generative commit

gpt_commit_msg.py

import sys
import os
from dotenv import load_dotenv, find_dotenv
import openai
import tiktoken

# Find and load the .env file from the project root
# load_dotenv('.env')

def num_tokens_from_string(string: str) -> int:
    """Returns the number of tokens in a text string."""
    encoding = tiktoken.encoding_for_model("gpt-3.5-turbo")
    num_tokens = len(encoding.encode(string))
    return num_tokens

def generate_summary(code_chunks):
    openai.api_key = os.getenv("OPENAI_API_KEY")

    summaries = []


    for chunk in code_chunks:
        system_intro = "You are a commit message generator used in prepare-commit-msg.\n \
                        1. Review the provided diff\n 2. Identify which category the diff falls into \
                        \n\t [fix][feat][chore][docs][style][refactor][perf][test]   \
                        \n3. respond with [category] <commit message summarizing +/- changes> \
                        Constraint: Respond with the unpunctuated commit message related to the provided code change.\n"
        prompt = f"{chunk}\n<<END_DIFF>>\n\n"
        


        # Count the tokens in the prompt
        tokens = num_tokens_from_string(prompt)

        # Ensure the number of tokens is within the model's limits (4096 tokens for gpt-3.5-turbo)
        if tokens > 4096:
            print(f"Warning: Skipping file '{code_chunks}' due to token limit exceeded ({tokens} tokens).", file=sys.stderr)
            continue

        completion = openai.ChatCompletion.create(
            model="gpt-3.5-turbo",
            temperature=0,
            messages=[
                {"role": "system", "content": system_intro},
                {"role": "user", "content": prompt}
            ]
        )
        summary = completion.choices[0].message.content
        summaries.append(summary)

    return summaries

if __name__ == "__main__":
    code_chunks = sys.argv[1:]
    summaries = generate_summary(code_chunks)
    print("\n".join(summaries))

prepare-commit-msg.sh

#!/bin/sh

COMMIT_MSG_FILE=$1
COMMIT_SOURCE=$2
SHA1=$3

contains_use_gpt() {
    while IFS= read -r line; do
        if [[ $line == *"<USE_GPT>"* ]]; then
            return 0
        fi
    done < "$1"
    return 1
}

if contains_use_gpt "$COMMIT_MSG_FILE"; then
    #    # Get the added code chunks
    ADDED_CODE=$(git diff --cached --patch)
    # echo "Added code chunks:"
    # echo "$ADDED_CODE"

    # Split the added code into chunks (separated by diff headers)
    IFS=$'\n' read -ra CHUNKS <<< "$(echo "$ADDED_CODE" | awk '/^diff/{if (p) {print s}; s=""; p=1} {s=s $0 RS} END{if (p) print s}')"
    # echo "Chunks array:"
    # printf '%s\n' "${CHUNKS[@]}"

    # Generate commit message summaries using the OpenAI completions endpoint
    SUMMARIES=$(python ".git/hooks/gpt_commit_msg.py" "${CHUNKS[@]}")

    # Remove the original commit message prompt and add the generated summaries
    /usr/bin/perl -i.bak -ne 'print unless(m/^. Please enter the commit message/..m/^#$/)' "$COMMIT_MSG_FILE"
    echo -n "" > "$COMMIT_MSG_FILE"
    echo -e "$SUMMARIES\n" >> "$COMMIT_MSG_FILE"

    # exit 1  # Uncomment this line to prevent commits while debugging
fi