Skip to content

A cross-platform installer for the Julia programming language

License

Notifications You must be signed in to change notification settings

johnnychen94/jill.py

Repository files navigation

JILL.py

The enhanced Python fork of JILL -- Julia Installer for Linux (and every other platform) -- Light

py version version Actions Status codecov OSCS release-date 中文README

Features

  • download Julia releases from the nearest mirror server
  • support all platforms and architectures
  • manage multiple julia releases
  • easy-to-use CLI tool

asciicast

Install JILL

For the first time users of jill, you will need to install it using pip: pip install jill --user -U. Also use this to upgrade JILL version.

Python >= 3.6 is required. For base docker images, you also need to make sure wget and gnupg are installed.

Installing Julias

When you type jill install, it does the following things:

  1. query the latest version
  2. download, verify, and install julia
  3. make symlinks, e.g., julia, julia-1, julia-1.6

For common Julia users:

  • Get the latest stable release: jill install
  • Get the latest 1.y.z release: jill install 1
  • Get the latest 1.6.z release: jill install 1.6
  • Get the specific version: jill install 1.6.2, jill install 1.7.0-beta3
  • Get the latest release (including unstable ones): jill install --unstable

Note that for Julia 1.10, you'll have to install it with jill install '"1.10"' because of the python-fire limit.

For Julia developers and maintainers:

  • Get the nightly builds: jill install latest. This gives you julia-latest.
  • Checkout CI build artifacts of specific commit in the Julia Repository: jill install 1.8.0+cc4be25c (<major>.<minor>.<patch>+<build> with at least the first 7 characters of the hash). This gives you julia-dev.

Some flags that can be useful:

  • No confirmation before installation: jill install --confirm
  • Download from Official source: jill install --upstream Official
  • Keep downloaded contents after installation: jill install --keep_downloads
  • Force a reinstallation: jill install --reinstall

The symlinks

To start Julia, you can use predefined JILL symlinks such as julia. jill install uses the following rule makes sure that you're always using the latest stable release.

Stable releases:

  • julia points to the latest Julia release.
  • julia-1 points to the latest 1.y.z Julia release.
  • julia-1.6 points to the latest 1.6.z Julia release.

For unstable releases such as 1.7.0-beta3, installing it via jill install 1.7 --unstable or jill install 1.7.0-beta3 will only give you julia-1.7; it won't make symlinks for julia or julia-1.

To dance on edge:

  • julia-latest points to the nightly build from jill install latest
  • julia-dev points to the julia CI build artifacts from, for example, jill install 1.8.0+cc4be25c.

List symlinks and their target versions

jill list [version] gives you every symlinks and their target Julia versions.

list

Change symlink target

For non-windows system, you are free to use ln command to change the symlink targets. For Windows it uses an entry .cmd file for this so you'll need to copy them. In the meantime, jill switch provides a simple and unified way to do this:

  • jill switch 1.6: let julia points to the latest julia 1.6.z release.
  • jill switch <path/to/my/own/julia/executable>: let julia points to custom executables.
  • jill switch 1.6 --target julia-1: let julia-1 points to the latest julia 1.6.z release.

About downloading upstreams

By default, JILL tries to be smart and will download contents from the nearest upstream. You can get the information of all upstreams via jill upstream. Here's what I get in my laptop, I live in China so the official upstreams aren't so accessible for me :(

upstream

To temporarily disable this feature, you can use flag --upstream <server_name>. For instance, jill install --upstream Official will faithfully download from the official julialang s3 bucket.

To permanently disable this feature, you can set environment variable JILL_UPSTREAM.

Note that flag is of higher priority than environment variable. For example, if JILL_UPSTREAM is set to mirror server "TUNA", you can still download from the official source via jill install --upstream Official.

About installation and symlink directories

Here's the default JILL installation and symlink directories:

system installation directory symlink directory
macOS /Applications ~/.local/bin
Linux/FreeBSD ~/packages/julias ~/.local/bin
Windows ~\AppData\Local\julias ~\AppData\Local\julias\bin

For example, on Linux jill install 1.6.2 will have a julia folder in ~/packages/julias/julia-1.6 and symlinks julia/julia-1/julia-1.6 created in ~/.local/bin.

Particularly, if you're using jill as root user, you will do a system-wide installation:

  • Installation directory will be /opt/julias for Linux/FreeBSD.
  • Symlink directory will be /usr/local/bin for Linux/FreeBSD/macOS.

To change the default JILL installation and symlink directories, you can set environment variables JILL_INSTALL_DIR and JILL_SYMLINK_DIR.

(Deprecated) jill install also provides two flag --install_dir <dirpath> and --symlink_dir <dirpath>, they have higher priority than the environment variables JILL_INSTALL_DIR and JILL_SYMLINK_DIR.

JILL environment variables

jill is made as a convenient tool and it can sometimes be annoying passing flags to it. There are some predefined environment variables that you can use to set the default values:

  • Specify a default downloading upstream JILL_UPSTREAM: --upstream
  • Override default symlink directory JILL_SYMLINK_DIR: --symlink_dir
  • Override default installation directory JILL_INSTALL_DIR: --install_dir

The flag version has higher priority than the environment variable version.


Advanced: Example with cron

If you're tired of seeing (xx days old master) in your nightly build version, then jill can make your nightly build always the latest version using cron:

# /etc/cron.d/jill
PATH=/usr/local/bin:/usr/sbin:/usr/sbin:/usr/bin:/sbin:/bin

# install a fresh nightly build every day
* 0 * * * root jill install latest --confirm --upstream Official

Advanced: Registering a new public releases upstream

If it's an public mirror and you want to share it worldwide to other users of JILL. You can add an entry to the public registry, make a PR, then I will tag a new release for that.

Please check the sources.json format for more detailed information on the format.

Advanced: Specifying custom (private) downloading upstream

To add new private upstream, you can create a file ~/.config/jill/sources.json (fow Windows it is ~/AppData/Local/julias/sources.json) and add your own upstream configuration just like the JILL sources.json does. Once this is done JILL will recognize this new upstream entry.

Please check the sources.json format for more detailed information on the format.

Advanced: make a Julia release mirror

There are two ways to do so:

  • use aws s3 sync, this should be the easiest way to do so I highly recommend this.
  • (Deprecated) use jill mirror command with mirror config example. I didn't know about the aws s3 sync stuff when I implemented this.

The Julia release mirror does not contain Julia package contents, to mirror all the Julia packages and artifacts (which requires >1.5Tb storage), you can use StorageMirrorServer.jl.

Advanced: The Python API

jill.py also provides a set of Python API:

from jill.install import install_julia
from jill.download import download_package

# equivalent to `jill install --confirm`
install_julia(confirm=True)
# equivalent to `jill download`
download_package()

You can read its docstring (e.g., ?install_julia) for more information.

FAQs

Why you should use JILL?

Distro package managers (e.g., apt, pac) is likely to provide a broken Julia with incorrect binary dependencies (e.g., LLVM ) versions. Hence it's recommended to download and extract the Julia binary provided in Julia Downloads. jill.py doesn't do anything magical, but just makes such operation even stupid.

Why I make the python fork of JILL?

At first I found myself needing a simple tool to download and install Julia on my macbook and servers in our lab, I made my own shell scripts and I'd like to share it with others. Then I found the jill.sh project, Abel knows a lot shell so I decide to contribute my macOS Julia installer to jill.sh.

There are three main reasons for why I decided to start my Python fork:

  • I live in China. Downloading resources from GitHub and AWS s3 buckets is a painful experience. Thus I want to support downloading from mirror servers. Adding mirror server support to jill.sh is quite complicated and can easily become a maintenance nightmare.
  • I want to make a cross platform installer that everyone can use, not just Linux/macOS users. Shell scripts doesn't allow this as far as I can tell. In contrast, Python allows this.
  • Most importantly, back to when I start this project, I knew very little shell, I knew nothing about C/C++/Rust/Go and whatever you think a good solution is. I happen to knew a few Python.

For some "obvious" reason, Julia People don't like Python and I understand it. (I also don't like Python after being advanced Julia user for more than 3 years) But to be honest, revisiting this project, I find using Python is one of the best-made decision during the entire project. Here is the reason: no matter how you enjoy Julia (or C++, Rust), Python is one of the best successful programming language for sever maintenance purpose. Users can easily found tons of "how-to" solutions about Python and it's easy to write, deploy, and ship Python codes to the world via PyPI.

And again, I live in China so I want to rely on services that are easily accessible in China, PyPI is, GitHub and AWS S3 bucket aren't. A recent Julia installer project juliaup written in Rust solves the Python dependency problem very well, but the tradeoff is that juliaup needs its own distributing system (currently GitHub and S3 bucket) to make sure it can be reliably downloaded to user machine. And for this it just won't be as good as PyPI in the foreseeable future.

Is it safe to use jill.py?

Yes, jill.py use GPG to check every tarballs after downloading. Also, *.dmg/*.pkg for macOS and .exe for Windows are already signed.

What's the difference between jill.sh and jill.py

jill.sh is a shell script that works quite well on Linux x86/x64 machines. jill.py is an enhanced python package that focus on Julia installation and version management, and brings a unified user experience on all platforms.

Why julia fails to start

The symlink julia are stored in JILL predefined symlinks dir thus you have to make sure this folder is in PATH. Search "how to add folder to PATH on xxx system" you will get a lot of solutions.

How do I use multiple patches releases (e.g., 1.6.1 and 1.6.2)

Generally, you should not care about patch version differences so jill.py make it explicitly that only one of 1.6.x can exist. If you insist to have multiple patch versions, you could use jill install --install_dir <some_other_folder> to install Julia in other folder, and then manually make a symlink back. As I just said, in most cases, common users should not care about this patch version difference and should just use the latest patch release.

How to only download contents without installation?

Use jill download [version] [--sys <system>] [--arch <arch>]. Check jill download --help for more details.

Linux with musl libc

For Julia (>= 1.5.0) in Linux with musl libc, you can just do jill install and it gives you the right Julia binary. To download the musl libc binary using jill download, you will need to pass --sys musl flag.

MacOS with Apple silicon (M1)

Yes it's supported. Because macOS ARM version is still of tier-3 support, jill.py will by default install the x86_64 version. If you want to use the ARM version, you can install it via jill install --preferred-arch arm64.

CERTIFICATE_VERIFY_FAILED error

If you're confident, try jill install --bypass-ssl.

Skip symbolic links generation

If for some reason you prefer to download julia without generating symbolic links jill install --skip-symlinks