#!/bin/bash

# Shared library for git AI commit message generation scripts

# Default model selection

# Returns the name of the preferred model to use

select_default_model() {

local models

models=$(llm models)

# Check if we got any models

if [ -z "$models" ]; then

echo "Error: No LLM models found. Please install at least one model." >&2

exit 1

fi

# Define preferred models in order of preference

local preferred_models=(

"ai-commit-message"

"claude-3.5-sonnet"

"gemini-2.0-flash-latest"

"o3-mini"

"deepseek-coder"

"gpt-4o-mini"

)

# Try each preferred model in order

for model in "${preferred_models[@]}"; do

if echo "$models" | grep -q "$model"; then

echo "$model"

return 0

fi

done

# If no preferred model is found, use the first available model

echo "$models" | head -n 1

}

# Resolve model alias to actual model name for display

# Arguments:

# $1 - model name (may be an alias)

# Returns: "alias: actual-model" or just the model name

resolve_model_display() {

local model="$1"

local model_display="$model"

if [[ "$model" == "ai-commit-message" ]]; then

local resolved_model

resolved_model=$(llm aliases list 2>/dev/null | grep "^ai-commit-message" | sed 's/^ai-commit-message[[:space:]]*:[[:space:]]*//')

if [[ -n "$resolved_model" ]]; then

model_display="ai-commit-message: $resolved_model"

fi

fi

echo "$model_display"

}

# Load additional instructions from global and project files

# Arguments:

# $1 - repository root path

# Returns: Combined instructions from global and project files

load_additional_instructions() {

local repo_root="$1"

local instructions=""

# Load global instructions

local global_file="$HOME/.config/ai-commit-instructions"

if [ -f "$global_file" ]; then

instructions=$(cat "$global_file")

fi

# Load project-local instructions

local project_file="$repo_root/.ai-commit-instructions"

if [ -f "$project_file" ]; then

if [ -n "$instructions" ]; then

instructions="$instructions

"

fi

instructions="$instructions$(cat "$project_file")"

fi

echo "$instructions"

}

# Sanitize UTF-8 input by replacing invalid sequences

# Reads from stdin, writes sanitized output to stdout

sanitize_utf8() {

# Try iconv first (faster), fall back to Python if not available

if command -v iconv >/dev/null 2>&1; then

# -c skips invalid sequences

iconv -f UTF-8 -t UTF-8 -c

else

# Python fallback: replace invalid UTF-8 with replacement character

python3 -c "

import sys

data = sys.stdin.buffer.read()

sys.stdout.write(data.decode('utf-8', errors='replace'))

"

fi

}

# Generate a commit message from diff using AI

# Arguments:

# $1 - model name

# $2 - custom prompt (optional, can be empty)

# $3 - current description (optional, can be empty)

# Reads diff from stdin

# Returns: Generated commit message

generate_commit_message() {

local model="$1"

local custom_prompt="$2"

local current_desc="$3"

# Read and sanitize diff from stdin

local diff_content

diff_content=$(sanitize_utf8)

# Check if diff is empty

if [ -z "$diff_content" ]; then

echo "Error: No diff content provided" >&2

return 1

fi

# Build the prompt based on whether custom prompt is provided

local prompt_text

if [[ -n "$custom_prompt" ]]; then

prompt_text="Analyze these changes and create a commit message according to the following instructions:

Changes:

\`\`\`

$diff_content

\`\`\`"

if [[ -n "$current_desc" ]]; then

prompt_text="$prompt_text

Current description:

\`\`\`

$current_desc

\`\`\`"

fi

prompt_text="$prompt_text

Instructions:

\`\`\`

$custom_prompt

\`\`\`

Format the response as a conventional commit message with a brief title line followed by a more detailed description if needed.

Do not include a summary paragraph after any list of changes.

FORMATTING RULES:

- Commit messages are viewed as plain text, not rendered markdown

- Use backticks for \`code\`, \`filenames\`, and \`identifiers\`

- Do NOT use **bold** or *italic* markdown

- Write bullet lists as plain text with simple dashes (-)

- Keep formatting minimal and readable as plain text

GOOD EXAMPLE:

feat: Add gradient compression pipeline

- Implement bucket-based quantization codec

- Add compression ratio calculation in \`metrics.py\`

- Support 8-bit and 16-bit quantization modes

- Update documentation with usage examples

AVOID (too much markdown):

feat: Add gradient compression pipeline

- **Implement** bucket-based quantization codec

- Add compression ratio calculation in **metrics.py**

- Support **8-bit** and **16-bit** quantization modes

- Update **documentation** with usage examples

Don't include any other text in the response, just the commit message."

else

prompt_text="Analyze these changes and create a conventional commit message:

\`\`\`

$diff_content

\`\`\`"

if [[ -n "$current_desc" ]]; then

prompt_text="$prompt_text

Current description (if any):

\`\`\`

$current_desc

\`\`\`"

fi

prompt_text="$prompt_text

Format the response as a conventional commit message with a brief title line followed by a more detailed description if needed.

Do not include a summary paragraph after any list of changes.

Follow the conventional commit format (e.g., feat:, fix:, docs:, chore:, refactor:, test:, style:).

FORMATTING RULES:

- Commit messages are viewed as plain text, not rendered markdown

- Use backticks for \`code\`, \`filenames\`, and \`identifiers\`

- Do NOT use **bold** or *italic* markdown

- Write bullet lists as plain text with simple dashes (-)

- Keep formatting minimal and readable as plain text

GOOD EXAMPLE:

feat: Add gradient compression pipeline

- Implement bucket-based quantization codec

- Add compression ratio calculation in \`metrics.py\`

- Support 8-bit and 16-bit quantization modes

- Update documentation with usage examples

AVOID (too much markdown):

feat: Add gradient compression pipeline

- **Implement** bucket-based quantization codec

- Add compression ratio calculation in **metrics.py**

- Support **8-bit** and **16-bit** quantization modes

- Update **documentation** with usage examples

Don't include any other text in the response, just the commit message."

fi

# Generate commit message using llm

# Pass prompt as argument, which may still be large but less likely to hit ARG_MAX

# than including the diff in the argument

local commit_msg

local llm_exit_code

commit_msg=$(echo "$prompt_text" | llm --model "$model")

llm_exit_code=$?

if [[ $llm_exit_code -ne 0 ]]; then

echo "$commit_msg" >&2

return 1

fi

# Strip markdown code fences if present

if [[ "$commit_msg" =~ ^\`\`\`.* ]] && [[ "$commit_msg" =~ \`\`\`$ ]]; then

commit_msg=$(echo "$commit_msg" | sed -e '1s/^```.*//' -e '$s/```$//' | sed '/^$/d')

fi

echo "$commit_msg"

}