Git Fetcher (git-source / gsrc)

git-source (or gsrc for short) is a command-line tool to fetch archives, latest tags, or branches from Git repositories hosted on GitHub, GitLab, Bitbucket, or Googlesource. It automates the retrieval of the latest version of a repository, saving archives locally.

---

Features

• Fetch the latest release tag or default branch automatically.
• Download the repository as a .tar.gz archive.
• Download a direct archive URL (for unsupported or unknown hosts).
• List all branches of a repository.
• Supports GitHub, GitLab, Bitbucket, and Googlesource.
• Optional verbose and debug output.
• Cross-platform: works anywhere Bash, curl, and git are available.
• Automatically generate MXE package (.mk) files from Git repository URLs

---

Requirements

The following tools must be installed for git-source to work properly:

• bash – The scripts are written in Bash. Most Linux/macOS systems include this by default.
• git – Needed to fetch repository info, list branches, and for fetch_latest_* functions.
• jq – Required to read/write the JSON database safely. Must be installed separately.
• curl or wget – Used by download_archive to fetch files from URLs.
• sha256sum – Used to compute checksums. On macOS, this may be shasum -a 256.
• make – Needed if building packages locally or running MXE build commands.
• tar – Needed to extract source archives.

Optional Tools

• MXE toolchain Only required if you plan to generate .mk files and build packages.
• meson – Required to detect and configure Meson-based projects.
• ninja-build – Needed by Meson as a backend to configure and build projects.
• cmake – Required to detect and configure CMake-based projects.
• patch – Optional, used if the build scripts apply patches to sources.
• pkg-config – Useful for detecting installed libraries when generating .mk files.

Installation

You can run git-source directly from the repository without installing:

# From the project root
./git-source.sh <repo_url_or_owner/repo>

# Install to default prefix (/usr/local)
./install.sh

# Install to a custom prefix, e.g., local user directory
./install.sh --prefix $HOME/.local

This installation process will:

• Copy the wrapper git-source to $PREFIX/bin/git-source
• Create a symlink gsrc pointing to the same executable
• Install the man page to $PREFIX/share/man/man1/git-source.1
• Record all installed files in $PREFIX/git-source.install-manifest

Make sure your PATH includes the installation bin directory:

export PATH="$PREFIX/bin:$PATH"

And if you want man pages accessible:

export MANPATH="$PREFIX/share/man:$MANPATH"

Enable Bash Autocomplete

git-source supports Tab completion for all flags and options. To enable it:

1. Source the completion script in your current shell:

source scripts/git-source-completion.sh

MXE Integration (Optional)

The git-source command can generate MXE .mk files for use in your MXE project using the --generate-mxe-makefile or --generate-mxe-makefile=libnameoption. If you want the generated files to be automatically copied into your MXE project, you can set the following environment variables before running the generator:

export MXE_ROOT="/path/to/your/mxe"  # Path to your MXE root directory, automatically set if you run git-source from mxe root
export MXE_TARGET="x86_64-w64-mingw32.static"  # Target triplet (optional, defaults to x86_64-w64-mingw32.static)
export GITHUB_TOKEN="<your_token_here>"  # optional, avoids API rate limits"

Uninstallation

To remove installed files, use the uninstall.sh script with the same prefix you used for installation:

# Default prefix
./uninstall.sh

# Custom prefix
./uninstall.sh --prefix $HOME/.local

Note: Uninstallation does not affect other system-wide tools or Git repositories you manage. It only removes the installed wrapper, symlink, and man page.

A.I. Automation

installing the Code Llama AI engine

$ sudo apt update
$ sudo apt install python3.12-venv
$ python3 -m venv ~/codellama-venv
$ source ~/codellama-venv/bin/activate # run this whenever you start a new terminal session
$ pip install --upgrade pip
$ pip install torch transformers accelerate sentencepiece  # This combined is the AI engine / runtime

• torch - the neural network engine (PyTorch created by Meta now used by almost everyone)
• transformers - loads and runs LLM models (Hugging Face Transformers support thousands of models)
• accelerate - helps with GPU/CPU optimization (Hugging Face Accelerate)
• sentencepiece - tokenizer used by Llama models created by Google

install the 7B model weights (~27GB)

$ mkdir -p ~/models
$ cd ~/models
$ git lfs install
$ git clone https://huggingface.co/codellama/CodeLlama-7b-hf  # These contain the trained neural network weights.

Test A.I.

import sys
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_path = "/home/a/models/CodeLlama-7b-hf"

# -----------------------------------------
# Get prompt from command line argument
# -----------------------------------------
if len(sys.argv) < 2:
    print("Usage: python test_codellama.py \"your prompt here\"")
    sys.exit(1)

prompt = sys.argv[1]

print("Loading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(model_path)

print("Loading model (this may take a while)...")
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    torch_dtype=torch.float16,
    device_map="auto"
)

print("Model loaded!")

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

output = model.generate(
    **inputs,
    max_new_tokens=200
)

print("\nAI response:\n")
print(tokenizer.decode(output[0], skip_special_tokens=True))

Sample Usage:

python test_codellama.py "Hi how are you?"