fix: Type chunk size and overlap during index key generation (#45)

Fixed index key generation and typing chunk size / overlap before hashing

Author: chandrasekharan-zipstack

src/unstract/sdk/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.24.0"
+__version__ = "0.24.1"
 
 
 def get_sdk_version():

src/unstract/sdk/index.py

@@ -31,7 +31,9 @@ def __init__(self, tool: BaseTool):
         # TODO: Inherit from StreamMixin and avoid using BaseTool
         self.tool = tool
 
-    def get_text_from_index(self, embedding_type: str, vector_db: str, doc_id: str):
+    def get_text_from_index(
+        self, embedding_type: str, vector_db: str, doc_id: str
+    ) -> Optional[str]:
         embedd_helper = ToolEmbedding(tool=self.tool)
         embedding_li = embedd_helper.get_embedding(adapter_instance_id=embedding_type)
         embedding_dimension = embedd_helper.get_embedding_length(embedding_li)
@@ -326,8 +328,10 @@ def generate_file_id(
             "vector_db_config": ToolAdapter.get_adapter_config(self.tool, vector_db),
             "embedding_config": ToolAdapter.get_adapter_config(self.tool, embedding),
             "x2text_config": ToolAdapter.get_adapter_config(self.tool, x2text),
-            "chunk_size": chunk_size,
-            "chunk_overlap": chunk_overlap,
+            # Typed and hashed as strings since the final hash is persisted
+            # and this is required to be backward compatible
+            "chunk_size": str(chunk_size),
+            "chunk_overlap": str(chunk_overlap),
         }
         # JSON keys are sorted to ensure that the same key gets hashed even in
         # case where the fields are reordered.