support local module imports in datacustomcode run

Author: diksha-sf

pyproject.toml

@@ -101,7 +101,6 @@ click = "^8.1.8"
 loguru = "^0.7.3"
 numpy = "*"
 pandas = "*"
-pipreqs = "*"
 pydantic = "2.13.1"
 pyspark = "3.5.1"
 python = ">=3.10,<3.12"
@@ -115,6 +114,7 @@ build = "*"
 coverage = ">=7.0.0,<8.0.0"
 ipykernel = "^6.29.5"
 mypy = "*"
+pipreqs = "*"
 poetry-dynamic-versioning = {extras = ["plugin"], version = "^1.8.2"}
 pre-commit = "*"
 pytest = "*"

src/datacustomcode/run.py

@@ -201,8 +201,20 @@ def run_function_with_test(entrypoint: str, test_file: str) -> None:
 
 
 def add_py_folder(entrypoint: str):
-    default_py_folder = "py-files"  # Hardcoded folder name
+    """Add py-files subfolder and entrypoint directory to sys.path.
+
+    This ensures:
+    1. py-files/ is available for additional dependencies
+    2. The entrypoint directory is available for local module imports
+    """
+    default_py_folder = "py-files"
     cwd = Path.cwd().joinpath(entrypoint)
-    py_folder = cwd.parent.joinpath(default_py_folder)
+    entrypoint_dir = cwd.parent
+    py_folder = entrypoint_dir.joinpath(default_py_folder)
+
+    # Add py-files folder if it exists
+    if py_folder.exists():
+        sys.path.insert(0, str(py_folder))
 
-    sys.path.append(str(py_folder))
+    # Add entrypoint directory to allow local module imports
+    sys.path.insert(0, str(entrypoint_dir))

src/datacustomcode/templates/function/payload/entrypoint.py

@@ -1,173 +1,67 @@
 import logging
-from typing import List
-from uuid import uuid4
 
-from datacustomcode.einstein_predictions.types import (
-    PredictionColumBuilder,
-    PredictionRequestBuilder,
-    PredictionType,
-)
+from utility import extract_citations, split_text_into_chunks
+
 from datacustomcode.function import Runtime
-from datacustomcode.llm_gateway.types.generate_text_request_builder import (
-    GenerateTextRequestBuilder,
+from datacustomcode.function.feature_types.chunking import (
+    ChunkType,
+    SearchIndexChunkingV1Output,
+    SearchIndexChunkingV1Request,
+    SearchIndexChunkingV1Response,
 )
 
 logger = logging.getLogger(__name__)
+logging.basicConfig(level=logging.INFO)
 
+# Default max chunk size (can be overridden if contract adds max_characters field)
+DEFAULT_MAX_CHUNK_SIZE = 50
 
-def chunk_text(text: str, chunk_size: int = 1000) -> List[str]:
-    """
-    Split text into chunks of approximately chunk_size characters.
-    Tries to split at sentence boundaries when possible.
+
+def function(
+    request: SearchIndexChunkingV1Request, runtime: Runtime
+) -> SearchIndexChunkingV1Response:
+    """Chunk documents into smaller pieces for search indexing.
+
+    Args:
+        request: SearchIndexChunkingV1Request with input documents
+        runtime: Runtime context (unused but required by contract)
+
+    Returns:
+        SearchIndexChunkingV1Response with chunked output
     """
-    if not text:
-        return []
+    print(f"Received {len(request.input)} documents to chunk")
 
     chunks = []
-    current_chunk = ""
-
-    # Split text into sentences (simple split by period)
-    sentences = text.split(". ")
-
-    for sentence in sentences:
-        if len(current_chunk) + len(sentence) <= chunk_size:
-            current_chunk += sentence + ". "
-        else:
-            if current_chunk:
-                chunks.append(current_chunk.strip())
-            current_chunk = sentence + ". "
-
-    if current_chunk:
-        chunks.append(current_chunk.strip())
-
-    return chunks
-
-
-def make_einstein_prediction(runtime: Runtime) -> None:
-    column = (
-        PredictionColumBuilder()
-        .set_column_name("col1")
-        .set_string_values(["str1", "str2"])
-        .build()
-    )
-    prediction_request = (
-        PredictionRequestBuilder()
-        .set_prediction_type(PredictionType.REGRESSION)
-        .set_model_api_name("regressionModel")
-        .set_prediction_columns([column])
-        .build()
-    )
-
-    prediction_response = runtime.einstein_predictions.predict(prediction_request)
-    logger.info(
-        f"Einstein prediction results - success: [{prediction_response.is_success}] "
-        f"response data: {prediction_response.data}"
-    )
-
-
-def generate_text(runtime: Runtime, prompt: str, model: str = "sfdc_ai__DefaultGPT52"):
-    builder = GenerateTextRequestBuilder()
-    llm_request = builder.set_prompt(prompt).set_model(model).build()
-    llm_response = runtime.llm_gateway.generate_text(llm_request)
-    logger.info(
-        f"LLM Gateway generate text results - success: [{llm_response.is_success}] "
-        f"response data: {llm_response.data}"
-    )
-
-
-def function(request: dict, runtime: Runtime) -> dict:
-    logger.info("Inside Function")
-    logger.info(request)
-
-    items = request["input"]
-    output_chunks = []
-    current_seq_no = 1  # Start sequence number from 1
+    seq_no = 1
 
-    """
-    You can use your AI models configured in Salesforce to generate texts
-    or predict an outcome. See README.md for how to test locally before
-    deploying to Data Cloud.
+    # Use default max chunk size
+    max_chunk_size = DEFAULT_MAX_CHUNK_SIZE
 
-    Example:
+    # Process each document
+    for doc_idx, doc in enumerate(request.input):
+        text = doc.text
+        metadata = doc.metadata
 
-        >>> generated_text = generate_text(runtime, "Generate a greeting message")
-        ... prediction = make_einstein_prediction(runtime)
+        print(f"Processing document {doc_idx + 1}: {len(text)} characters")
 
-    """
+        # Split the text using our simple chunking algorithm
+        text_chunks = split_text_into_chunks(text, max_chunk_size, overlap=20)
+
+        # Create chunk outputs
+        for chunk_text in text_chunks:
+            citations = extract_citations(metadata)
+
+            chunk_output = SearchIndexChunkingV1Output(
+                chunk_type=ChunkType.TEXT,
+                text=chunk_text.strip(),
+                seq_no=seq_no,
+                citations=citations,
+            )
+            chunks.append(chunk_output)
+
+            print(f"Chunk {seq_no}: {len(chunk_text)} chars")
+            seq_no += 1
+
+    print(f"Generated {len(chunks)} chunks total")
 
-    for item in items:
-        # Item is DocElement as dict
-        logger.info(f"Processing item: {item}")
-
-        text = item.get("text", "")
-        metadata = item.get("metadata", {})
-
-        # Create chunks from the text
-        text_chunks = chunk_text(text, chunk_size=100)  # Using a larger chunk size
-
-        # Create chunk dictionaries for each text chunk
-        for chunk_content in text_chunks:
-            chunk_dict = {
-                "text": chunk_content,
-                "metadata": metadata,
-                "seq_no": current_seq_no,
-                "chunk_type": "text",
-                "chunk_id": str(uuid4()),
-                "tag_metadata": {},
-                "citations": {},
-                "source_record": item,
-            }
-            output_chunks.append(chunk_dict)
-            current_seq_no += 1  # Increment sequence number for next chunk
-
-    logger.info("Completed chunking")
-    response = {
-        "output": output_chunks,
-        "status": {"status_type": "success", "status_message": "Chunking completed"},
-    }
-    logger.info(response)
-    return response
-
-
-# Test the function
-if __name__ == "__main__":
-    # Configure logging
-    logging.basicConfig(level=logging.INFO)
-
-    # Create test data with two DocElements
-    test_request = {
-        "input": [
-            {
-                "text": (
-                    """This is the first sentence of the first document, which is
-                    intentionally made longer to test chunking. """
-                    """Here is the second sentence of the first document, which is also
-                     quite long and should ensure that the chunking function splits
-                     this text into two chunks when the chunk size is set to 100."""
-                ),
-                "metadata": {"source": "test1", "type": "document"},
-            },
-            {
-                "text": (
-                    """This is the first sentence of the second document, and it is
-                    also extended to be longer than usual for testing purposes. """
-                    """The second sentence of the second document is similarly lengthy,
-                     so that the chunking function will again create two chunks for
-                     this document."""
-                ),
-                "metadata": {"source": "test2", "type": "document"},
-            },
-        ]
-    }
-
-    # Run the function
-    result = function(test_request, Runtime())
-
-    # Print the results in a more readable format
-    print("\nChunking Results:")
-    print("----------------")
-    for chunk in result["output"]:
-        print(f"\nChunk #{chunk['seq_no']}:")
-        print(f"Text: {chunk['text'][:100]}...")  # Print first 100 chars of each chunk
-        print(f"Source: {chunk['metadata']['source']}")
-        print(f"Chunk ID: {chunk['chunk_id']}")
+    return SearchIndexChunkingV1Response(output=chunks)

src/datacustomcode/templates/function/payload/utility.py

@@ -0,0 +1,104 @@
+"""Utility functions for text chunking operations."""
+
+import logging
+from typing import (
+    Dict,
+    List,
+    Optional,
+)
+
+from datacustomcode.function.feature_types.chunking import SearchIndexChunkingV1Metadata
+
+logger = logging.getLogger(__name__)
+
+
+def split_text_into_chunks(text: str, max_size: int, overlap: int = 20) -> List[str]:
+    """Split text into chunks with overlap, trying to break at natural boundaries.
+
+    Tries to break at natural boundaries in order of preference:
+    1. Paragraph boundaries (\\n\\n)
+    2. Line boundaries (\\n)
+    3. Sentence boundaries (. ! ?)
+    4. Word boundaries (space)
+    5. Hard cut if no good boundary found
+
+    Args:
+        text: Text to split
+        max_size: Maximum characters per chunk
+        overlap: Number of characters to overlap between chunks
+
+    Returns:
+        List of text chunks
+    """
+    if len(text) <= max_size:
+        return [text]
+
+    chunks = []
+    start = 0
+
+    while start < len(text):
+        # Determine end position for this chunk
+        end = start + max_size
+
+        if end >= len(text):
+            # Last chunk
+            chunks.append(text[start:])
+            break
+
+        # Try to find a good breaking point (in order of preference)
+        chunk_text = text[start:end]
+        break_point = None
+
+        # Try to break at paragraph boundary (\n\n)
+        last_paragraph = chunk_text.rfind("\n\n")
+        if last_paragraph > max_size * 0.5:  # Only if it's past halfway
+            break_point = start + last_paragraph + 2  # +2 to skip the \n\n
+
+        # Try to break at line boundary (\n)
+        if break_point is None:
+            last_newline = chunk_text.rfind("\n")
+            if last_newline > max_size * 0.5:
+                break_point = start + last_newline + 1
+
+        # Try to break at sentence boundary (. ! ?)
+        if break_point is None:
+            for punct in [". ", "! ", "? "]:
+                last_sentence = chunk_text.rfind(punct)
+                if last_sentence > max_size * 0.5:
+                    break_point = start + last_sentence + len(punct)
+                    break
+
+        # Try to break at word boundary (space)
+        if break_point is None:
+            last_space = chunk_text.rfind(" ")
+            if last_space > max_size * 0.5:
+                break_point = start + last_space + 1
+
+        # If no good breaking point, just hard cut
+        if break_point is None:
+            break_point = end
+
+        chunks.append(text[start:break_point].strip())
+
+        # Move start position with overlap
+        start = max(break_point - overlap, start + 1)
+
+    return chunks
+
+
+def extract_citations(
+    metadata: Optional[SearchIndexChunkingV1Metadata],
+) -> Dict[str, str]:
+    """Extract citations from document metadata.
+
+    Args:
+        metadata: Document metadata containing source DMO fields
+
+    Returns:
+        Dictionary of citation key-value pairs
+    """
+    citations = {}
+    if metadata and metadata.source_dmo_fields:
+        for key, value in metadata.source_dmo_fields.items():
+            citations[key] = str(value)
+    return citations