Code-aware text splitting in LangChain

Introduction

Code-aware text splitting helps break long code or text into smaller parts without cutting code blocks or important sections. This keeps the code readable and easy to process.

When you want to process large code files in smaller chunks for analysis or AI models.

When you need to split code for display in a user interface without breaking syntax.

When preparing code snippets for documentation or tutorials to keep examples clear.

When feeding code into language models that have input size limits.

When extracting meaningful parts of code while preserving structure.

Syntax

LangChain

from langchain_experimental.text_splitter import PythonCodeTextSplitter

splitter = PythonCodeTextSplitter()
chunks = splitter.split_text(long_code_string)

The PythonCodeTextSplitter automatically detects code blocks and splits accordingly.

You can customize splitting behavior by passing parameters like chunk_size and chunk_overlap.

Examples

Splits a small Python code snippet into chunks without breaking lines inside functions.

LangChain

from langchain_experimental.text_splitter import PythonCodeTextSplitter

splitter = PythonCodeTextSplitter()
chunks = splitter.split_text('def hello():\n    print("Hi")\n\nhello()')

Splits code into chunks of max 50 characters with 10 characters overlapping between chunks to keep context.

LangChain

splitter = PythonCodeTextSplitter(chunk_size=50, chunk_overlap=10)
chunks = splitter.split_text(large_code_string)

Sample Program

This example splits a small Python script into chunks of about 40 characters with some overlap. It prints each chunk clearly separated.

LangChain

from langchain_experimental.text_splitter import PythonCodeTextSplitter

code = '''
def greet(name):
    print(f"Hello, {name}!")

for person in ['Alice', 'Bob']:
    greet(person)
'''

splitter = PythonCodeTextSplitter(chunk_size=40, chunk_overlap=5)
chunks = splitter.split_text(code)

for i, chunk in enumerate(chunks, 1):
    print(f"Chunk {i}:\n{chunk}\n---")

Important Notes

Code-aware splitting keeps code blocks intact, avoiding syntax errors when processing chunks.

Overlap helps maintain context between chunks, useful for AI models understanding code.

Always test splitting on your specific code to ensure it fits your use case.

Summary

Code-aware text splitting breaks code into smaller, meaningful parts.

It avoids cutting inside code blocks to keep code readable and valid.

Useful for processing, displaying, or analyzing code in pieces.

Practice

1. What is the main purpose of code-aware text splitting in langchain?

easy

A. To convert code into plain text

B. To merge multiple code files into one

C. To remove comments from code automatically

D. To split text into smaller parts without breaking code blocks

Solution

1. Step 1: Understand the goal of text splitting

   Text splitting breaks large text into smaller pieces for easier handling.

2. Step 2: Recognize the importance of code awareness

   Code-aware splitting avoids breaking inside code blocks to keep code valid and readable.

3. Final Answer:

   To split text into smaller parts without breaking code blocks -> Option D

4. Quick Check:

   Code-aware splitting = split without breaking code blocks [OK]

Hint: Remember: split text but keep code blocks intact [OK]

Common Mistakes:

• Thinking it merges code files
• Assuming it removes comments
• Believing it converts code to plain text

2. Which of the following is the correct way to create a code-aware text splitter in langchain?

easy

A. TextSplitter(chunk_size=1000, overlap=100)

B. CodeAwareTextSplitter(chunk_size=1000, chunk_overlap=100)

C. CodeSplitter(chunk=1000, overlap=100)

D. CodeAwareSplitter(size=1000, overlap=100)

Solution

1. Step 1: Identify the correct class name

   The langchain class for code-aware splitting is CodeAwareTextSplitter.

2. Step 2: Check parameter names

   It uses chunk_size and chunk_overlap as parameters.

3. Final Answer:

   CodeAwareTextSplitter(chunk_size=1000, chunk_overlap=100) -> Option B

4. Quick Check:

   Correct class and parameters = CodeAwareTextSplitter(chunk_size=1000, chunk_overlap=100) [OK]

Hint: Look for exact class and parameter names [OK]

Common Mistakes:

• Using wrong class names like CodeSplitter
• Using incorrect parameter names like overlap
• Confusing with generic TextSplitter

3. Given the following code snippet using langchain's CodeAwareTextSplitter:

splitter = CodeAwareTextSplitter(chunk_size=50, chunk_overlap=10)
text = '''def add(a, b):\n    return a + b\n\nprint(add(2, 3))'''
chunks = splitter.split_texts([text])
print(len(chunks))

What will be the output?

medium

A. 1

B. 2

C. 3

D. 4

Solution

1. Step 1: Understand chunk size and overlap

   The chunk size is 50 characters, overlap 10 characters, so text splits into chunks of about 50 chars with 10 chars repeated.

2. Step 2: Analyze the text length and splitting

   The text has 49 characters including the '\n' sequences. It will split into 1 chunk because 49 <= 50.

3. Final Answer:

   1 -> Option A

4. Quick Check:

   Text length <= chunk_size, so splits into 1 chunk [OK]

Hint: Compare text length to chunk size to estimate chunks [OK]

Common Mistakes:

• Assuming only 1 chunk regardless of size
• Ignoring chunk overlap effect
• Counting lines instead of characters

4. You wrote this code to split Python code using langchain's CodeAwareTextSplitter:

splitter = CodeAwareTextSplitter(chunk_size=40, chunk_overlap=5)
text = 'def foo():\n    print("Hello")\n    return 42'
chunks = splitter.split_text(text)
print(chunks)

But you get an error: AttributeError: 'CodeAwareTextSplitter' object has no attribute 'split_text'. What is the likely cause?

medium

A. CodeAwareTextSplitter does not support Python code

B. The chunk_size value is too small

C. The method name should be split_documents instead of split_text

D. The text variable is not a list

Solution

1. Step 1: Check method names in CodeAwareTextSplitter

   The correct method to split text is split_documents or split_texts, not split_text.

2. Step 2: Confirm method usage

   Using split_text causes AttributeError because it does not exist.

3. Final Answer:

   The method name should be split_documents instead of split_text -> Option C

4. Quick Check:

   Method name mismatch causes AttributeError [OK]

Hint: Check exact method names in docs or autocomplete [OK]

Common Mistakes:

• Using wrong method name
• Assuming chunk_size causes error
• Passing string instead of list without checking method

5. You want to split a large Python script into chunks for analysis, but you must ensure no code block is broken and each chunk overlaps by 20 characters. Which approach using langchain's CodeAwareTextSplitter is best?

hard

A. Set chunk_size to 500 and chunk_overlap to 20, then call split_documents on the script wrapped as a Document

B. Set chunk_size to 20 and chunk_overlap to 500, then call split_text on the raw string

C. Use TextSplitter with chunk_size=500 and chunk_overlap=20 on the script string

D. Split the script manually by lines and ignore chunk_overlap

Solution

1. Step 1: Choose correct splitter and parameters

   Use CodeAwareTextSplitter to avoid breaking code blocks, with chunk_size=500 and chunk_overlap=20 for desired chunk size and overlap.

2. Step 2: Use correct input and method

   Wrap the script as a Document and call split_documents to get properly split chunks.

3. Final Answer:

   Set chunk_size to 500 and chunk_overlap to 20, then call split_documents on the script wrapped as a Document -> Option A

4. Quick Check:

   Use CodeAwareTextSplitter + split_documents + proper chunk sizes [OK]

Hint: Use CodeAwareTextSplitter with Documents and split_documents method [OK]

Common Mistakes:

• Using wrong method like split_text
• Setting chunk_overlap larger than chunk_size
• Using generic TextSplitter instead of code-aware