Code-aware text splitting in LangChain - Step-by-Step Execution

Concept Flow - Code-aware text splitting

Input: Large Code Text

↓

Identify Code Boundaries

↓

Split Text at Logical Points

↓

Create Smaller Code Chunks

↓

Output: List of Code Chunks

The process starts with a large code text, then finds logical places to split it, and finally outputs smaller code chunks.

Execution Sample

LangChain

from langchain.text_splitter import CodeTextSplitter

text = '''def add(a, b):\n    return a + b\n\nprint(add(2, 3))'''
splitter = CodeTextSplitter()
chunks = splitter.split_text(text)

This code splits a Python code string into smaller chunks respecting code structure.

Execution Table

Step	Action	Input Text	Split Points Found	Chunks Created
1	Receive full code text	def add(a, b):\n return a + b\n\nprint(add(2, 3))	None yet	None yet
2	Analyze text for code boundaries	Same as input	After function definition and before print statement	None yet
3	Split text at identified boundaries	Same as input	Confirmed split points	['def add(a, b):\n return a + b', 'print(add(2, 3))']
4	Return list of code chunks	Same as input	Split points used	2 chunks created

💡 All code text processed and split into logical chunks

Variable Tracker

Variable	Start	After Step 2	After Step 3	Final
text	Full code string	Full code string	Full code string	Full code string
split_points	None	Positions after function and blank line	Positions confirmed	Positions used for splitting
chunks	None	None	List with 2 code chunks	List with 2 code chunks

Key Moments - 2 Insights

Why doesn't the splitter just split by every newline?

Because splitting by every newline would break code logic. The splitter finds logical boundaries like function ends to keep code chunks meaningful, as shown in execution_table step 2 and 3.

What if the code has comments or blank lines?

The splitter treats comments and blank lines as part of code blocks or boundaries depending on context, ensuring chunks remain valid code, as seen in the split points found in step 2.

Visual Quiz - 3 Questions

Look at the execution table, how many chunks are created after splitting?

A 1

B 2

C 3

D 4

💡 Hint

Check the 'Chunks Created' column in step 3 and 4 of the execution_table

At which step are the split points identified?

A Step 2

B Step 1

C Step 3

D Step 4

💡 Hint

Look at the 'Split Points Found' column in the execution_table

If the input text had no logical boundaries, what would happen to chunks?

A Multiple small chunks created

B No chunks created

C One chunk containing the whole text

D Error thrown

💡 Hint

Refer to how chunks are created only at logical split points in variable_tracker and execution_table

Concept Snapshot

Code-aware text splitting:
- Input large code text
- Detect logical code boundaries (functions, classes)
- Split text at these points
- Output list of smaller code chunks
- Keeps code chunks meaningful and valid

Full Transcript

Code-aware text splitting takes a large piece of code and breaks it into smaller parts without breaking the code logic. It looks for places like the end of functions or classes to split. This way, each chunk is a meaningful piece of code. The example shows splitting a Python function and a print statement into two chunks. The process involves reading the full text, finding split points, splitting, and returning the chunks. This helps when processing code in smaller parts, like for analysis or display.

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