haystack-velesdb 8.0.0__tar.gz → 8.2.0__tar.gz

{haystack_velesdb-8.0.0/src/haystack_velesdb.egg-info → haystack_velesdb-8.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: haystack-velesdb
-Version: 8.0.0
+Version: 8.2.0
 Summary: Haystack 2.x DocumentStore for VelesDB: The Local AI Memory Database.
 Author-email: VelesDB Team <contact@wiscale.fr>
 License: MIT
@@ -21,7 +21,7 @@ Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: haystack-ai>=2.0.0
 Requires-Dist: velesdb>=5.1.0
-Requires-Dist: velesdb-common>=8.0.0
+Requires-Dist: velesdb-common>=8.2.0
 Provides-Extra: dev
 Requires-Dist: pytest<9.0,>=7.0; extra == "dev"
 Dynamic: license-file
@@ -82,6 +82,9 @@ and semantic search example using `SentenceTransformersDocumentEmbedder`.
 ```python
 from haystack import Pipeline
 from haystack.components.converters import PyPDFToDocument
+# haystack-ai 2.x path; on haystack-ai 3.x the SentenceTransformers embedders
+# moved to the `sentence-transformers-haystack` package:
+#   from haystack_integrations.components.embedders.sentence_transformers import (...)
 from haystack.components.embedders import (
     SentenceTransformersDocumentEmbedder,
     SentenceTransformersTextEmbedder,
@@ -139,7 +142,7 @@ print(result["retriever"]["documents"])
 | `write_documents_streaming(documents, policy, sparse_vectors=None, batch_size=100)` | Same `DuplicatePolicy` semantics as `write_documents`, but sent through the streaming channel in `batch_size`-sized chunks — better throughput for large bulk loads. See "Note on streaming ingestion" below |
 | `flush()` | Flush pending changes to disk. See "Note on streaming ingestion" below |
 | `filter_documents(filters)` | Scroll documents matching a VelesDB filter dict, up to `scroll_limit` |
-| `embedding_retrieval(query_embedding, top_k, filters, scale_score, fusion=None, fusion_params=None)` | Vector similarity search. `fusion` selects a `velesdb.FusionStrategy` (`"average"`, `"maximum"`, `"rrf"`, `"weighted"`, `"relative_score"` / `"rsf"`) applied via `Collection.multi_query_search`, changing the ranking; `fusion_params` configures it (e.g. `{"dense_weight": 0.7, "sparse_weight": 0.3}` for `"rsf"`). `filters` cannot be combined with `fusion` |
+| `embedding_retrieval(query_embedding, top_k, filters, scale_score, fusion=None, fusion_params=None)` | Vector similarity search. `fusion` selects a `velesdb.FusionStrategy` (`"average"`, `"maximum"`, `"rrf"`, `"weighted_rrf"`, `"weighted"`, `"relative_score"` / `"rsf"`) applied via `Collection.multi_query_search`, changing the ranking; `fusion_params` configures it (e.g. `{"dense_weight": 0.7, "sparse_weight": 0.3}` for `"rsf"`, `{"weights": [...]}` for `"weighted_rrf"`). `filters` cannot be combined with `fusion` |
 | `count_documents()` | Total document count |
 | `delete_documents(document_ids)` | Delete by Haystack string IDs |
 | `train_pq(m=8, k=256, opq=False)` | Train Product Quantization on the collection |
@@ -150,9 +153,11 @@ print(result["retriever"]["documents"])
 | `to_dict()` / `from_dict()` | Haystack pipeline serialisation |
 
 **Note on `DuplicatePolicy`:** `NONE` and `OVERWRITE` use VelesDB upsert semantics
-and always overwrite on collision.  `FAIL` is fully enforced: a pre-scan is
-performed before writing and `DuplicateDocumentError` is raised if any document
-already exists (prefer `OVERWRITE` or `NONE` for bulk loads to skip the scan cost).
+and always overwrite on collision. `SKIP` drops incoming documents whose ID
+already exists and writes only the new subset. `FAIL` is fully enforced: a
+pre-scan is performed before writing and `DuplicateDocumentError` is raised if
+any document already exists (prefer `OVERWRITE` or `NONE` for bulk loads to
+skip the scan cost).
 
 **Note on document IDs and SHA-256:** Haystack string IDs are mapped to 63-bit
 integers using the first 8 bytes of SHA-256 (~9.2 × 10¹⁸ slots).  For a

{haystack_velesdb-8.0.0 → haystack_velesdb-8.2.0}/README.md

@@ -54,6 +54,9 @@ and semantic search example using `SentenceTransformersDocumentEmbedder`.
 ```python
 from haystack import Pipeline
 from haystack.components.converters import PyPDFToDocument
+# haystack-ai 2.x path; on haystack-ai 3.x the SentenceTransformers embedders
+# moved to the `sentence-transformers-haystack` package:
+#   from haystack_integrations.components.embedders.sentence_transformers import (...)
 from haystack.components.embedders import (
     SentenceTransformersDocumentEmbedder,
     SentenceTransformersTextEmbedder,
@@ -111,7 +114,7 @@ print(result["retriever"]["documents"])
 | `write_documents_streaming(documents, policy, sparse_vectors=None, batch_size=100)` | Same `DuplicatePolicy` semantics as `write_documents`, but sent through the streaming channel in `batch_size`-sized chunks — better throughput for large bulk loads. See "Note on streaming ingestion" below |
 | `flush()` | Flush pending changes to disk. See "Note on streaming ingestion" below |
 | `filter_documents(filters)` | Scroll documents matching a VelesDB filter dict, up to `scroll_limit` |
-| `embedding_retrieval(query_embedding, top_k, filters, scale_score, fusion=None, fusion_params=None)` | Vector similarity search. `fusion` selects a `velesdb.FusionStrategy` (`"average"`, `"maximum"`, `"rrf"`, `"weighted"`, `"relative_score"` / `"rsf"`) applied via `Collection.multi_query_search`, changing the ranking; `fusion_params` configures it (e.g. `{"dense_weight": 0.7, "sparse_weight": 0.3}` for `"rsf"`). `filters` cannot be combined with `fusion` |
+| `embedding_retrieval(query_embedding, top_k, filters, scale_score, fusion=None, fusion_params=None)` | Vector similarity search. `fusion` selects a `velesdb.FusionStrategy` (`"average"`, `"maximum"`, `"rrf"`, `"weighted_rrf"`, `"weighted"`, `"relative_score"` / `"rsf"`) applied via `Collection.multi_query_search`, changing the ranking; `fusion_params` configures it (e.g. `{"dense_weight": 0.7, "sparse_weight": 0.3}` for `"rsf"`, `{"weights": [...]}` for `"weighted_rrf"`). `filters` cannot be combined with `fusion` |
 | `count_documents()` | Total document count |
 | `delete_documents(document_ids)` | Delete by Haystack string IDs |
 | `train_pq(m=8, k=256, opq=False)` | Train Product Quantization on the collection |
@@ -122,9 +125,11 @@ print(result["retriever"]["documents"])
 | `to_dict()` / `from_dict()` | Haystack pipeline serialisation |
 
 **Note on `DuplicatePolicy`:** `NONE` and `OVERWRITE` use VelesDB upsert semantics
-and always overwrite on collision.  `FAIL` is fully enforced: a pre-scan is
-performed before writing and `DuplicateDocumentError` is raised if any document
-already exists (prefer `OVERWRITE` or `NONE` for bulk loads to skip the scan cost).
+and always overwrite on collision. `SKIP` drops incoming documents whose ID
+already exists and writes only the new subset. `FAIL` is fully enforced: a
+pre-scan is performed before writing and `DuplicateDocumentError` is raised if
+any document already exists (prefer `OVERWRITE` or `NONE` for bulk loads to
+skip the scan cost).
 
 **Note on document IDs and SHA-256:** Haystack string IDs are mapped to 63-bit
 integers using the first 8 bytes of SHA-256 (~9.2 × 10¹⁸ slots).  For a

{haystack_velesdb-8.0.0 → haystack_velesdb-8.2.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "haystack-velesdb"
-version = "8.0.0"
+version = "8.2.0"
 description = "Haystack 2.x DocumentStore for VelesDB: The Local AI Memory Database."
 readme = "README.md"
 license = {text = "MIT"}
@@ -32,7 +32,7 @@ requires-python = ">=3.10"
 dependencies = [
     "haystack-ai>=2.0.0",
     "velesdb>=5.1.0",
-    "velesdb-common>=8.0.0",
+    "velesdb-common>=8.2.0",
 ]
 
 [project.optional-dependencies]

{haystack_velesdb-8.0.0 → haystack_velesdb-8.2.0}/src/haystack_velesdb/__init__.py

@@ -4,4 +4,4 @@ from haystack_velesdb.document_store import VelesDBDocumentStore
 from haystack_velesdb.retriever import VelesDBEmbeddingRetriever
 
 __all__ = ["VelesDBDocumentStore", "VelesDBEmbeddingRetriever"]
-__version__ = "8.0.0"
+__version__ = "8.2.0"

{haystack_velesdb-8.0.0 → haystack_velesdb-8.2.0}/src/haystack_velesdb/document_store.py

@@ -23,6 +23,7 @@ from velesdb_common.scores import (
     to_similarity,
 )
 from velesdb_common.security import (
+    ALLOWED_CONDITION_TYPES,
     validate_collection_name,
     validate_metric,
     validate_named_sparse_vector,
@@ -60,6 +61,21 @@ _HAYSTACK_LOGIC_TO_VELES: Dict[str, str] = {
     "NOT": "not",
 }
 
+# The TARGET tags are not re-listed as free literals: every value in both
+# maps must be a member of ``ALLOWED_CONDITION_TYPES`` (single-sourced from
+# velesdb-core through ``velesdb_common``), which this import-time guard
+# enforces so the maps cannot drift from the engine's vocabulary — the same
+# guard ``llamaindex_velesdb.filter_ops`` already carries (#2095 item B).
+_UNKNOWN_CONDITION_TYPES = (
+    set(_HAYSTACK_COMPARISON_TO_VELES.values())
+    | set(_HAYSTACK_LOGIC_TO_VELES.values())
+) - ALLOWED_CONDITION_TYPES
+if _UNKNOWN_CONDITION_TYPES:
+    raise RuntimeError(
+        "haystack_velesdb emits condition types unknown to velesdb-core: "
+        f"{sorted(_UNKNOWN_CONDITION_TYPES)}"
+    )
+
 
 def _strip_meta_prefix(field: str) -> str:
     """Drop the leading ``meta.`` namespace from a Haystack field path.
@@ -747,7 +763,7 @@ class VelesDBDocumentStore(CollectionAdminMixin):
                 keeps (``average``, ``maximum`` or ``weighted`` over a distance
                 metric) is mapped to ``1 / (1 + d)`` either way.
             fusion: Optional fusion strategy name applied to the ranking —
-                one of ``"average"``, ``"maximum"``, ``"rrf"``,
+                one of ``"average"``, ``"maximum"``, ``"rrf"``, ``"weighted_rrf"``,
                 ``"weighted"``, ``"relative_score"`` / ``"rsf"``. When set,
                 the query is ranked through the chosen
                 :class:`velesdb.FusionStrategy`, which changes the result

{haystack_velesdb-8.0.0 → haystack_velesdb-8.2.0/src/haystack_velesdb.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: haystack-velesdb
-Version: 8.0.0
+Version: 8.2.0
 Summary: Haystack 2.x DocumentStore for VelesDB: The Local AI Memory Database.
 Author-email: VelesDB Team <contact@wiscale.fr>
 License: MIT
@@ -21,7 +21,7 @@ Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: haystack-ai>=2.0.0
 Requires-Dist: velesdb>=5.1.0
-Requires-Dist: velesdb-common>=8.0.0
+Requires-Dist: velesdb-common>=8.2.0
 Provides-Extra: dev
 Requires-Dist: pytest<9.0,>=7.0; extra == "dev"
 Dynamic: license-file
@@ -82,6 +82,9 @@ and semantic search example using `SentenceTransformersDocumentEmbedder`.
 ```python
 from haystack import Pipeline
 from haystack.components.converters import PyPDFToDocument
+# haystack-ai 2.x path; on haystack-ai 3.x the SentenceTransformers embedders
+# moved to the `sentence-transformers-haystack` package:
+#   from haystack_integrations.components.embedders.sentence_transformers import (...)
 from haystack.components.embedders import (
     SentenceTransformersDocumentEmbedder,
     SentenceTransformersTextEmbedder,
@@ -139,7 +142,7 @@ print(result["retriever"]["documents"])
 | `write_documents_streaming(documents, policy, sparse_vectors=None, batch_size=100)` | Same `DuplicatePolicy` semantics as `write_documents`, but sent through the streaming channel in `batch_size`-sized chunks — better throughput for large bulk loads. See "Note on streaming ingestion" below |
 | `flush()` | Flush pending changes to disk. See "Note on streaming ingestion" below |
 | `filter_documents(filters)` | Scroll documents matching a VelesDB filter dict, up to `scroll_limit` |
-| `embedding_retrieval(query_embedding, top_k, filters, scale_score, fusion=None, fusion_params=None)` | Vector similarity search. `fusion` selects a `velesdb.FusionStrategy` (`"average"`, `"maximum"`, `"rrf"`, `"weighted"`, `"relative_score"` / `"rsf"`) applied via `Collection.multi_query_search`, changing the ranking; `fusion_params` configures it (e.g. `{"dense_weight": 0.7, "sparse_weight": 0.3}` for `"rsf"`). `filters` cannot be combined with `fusion` |
+| `embedding_retrieval(query_embedding, top_k, filters, scale_score, fusion=None, fusion_params=None)` | Vector similarity search. `fusion` selects a `velesdb.FusionStrategy` (`"average"`, `"maximum"`, `"rrf"`, `"weighted_rrf"`, `"weighted"`, `"relative_score"` / `"rsf"`) applied via `Collection.multi_query_search`, changing the ranking; `fusion_params` configures it (e.g. `{"dense_weight": 0.7, "sparse_weight": 0.3}` for `"rsf"`, `{"weights": [...]}` for `"weighted_rrf"`). `filters` cannot be combined with `fusion` |
 | `count_documents()` | Total document count |
 | `delete_documents(document_ids)` | Delete by Haystack string IDs |
 | `train_pq(m=8, k=256, opq=False)` | Train Product Quantization on the collection |
@@ -150,9 +153,11 @@ print(result["retriever"]["documents"])
 | `to_dict()` / `from_dict()` | Haystack pipeline serialisation |
 
 **Note on `DuplicatePolicy`:** `NONE` and `OVERWRITE` use VelesDB upsert semantics
-and always overwrite on collision.  `FAIL` is fully enforced: a pre-scan is
-performed before writing and `DuplicateDocumentError` is raised if any document
-already exists (prefer `OVERWRITE` or `NONE` for bulk loads to skip the scan cost).
+and always overwrite on collision. `SKIP` drops incoming documents whose ID
+already exists and writes only the new subset. `FAIL` is fully enforced: a
+pre-scan is performed before writing and `DuplicateDocumentError` is raised if
+any document already exists (prefer `OVERWRITE` or `NONE` for bulk loads to
+skip the scan cost).
 
 **Note on document IDs and SHA-256:** Haystack string IDs are mapped to 63-bit
 integers using the first 8 bytes of SHA-256 (~9.2 × 10¹⁸ slots).  For a

{haystack_velesdb-8.0.0 → haystack_velesdb-8.2.0}/src/haystack_velesdb.egg-info/requires.txt

@@ -1,6 +1,6 @@
 haystack-ai>=2.0.0
 velesdb>=5.1.0
-velesdb-common>=8.0.0
+velesdb-common>=8.2.0
 
 [dev]
 pytest<9.0,>=7.0

{haystack_velesdb-8.0.0 → haystack_velesdb-8.2.0}/tests/test_document_store.py

@@ -339,6 +339,16 @@ def _load_module() -> types.ModuleType:
     vc_sec.validate_metric = _passthrough  # type: ignore[attr-defined]
     vc_sec.validate_named_sparse_vector = _passthrough  # type: ignore[attr-defined]
     vc_sec.SecurityError = ValueError  # type: ignore[attr-defined]
+    # Mirrors velesdb_common.security._FALLBACK_CONDITION_TYPES: the store's
+    # import-time drift guard reads this set.
+    vc_sec.ALLOWED_CONDITION_TYPES = frozenset(  # type: ignore[attr-defined]
+        {
+            "eq", "neq", "gt", "gte", "lt", "lte", "in", "contains",
+            "is_null", "is_not_null", "and", "or", "not", "like", "ilike",
+            "array_contains", "array_contains_any", "array_contains_all",
+            "geo_distance", "geo_bbox",
+        }
+    )
     sys.modules["velesdb_common.security"] = vc_sec
 
     vc_fusion = types.ModuleType("velesdb_common.fusion")