metaxy 0.0.1.dev3__py3-none-any.whl

metaxy/graph/diff/differ.py

@@ -0,0 +1,769 @@
+"""Graph diffing logic and snapshot resolution."""
+
+import json
+import warnings
+from collections.abc import Mapping
+from typing import Any
+
+from metaxy.graph.diff.diff_models import (
+    AddedNode,
+    FieldChange,
+    GraphDiff,
+    NodeChange,
+    RemovedNode,
+)
+from metaxy.metadata_store.base import MetadataStore
+from metaxy.metadata_store.system import FEATURE_VERSIONS_KEY
+from metaxy.models.feature import FeatureGraph
+from metaxy.models.types import FeatureKey, FieldKey
+
+
+class SnapshotResolver:
+    """Resolves snapshot version literals to actual snapshot hashes."""
+
+    def resolve_snapshot(
+        self, literal: str, store: MetadataStore | None, graph: FeatureGraph | None
+    ) -> str:
+        """Resolve a snapshot literal to its actual version hash.
+
+        Args:
+            literal: Snapshot identifier ("latest", "current", or version hash)
+            store: Metadata store to query for snapshots (required for "latest")
+            graph: Optional active graph for "current" resolution
+
+        Returns:
+            Resolved snapshot version hash
+
+        Raises:
+            ValueError: If literal is invalid or cannot be resolved
+        """
+        if literal == "latest":
+            if store is None:
+                raise ValueError(
+                    "Cannot resolve 'latest': no metadata store provided. "
+                    "Provide a store to query for snapshots."
+                )
+            return self._resolve_latest(store)
+        elif literal == "current":
+            return self._resolve_current(graph)
+        else:
+            # Treat as explicit snapshot version
+            return literal
+
+    def _resolve_latest(self, store: MetadataStore) -> str:
+        """Resolve 'latest' to most recent snapshot in store."""
+        from metaxy.metadata_store.system.storage import SystemTableStorage
+
+        with store:
+            storage = SystemTableStorage(store)
+            snapshots_df = storage.read_graph_snapshots()
+
+        if snapshots_df.height == 0:
+            raise ValueError(
+                "No snapshots found in store. Cannot resolve 'latest'. "
+                "Run 'metaxy graph push' to record a snapshot."
+            )
+
+        # read_graph_snapshots() returns sorted by recorded_at descending
+        latest_snapshot = snapshots_df["metaxy_snapshot_version"][0]
+        return latest_snapshot
+
+    def _resolve_current(self, graph: FeatureGraph | None) -> str:
+        """Resolve 'current' to active graph's snapshot version."""
+        if graph is None:
+            raise ValueError(
+                "Cannot resolve 'current': no active graph provided. "
+                "Ensure features are loaded before using 'current'."
+            )
+
+        if len(graph.features_by_key) == 0:
+            raise ValueError(
+                "Cannot resolve 'current': active graph is empty. "
+                "Ensure features are loaded before using 'current'."
+            )
+
+        return graph.snapshot_version
+
+
+class GraphDiffer:
+    """Compares two graph snapshots and produces a diff."""
+
+    def diff(
+        self,
+        snapshot1_data: Mapping[str, Mapping[str, Any]],
+        snapshot2_data: Mapping[str, Mapping[str, Any]],
+        from_snapshot_version: str = "unknown",
+        to_snapshot_version: str = "unknown",
+    ) -> GraphDiff:
+        """Compute diff between two snapshots.
+
+        Args:
+            snapshot1_data: First snapshot (feature_key -> {feature_version, feature_spec, fields})
+            snapshot2_data: Second snapshot (feature_key -> {feature_version, feature_spec, fields})
+            from_snapshot_version: Source snapshot version
+            to_snapshot_version: Target snapshot version
+
+        Returns:
+            GraphDiff with added, removed, and changed features
+        """
+        # Extract feature keys
+        keys1 = set(snapshot1_data.keys())
+        keys2 = set(snapshot2_data.keys())
+
+        # Identify added and removed features
+        added_keys = keys2 - keys1
+        removed_keys = keys1 - keys2
+        common_keys = keys1 & keys2
+
+        # Build added nodes
+        added_nodes = []
+        for key_str in sorted(added_keys):
+            feature_data = snapshot2_data[key_str]
+            feature_spec = feature_data.get("feature_spec", {})
+
+            # Extract fields
+            fields_list = []
+            for field_dict in feature_spec.get("fields", []):
+                field_key_list = field_dict.get("key", [])
+                field_key_str = (
+                    "/".join(field_key_list)
+                    if isinstance(field_key_list, list)
+                    else field_key_list
+                )
+                fields_list.append(
+                    {
+                        "key": field_key_str,
+                        "version": feature_data.get("fields", {}).get(
+                            field_key_str, ""
+                        ),
+                        "code_version": field_dict.get("code_version"),
+                    }
+                )
+
+            # Extract dependencies
+            deps = []
+            if feature_spec.get("deps"):
+                for dep in feature_spec["deps"]:
+                    # Check for 'feature' field (new format) or 'key' field (old format)
+                    dep_key = dep.get("feature") or dep.get("key", [])
+                    if isinstance(dep_key, list):
+                        deps.append(FeatureKey(dep_key))
+                    else:
+                        deps.append(FeatureKey(dep_key.split("/")))
+
+            added_nodes.append(
+                AddedNode(
+                    feature_key=FeatureKey(key_str.split("/")),
+                    version=feature_data["metaxy_feature_version"],
+                    code_version=feature_spec.get("code_version"),
+                    fields=fields_list,
+                    dependencies=deps,
+                )
+            )
+
+        # Build removed nodes
+        removed_nodes = []
+        for key_str in sorted(removed_keys):
+            feature_data = snapshot1_data[key_str]
+            feature_spec = feature_data.get("feature_spec", {})
+
+            # Extract fields
+            fields_list = []
+            for field_dict in feature_spec.get("fields", []):
+                field_key_list = field_dict.get("key", [])
+                field_key_str = (
+                    "/".join(field_key_list)
+                    if isinstance(field_key_list, list)
+                    else field_key_list
+                )
+                fields_list.append(
+                    {
+                        "key": field_key_str,
+                        "version": feature_data.get("fields", {}).get(
+                            field_key_str, ""
+                        ),
+                        "code_version": field_dict.get("code_version"),
+                    }
+                )
+
+            # Extract dependencies
+            deps = []
+            if feature_spec.get("deps"):
+                for dep in feature_spec["deps"]:
+                    # Check for 'feature' field (new format) or 'key' field (old format)
+                    dep_key = dep.get("feature") or dep.get("key", [])
+                    if isinstance(dep_key, list):
+                        deps.append(FeatureKey(dep_key))
+                    else:
+                        deps.append(FeatureKey(dep_key.split("/")))
+
+            removed_nodes.append(
+                RemovedNode(
+                    feature_key=FeatureKey(key_str.split("/")),
+                    version=feature_data["metaxy_feature_version"],
+                    code_version=feature_spec.get("code_version"),
+                    fields=fields_list,
+                    dependencies=deps,
+                )
+            )
+
+        # Identify changed features
+        changed_nodes = []
+        for key_str in sorted(common_keys):
+            feature1 = snapshot1_data[key_str]
+            feature2 = snapshot2_data[key_str]
+
+            version1 = feature1["metaxy_feature_version"]
+            version2 = feature2["metaxy_feature_version"]
+
+            spec1 = feature1.get("feature_spec", {})
+            spec2 = feature2.get("feature_spec", {})
+
+            fields1 = feature1.get("fields", {})
+            fields2 = feature2.get("fields", {})
+
+            # Get tracking versions for migration detection
+            # Use tracking version if available (new system), otherwise fall back to feature_version
+            tracking_version1 = feature1.get("metaxy_full_definition_version", version1)
+            tracking_version2 = feature2.get("metaxy_full_definition_version", version2)
+
+            # Check if feature tracking version changed (indicates migration needed)
+            if tracking_version1 != tracking_version2:
+                # Compute field changes
+                field_changes = self._compute_field_changes(fields1, fields2)
+
+                changed_nodes.append(
+                    NodeChange(
+                        feature_key=FeatureKey(key_str.split("/")),
+                        old_version=version1,
+                        new_version=version2,
+                        old_code_version=spec1.get("code_version"),
+                        new_code_version=spec2.get("code_version"),
+                        added_fields=[fc for fc in field_changes if fc.is_added],
+                        removed_fields=[fc for fc in field_changes if fc.is_removed],
+                        changed_fields=[fc for fc in field_changes if fc.is_changed],
+                    )
+                )
+
+        return GraphDiff(
+            from_snapshot_version=from_snapshot_version,
+            to_snapshot_version=to_snapshot_version,
+            added_nodes=added_nodes,
+            removed_nodes=removed_nodes,
+            changed_nodes=changed_nodes,
+        )
+
+    def _compute_field_changes(
+        self, fields1: dict[str, str], fields2: dict[str, str]
+    ) -> list[FieldChange]:
+        """Compute changes between two field version mappings.
+
+        Args:
+            fields1: Field key (string) -> field version (hash) from snapshot1
+            fields2: Field key (string) -> field version (hash) from snapshot2
+
+        Returns:
+            List of FieldChange objects
+        """
+        field_keys1 = set(fields1.keys())
+        field_keys2 = set(fields2.keys())
+
+        added_fields = field_keys2 - field_keys1
+        removed_fields = field_keys1 - field_keys2
+        common_fields = field_keys1 & field_keys2
+
+        changes = []
+
+        # Added fields
+        for field_key_str in sorted(added_fields):
+            changes.append(
+                FieldChange(
+                    field_key=FieldKey(field_key_str.split("/")),
+                    old_version=None,
+                    new_version=fields2[field_key_str],
+                    old_code_version=None,
+                    new_code_version=None,  # TODO: Extract from spec if available
+                )
+            )
+
+        # Removed fields
+        for field_key_str in sorted(removed_fields):
+            changes.append(
+                FieldChange(
+                    field_key=FieldKey(field_key_str.split("/")),
+                    old_version=fields1[field_key_str],
+                    new_version=None,
+                    old_code_version=None,  # TODO: Extract from spec if available
+                    new_code_version=None,
+                )
+            )
+
+        # Changed fields
+        for field_key_str in sorted(common_fields):
+            version1 = fields1[field_key_str]
+            version2 = fields2[field_key_str]
+
+            if version1 != version2:
+                changes.append(
+                    FieldChange(
+                        field_key=FieldKey(field_key_str.split("/")),
+                        old_version=version1,
+                        new_version=version2,
+                        old_code_version=None,  # TODO: Extract from spec if available
+                        new_code_version=None,  # TODO: Extract from spec if available
+                    )
+                )
+
+        return changes
+
+    def create_merged_graph_data(
+        self,
+        snapshot1_data: Mapping[str, Mapping[str, Any]],
+        snapshot2_data: Mapping[str, Mapping[str, Any]],
+        diff: GraphDiff,
+    ) -> dict[str, Any]:
+        """Create merged graph data structure with status annotations.
+
+        This combines features from both snapshots into a single unified view,
+        annotating each feature with its status (added/removed/changed/unchanged).
+
+        Args:
+            snapshot1_data: First snapshot data (feature_key -> {feature_version, fields})
+            snapshot2_data: Second snapshot data (feature_key -> {feature_version, fields})
+            diff: Computed diff between snapshots
+
+        Returns:
+            Dict with structure:
+            {
+                'nodes': {
+                    feature_key_str: {
+                        'status': 'added' | 'removed' | 'changed' | 'unchanged',
+                        'old_version': str | None,
+                        'new_version': str | None,
+                        'fields': {...},  # fields from relevant snapshot
+                        'field_changes': [...],  # FieldChange objects for changed nodes
+                        'dependencies': [feature_key_str, ...],  # deps from relevant snapshot
+                    }
+                },
+                'edges': [
+                    {'from': feature_key_str, 'to': feature_key_str}
+                ]
+            }
+        """
+        # Create status mapping for efficient lookup
+        added_keys = {node.feature_key.to_string() for node in diff.added_nodes}
+        removed_keys = {node.feature_key.to_string() for node in diff.removed_nodes}
+        changed_keys = {
+            node.feature_key.to_string(): node for node in diff.changed_nodes
+        }
+
+        # Get all feature keys from both snapshots
+        all_keys = set(snapshot1_data.keys()) | set(snapshot2_data.keys())
+
+        nodes = {}
+        edges = []
+
+        for feature_key_str in all_keys:
+            # Determine status
+            if feature_key_str in added_keys:
+                status = "added"
+                old_version = None
+                new_version = snapshot2_data[feature_key_str]["metaxy_feature_version"]
+                fields = snapshot2_data[feature_key_str].get("fields", {})
+                field_changes = []
+                # Dependencies from snapshot2
+                deps = self._extract_dependencies(
+                    snapshot2_data[feature_key_str].get("feature_spec", {})
+                )
+            elif feature_key_str in removed_keys:
+                status = "removed"
+                old_version = snapshot1_data[feature_key_str]["metaxy_feature_version"]
+                new_version = None
+                fields = snapshot1_data[feature_key_str].get("fields", {})
+                field_changes = []
+                # Dependencies from snapshot1
+                deps = self._extract_dependencies(
+                    snapshot1_data[feature_key_str].get("feature_spec", {})
+                )
+            elif feature_key_str in changed_keys:
+                status = "changed"
+                node_change = changed_keys[feature_key_str]
+                old_version = node_change.old_version
+                new_version = node_change.new_version
+                fields = snapshot2_data[feature_key_str].get("fields", {})
+                # Combine all field changes from the NodeChange
+                field_changes = (
+                    node_change.added_fields
+                    + node_change.removed_fields
+                    + node_change.changed_fields
+                )
+                # Dependencies from snapshot2 (current version)
+                deps = self._extract_dependencies(
+                    snapshot2_data[feature_key_str].get("feature_spec", {})
+                )
+            else:
+                # Unchanged
+                status = "unchanged"
+                old_version = snapshot1_data[feature_key_str]["metaxy_feature_version"]
+                new_version = snapshot2_data[feature_key_str]["metaxy_feature_version"]
+                fields = snapshot2_data[feature_key_str].get("fields", {})
+                field_changes = []
+                # Dependencies from snapshot2
+                deps = self._extract_dependencies(
+                    snapshot2_data[feature_key_str].get("feature_spec", {})
+                )
+
+            nodes[feature_key_str] = {
+                "status": status,
+                "old_version": old_version,
+                "new_version": new_version,
+                "fields": fields,
+                "field_changes": field_changes,
+                "dependencies": deps,
+            }
+
+            # Create edges for dependencies (arrow points from dependency to feature)
+            for dep_key in deps:
+                edges.append({"from": dep_key, "to": feature_key_str})
+
+        return {
+            "nodes": nodes,
+            "edges": edges,
+        }
+
+    def _extract_dependencies(self, feature_spec: dict[str, Any]) -> list[str]:
+        """Extract dependency feature keys from a feature spec.
+
+        Args:
+            feature_spec: Parsed feature spec dict
+
+        Returns:
+            List of dependency feature keys as strings
+        """
+        deps = feature_spec.get("deps", [])
+        if deps is None:
+            return []
+
+        dep_keys = []
+        for dep in deps:
+            # Check for 'feature' field (new format) or 'key' field (old format)
+            dep_key = dep.get("feature") or dep.get("key", [])
+            if isinstance(dep_key, list):
+                dep_keys.append("/".join(dep_key))
+            else:
+                dep_keys.append(dep_key)
+
+        return dep_keys
+
+    def filter_merged_graph(
+        self,
+        merged_data: dict[str, Any],
+        focus_feature: str | None = None,
+        up: int | None = None,
+        down: int | None = None,
+    ) -> dict[str, Any]:
+        """Filter merged graph to show only relevant features.
+
+        Args:
+            merged_data: Merged graph data with nodes and edges
+            focus_feature: Feature key to focus on (string format with / or __)
+            up: Number of upstream levels (None = all if focus_feature is set, 0 otherwise)
+            down: Number of downstream levels (None = all if focus_feature is set, 0 otherwise)
+
+        Returns:
+            Filtered merged graph data with same structure
+
+        Raises:
+            ValueError: If focus_feature is specified but not found in graph
+        """
+        if focus_feature is None:
+            # No filtering
+            return merged_data
+
+        # Parse feature key (support both / and __ formats)
+        if "/" in focus_feature:
+            focus_key = focus_feature
+        else:
+            focus_key = focus_feature.replace("__", "/")
+
+        # Check if focus feature exists
+        if focus_key not in merged_data["nodes"]:
+            raise ValueError(f"Feature '{focus_feature}' not found in graph")
+
+        # Build dependency graph for traversal
+        # Build forward edges (feature -> dependents) and backward edges (feature -> dependencies)
+        forward_edges: dict[str, list[str]] = {}  # feature -> list of dependents
+        backward_edges: dict[str, list[str]] = {}  # feature -> list of dependencies
+
+        for edge in merged_data["edges"]:
+            dep = edge["from"]  # dependency
+            feat = edge["to"]  # dependent feature
+
+            if feat not in backward_edges:
+                backward_edges[feat] = []
+            backward_edges[feat].append(dep)
+
+            if dep not in forward_edges:
+                forward_edges[dep] = []
+            forward_edges[dep].append(feat)
+
+        # Find features to include
+        features_to_include = {focus_key}
+
+        # Add upstream (dependencies)
+        # Default behavior: if focus_feature is set but up is not specified, include all upstream
+        if up is None:
+            # Include all upstream
+            upstream = self._get_upstream_features(
+                focus_key, backward_edges, max_levels=None
+            )
+            features_to_include.update(upstream)
+        elif up > 0:
+            # Include specified number of levels
+            upstream = self._get_upstream_features(
+                focus_key, backward_edges, max_levels=up
+            )
+            features_to_include.update(upstream)
+        # else: up == 0, don't include upstream
+
+        # Add downstream (dependents)
+        # Default behavior: if focus_feature is set but down is not specified, include all downstream
+        if down is None:
+            # Include all downstream
+            downstream = self._get_downstream_features(
+                focus_key, forward_edges, max_levels=None
+            )
+            features_to_include.update(downstream)
+        elif down > 0:
+            # Include specified number of levels
+            downstream = self._get_downstream_features(
+                focus_key, forward_edges, max_levels=down
+            )
+            features_to_include.update(downstream)
+        # else: down == 0, don't include downstream
+
+        # Filter nodes and edges
+        filtered_nodes = {
+            k: v for k, v in merged_data["nodes"].items() if k in features_to_include
+        }
+        filtered_edges = [
+            e
+            for e in merged_data["edges"]
+            if e["from"] in features_to_include and e["to"] in features_to_include
+        ]
+
+        return {
+            "nodes": filtered_nodes,
+            "edges": filtered_edges,
+        }
+
+    def _get_upstream_features(
+        self,
+        start_key: str,
+        backward_edges: dict[str, list[str]],
+        max_levels: int | None = None,
+        visited: set[str] | None = None,
+        level: int = 0,
+    ) -> set[str]:
+        """Get upstream features (dependencies) recursively."""
+        if visited is None:
+            visited = set()
+
+        if start_key in visited:
+            return set()
+
+        if max_levels is not None and level >= max_levels:
+            return set()
+
+        visited.add(start_key)
+        upstream: set[str] = set()
+
+        deps = backward_edges.get(start_key, [])
+        for dep in deps:
+            if dep not in visited:
+                upstream.add(dep)
+                # Recurse
+                upstream.update(
+                    self._get_upstream_features(
+                        dep, backward_edges, max_levels, visited, level + 1
+                    )
+                )
+
+        return upstream
+
+    def _get_downstream_features(
+        self,
+        start_key: str,
+        forward_edges: dict[str, list[str]],
+        max_levels: int | None = None,
+        visited: set[str] | None = None,
+        level: int = 0,
+    ) -> set[str]:
+        """Get downstream features (dependents) recursively."""
+        if visited is None:
+            visited = set()
+
+        if start_key in visited:
+            return set()
+
+        if max_levels is not None and level >= max_levels:
+            return set()
+
+        visited.add(start_key)
+        downstream: set[str] = set()
+
+        dependents = forward_edges.get(start_key, [])
+        for dependent in dependents:
+            if dependent not in visited:
+                downstream.add(dependent)
+                # Recurse
+                downstream.update(
+                    self._get_downstream_features(
+                        dependent, forward_edges, max_levels, visited, level + 1
+                    )
+                )
+
+        return downstream
+
+    def load_snapshot_data(
+        self, store: MetadataStore, snapshot_version: str, project: str | None = None
+    ) -> Mapping[str, Mapping[str, Any]]:
+        """Load snapshot data from store.
+
+        Args:
+            store: Metadata store to query
+            snapshot_version: Snapshot version to load
+            project: Optional project name to filter by (None means all projects)
+
+        Returns:
+            Dict mapping feature_key (string) -> {feature_version, feature_spec, fields}
+            where fields is dict mapping field_key (string) -> field_version (hash)
+
+        Raises:
+            ValueError: If snapshot not found in store
+        """
+        # Import at function level to avoid circular imports
+
+        # Auto-open store if not already open
+        if not store._is_open:
+            with store.open("read"):
+                return self.load_snapshot_data(store, snapshot_version, project)
+
+        # Query feature_versions table for this snapshot
+        try:
+            features_lazy = store.read_metadata_in_store(FEATURE_VERSIONS_KEY)
+            if features_lazy is None:
+                raise ValueError(
+                    f"No feature_versions table found in store. Cannot load snapshot {snapshot_version}."
+                )
+
+            # Filter by snapshot_version and project
+            import narwhals as nw
+
+            features_df = (
+                features_lazy.filter(
+                    nw.col("metaxy_snapshot_version") == snapshot_version
+                )
+                .collect()
+                .to_polars()
+            )
+
+            if features_df.height == 0:
+                raise ValueError(
+                    f"Snapshot {snapshot_version} not found in store. "
+                    "Run 'metaxy graph push' to record snapshots or check the version hash."
+                )
+
+        except Exception as e:
+            raise ValueError(f"Failed to load snapshot {snapshot_version}: {e}") from e
+
+        # Build snapshot data structure
+        snapshot_dict = {}
+        for row in features_df.iter_rows(named=True):
+            feature_key_str = row["feature_key"]
+            feature_version = row["metaxy_feature_version"]
+            feature_spec_json = row["feature_spec"]
+            feature_class_path = row.get("feature_class_path", "")
+
+            feature_spec_dict = json.loads(feature_spec_json)
+
+            snapshot_dict[feature_key_str] = {
+                "metaxy_feature_version": feature_version,
+                "feature_spec": feature_spec_dict,
+                "feature_class_path": feature_class_path,
+                "metaxy_full_definition_version": row.get(
+                    "metaxy_full_definition_version", feature_version
+                ),  # Fallback for backward compatibility
+            }
+
+        # Try to reconstruct FeatureGraph from snapshot to compute field versions
+        # This may fail if features have been removed/moved, so we handle that gracefully
+        graph: FeatureGraph | None = None
+        try:
+            graph = FeatureGraph.from_snapshot(snapshot_dict)
+            graph_available = True
+        except ImportError:
+            # Some features can't be imported (likely removed) - proceed without graph
+            # For diff purposes, we can still show feature-level changes
+            # We'll use feature_version as a fallback for all field versions
+            graph_available = False
+            warnings.warn(
+                "Using feature_version as field_version fallback for features that cannot be imported. "
+                "This may occur when features have been removed or moved.",
+                UserWarning,
+                stacklevel=2,
+            )
+
+        # Compute field versions using the reconstructed graph (if available)
+        from metaxy.models.plan import FQFieldKey
+
+        snapshot_data = {}
+        for feature_key_str in snapshot_dict.keys():
+            feature_version = snapshot_dict[feature_key_str]["metaxy_feature_version"]
+            feature_spec = snapshot_dict[feature_key_str]["feature_spec"]
+            feature_key_obj = FeatureKey(feature_key_str.split("/"))
+
+            # Compute field versions using graph (if available)
+            fields_data = {}
+            if (
+                graph_available
+                and graph is not None
+                and feature_key_obj in graph.features_by_key
+            ):
+                # Feature exists in reconstructed graph - compute precise field versions
+                for field_dict in feature_spec.get("fields", []):
+                    field_key_list = field_dict.get("key")
+                    if isinstance(field_key_list, list):
+                        field_key = FieldKey(field_key_list)
+                        field_key_str_normalized = "/".join(field_key_list)
+                    else:
+                        field_key = FieldKey([field_key_list])
+                        field_key_str_normalized = field_key_list
+
+                    # Compute field version using the graph
+                    fq_key = FQFieldKey(feature=feature_key_obj, field=field_key)
+                    field_version = graph.get_field_version(fq_key)
+                    fields_data[field_key_str_normalized] = field_version
+            else:
+                # Feature doesn't exist in graph (removed/moved) - use feature_version as fallback
+                # All fields get the same version (the feature version)
+                for field_dict in feature_spec.get("fields", []):
+                    field_key_list = field_dict.get("key")
+                    if isinstance(field_key_list, list):
+                        field_key_str_normalized = "/".join(field_key_list)
+                    else:
+                        field_key_str_normalized = field_key_list
+
+                    # Use feature_version directly as fallback
+                    fields_data[field_key_str_normalized] = feature_version
+
+            snapshot_data[feature_key_str] = {
+                "metaxy_feature_version": feature_version,
+                "fields": fields_data,
+                "feature_spec": feature_spec,
+                "metaxy_full_definition_version": snapshot_dict[feature_key_str].get(
+                    "metaxy_full_definition_version", feature_version
+                ),
+            }
+
+        return snapshot_data