancestor_hash = self.find_common_ancestor(source_commit_hash, target_commit_hash)
-        
+
         if ancestor_hash == source_commit_hash:
             # Already up to date
             return MergeResult(
                 success=True,
                 commit_hash=target_commit_hash,
                 conflicts=[],
-                message="Already up to date"
+                message="Already up to date",
             )
-        
+
         if ancestor_hash == target_commit_hash:
             # Fast-forward
             if not target_branch:
                 target_branch = self.repo.refs.get_current_branch()
-            
+
             self.repo.refs.set_branch_commit(target_branch, source_commit_hash)
-            
+
             return MergeResult(
                 success=True,
                 commit_hash=source_commit_hash,
                 conflicts=[],
-                message=f"Fast-forward to {source_branch}"
+                message=f"Fast-forward to {source_branch}",
             )
-        
+
         # Three-way merge
         # Get trees
         ancestor_commit = Commit.load(self.object_store, ancestor_hash)
         ours_commit = Commit.load(self.object_store, target_commit_hash)
         theirs_commit = Commit.load(self.object_store, source_commit_hash)
-        
+
         base_files = self.get_tree_files(ancestor_commit.tree)
         ours_files = self.get_tree_files(ours_commit.tree)
         theirs_files = self.get_tree_files(theirs_commit.tree)
-        
+
         # Merge files
         merged_files, conflicts = self.merge_files(base_files, ours_files, theirs_files)
-        
+
         if conflicts:
             # Stage merged files for manual resolution
             for path, hash_id in merged_files.items():
                 content = Blob.load(self.object_store, hash_id).content
                 self.repo.staging.add(path, hash_id, content)
-            
+
             return MergeResult(
                 success=False,
                 commit_hash=None,
                 conflicts=conflicts,
-                message=f"Merge conflict in {len(conflicts)} file(s). Resolve conflicts and commit."
+                message=f"Merge conflict in {len(conflicts)} file(s). Resolve conflicts and commit.",
             )
-        
+
         # Create merge commit
         # Build tree from merged files
         entries = []
         for path, hash_id in merged_files.items():
             path_obj = Path(path)
-            entries.append(TreeEntry(
-                mode='100644',
-                obj_type='blob',
-                hash=hash_id,
-                name=path_obj.name,
-                path=str(path_obj.parent) if str(path_obj.parent) != '.' else ''
-            ))
-        
+            entries.append(
+                TreeEntry(
+                    mode="100644",
+                    obj_type="blob",
+                    hash=hash_id,
+                    name=path_obj.name,
+                    path=str(path_obj.parent) if str(path_obj.parent) != "." else "",
+                )
+            )
+
         tree = Tree(entries=entries)
         tree_hash = tree.store(self.object_store)
 
         merge_message = message or f"Merge branch '{source_branch}'"
-        
+
         merge_commit = Commit(
             tree=tree_hash,
             parents=[target_commit_hash, source_commit_hash],
             author=self.repo.get_author(),
-            timestamp=datetime.utcnow().isoformat() + 'Z',
+            timestamp=datetime.utcnow().isoformat() + "Z",
             message=merge_message,
-            metadata={'merge': True, 'source_branch': source_branch}
+            metadata={"merge": True, "source_branch": source_branch},
         )
-        
+
         merge_hash = merge_commit.store(self.object_store)
-        
+
         # Update target branch
         if not target_branch:
             target_branch = self.repo.refs.get_current_branch()
-        
+
         if target_branch:
             self.repo.refs.set_branch_commit(target_branch, merge_hash)
         else:
             # Detached HEAD
             self.repo.refs.set_head_detached(merge_hash)
-        
+
         return MergeResult(
             success=True,
             commit_hash=merge_hash,
             conflicts=[],
-            message=f"Successfully merged {source_branch}"
+            message=f"Successfully merged {source_branch}",
         )