Package managers are the foundation of every stack yet they haven’t evolved in decades—they still work the same way they always have.
- Why are they so slow? So clunky? So incapable?
- Why do you need to use a different one (with subtly different syntax) on every platform?
- Why is it that you can have any version (as long as its the latest) but you have to wait weeks for it?
Introducing tea
, the next-generation, cross-platform package manager from
the creator of brew
.
$ node --eval 'console.log("Hello World!")'
command not found: node
$ sh <(curl tea.xyz) --yes
installing ~/.tea…
$ node --eval 'console.log("Hello World!")'
tea: installing ~/.tea/nodejs.org/v19.5.0
Hello World!
With tea there is no install packages step. Just type the commands you need and tea takes care of the rest: fetching packages, constructing a virtual environment isolated from the rest of your system and then running your commands.
Change your thinking from “I want to install foo” to “I want to use foo”.
Check out # Magic to learn how this works.
Scripting’s been stuck in a dark age of Bash because it’s the only thing you can be sure is installed. Lame af right?
$ tea ./hello.go
installing go…
go: running hello.go
#…
$ tea https://examples.deno.land/color-logging.ts
tea: installing deno…
#…
# need more dependencies than the interpreter? tea reads YAML front matter
$ tea ./favicon-generator input.png
tea: installing image-magick, optipng, guetzli and 3 other packages…
# …
output: favicon-128.png…
$ cat favicon-generator
#!/usr/bin/env ruby
# ^^ tea reads the shebang and automatically installs ruby
#---
# dependencies:
# imagemagick.org: 4
# optipng.sourceforge.net: 1
#---
You’re a developer, you need specific versions of tools, not just whatever the package manager happens to have today.
$ node^16 --version
tea: installing node^16
v16.19.0
$ node~16.18 --version
tea: installing node~16.18
v16.18.1
# when you need even more control we support the full https://semver.org spec
# though you will need to invoke tea directly
$ tea +nodejs.org'>=15,<17' node --version
v16.19.0
In fact you can construct virtual environments of specific tools and versions encapsulated separately from the rest of your system.
$ tea +rust-lang.org^1.63 +python.org~3.11 sh
tea: this is a temporary shell containing rust-lang.org, python.org and 13 other packages
tea: type `exit` when done
tea $ rustc --version
rustc 1.65.0
tea $ python --version
Python 3.11.1
tea $ which rustc
~/.tea/rust-lang.org/v1.65.0/bin/rustc
# ^^ inside environments tools *are* in `PATH`
tea $ exit
$ which rustc
rustc not found
# remember: tea doesn’t install packages
And since typically different projects will need different versions, we support that too:
$ node --version
v19.3.0
$ jq '.tea += {
"dependencies": {"nodejs.org": "^18.4"}
}' package.json | sponge package.json
# ^^ specify the version of the tool you need in *that tool’s config*.
# we strongly prefer YAML front matter in a comment
# but here we add an isolated `tea` namespace since JSON can’t have comments
# NOTE tea will automatically fetch both jq and sponge when you run the above
$ cat <<EOF >>my-project/README.md
# Dependencies
| Project | Version |
| ---------- | ------- |
| nodejs.org | ^18.4 |
EOF
# ^^ optionally, tea can also parse your README
$ cd my-project
$ node --version
v18.13.0
$ cd ..
$ node --version
v19.3.0
Using existing files further cements the notion that using tea is completely optional for both you and your users.
They either use tea or they source the dependencies themselves.
Some files are indicators of what tools they need, so if you choose not to
explain them in a README that means you expect your users "just know" package.json
means node
, cargo.toml
means rust
and so on.
Which is why you may prefer to encode the data in the README
since then
it’s readable by both humans and tea.
tea is a standalone, cross-platform† binary. So if you wanted you could just cURL it:
$ curl -Lo tea https://tea.xyz/$(uname)/$(uname -m)
$ chmod u+x ./tea
$ echo '# tea *really is* a standalone binary' | ./tea --sync glow -
tea: installing charm.sh/glow
# `tea --sync` updates pkgs, but you have to call it *at least once*
# our installer does this for you normally
However, if you want tea’s shell magic, you’ll need our installer:
sh <(curl https://tea.xyz)
# • asks you to confirm before it sets up `~/.tea`
# • asks you to confirm before adding one line to your `~/.shellrc`
# • DOESN’T need `sudo` or the Xcode Command Line Tools
† we support macOS >= 11 and Linux (glibc >= 23) and WSL, Windows native is being built as we speak.
In fact, our one-liner abstracts away installing tea itself:
# will your node project work with bun.sh? find out without having to install it!
$ sh <(curl tea.xyz) bun run start
# if tea is installed, our one-liner uses the tea installation, if it’s not
# installed then it **doesn’t install tea** or any dependencies, it creates a
# sandbox and runs everything in there
So now, in your blog posts, tweets, and tutorials, you don’t have to start with any “how to install tea” preamble nor will readers need to google anything. If they want to learn about tea first, they can go to the same URL as they’re curl’ing. We already work on Linux, macOS, and WSL; soon we’ll support Windows natively.
As a bonus the installer also updates tea.
Package managers can’t install themselves (#cantfix
).
But there’s other ways:
$ brew install teaxyz/pkgs/tea-cli
# or just try it out first
$ docker run --rm -it teaxyz/cli
Even more ways to install tea
Take your pick:
Grab a standalone binary from GitHub release.
Or get a plain text listing of standalone binary download URLs:
$ curl dist.tea.xyz # then copy, paste and cURL again
Or here’s a fancy one-liner:
$ sudo install -m 755 <(curl --compressed -LSsf https://tea.xyz/$(uname)/$(uname -m)) /usr/local/bin/tea
On macOS you may need to unquarantine the binary before it will execute:
# download from https://github.com/teaxyz/cli/releases first $ xattr -d com.apple.quarantine ./tea
- uses: teaxyz/setup@v0
# ^^ https://github.com/teaxyz/setup
Our action installs your deps and makes them (and tea) accessible to the rest of the workflow.
Every project you work on needs different tools with different versions. Installing those tools globally makes no sense and could even cause subtle bugs during dev.
tea can determine the tools a project directory needs based on the files it finds. With our shell magic just step into the project directory and type commands; tea automatically fetches the specific versions those projects need and runs them.
We try to be as clever as possible, eg. we parse the node version out of
a GitHub Action’s action.yml
. If we see a .node-version
file, we add that
version of node to the environment. We want to support everything that makes
sense (your PRs are very welcome!)
Where there isn’t a convention for specifying tool versions, you can add
YAML front matter to its configuration file. For example, if we’re
talking a python poetry project, then you can specify your python version by
adding YAML front matter to its pyproject.toml
:
$ cat <<EoYAML >> pyproject.toml
# ---
# dependencies:
# python.org: ^3.11.3
# ---
EoYAML
If your poetry project needs other non pypa dependencies (like a c compiler)
then you can add them there too (eg. llvm.org
, our deps are always named
after project homepages because then you just google it and there’s no
ambiguity).
If you prefer we also support extracting your deps from a markdown table
under a # Dependencies
section in your README.md
:
$ cat <<EOF >>my-project/README.md
# Dependencies
| Project | Version |
| ---------- | ------- |
| go.dev | ^1 |
EOF
Using your README
is kinda neat; it makes this data both machine and human
readable.
PSA: Stop using Docker
Docker is great for deployment and cross compilation, but… let’s face it: it sucks for dev.
Docker stifles builders. It constricts you; you’re immalleable; tech marches onwards but your docker container remains immobile. Nobody knows how to use
docker
. Once thatDockerfile
is set up, nobody dares touch it.And let’s face it, getting your personal dev and debug tools working inside that image is incredibly frustrating. Why limit your potential?
Keep deploying with Docker, but use tea to develop.
Then when you do deploy you may as well install those deps with tea.
Frankly, tea is properly versioned (unlike system packagers) so with tea your deployments actually remain more stable.
We still need to do a bit of work here. We don’t install any packages when
you step into a directory as that would violate the principle of least
surprise. But this may mean the full environment doesn’t work initially.
You can fix this with a tea -SE && cd .
. We’re working on improving this.
- After changing any files you need to do a
cd .
to reload the environment. - While debugging you can do
tea -E
to see if we’re picking up your changes. - If it still doesn’t work open a ticket. This is either a bug or we don’t support your toolset yet.
There are all sorts of variables a developer needs when working on a project
and tea aims to make them available to you. Thus we provide SRCROOT
and
VERSION
‡ in addition to everything else required to make your devenv
function. To see the full environment for your project run tea -En
or simply
env
.
‡ extracted from the
README.md
orVERSION
files.
Our magic puts the entire open source ecosystem at your fingertips. Our installer enables it by adding some hooks to your shell:
- A hook when changing directory that sets up project environments
- Environments are just shell environment variables
- A hook for the “command not found” scenario that installs that command before running it †
Magic is entirely optional, tea is still entirely usable without it.
Generally we’d say our magic is for devs and not for ops.
† Our “command not found” magic only works at a terminal prompt. Thus eg. VSCode won’t magically find
deno
. Shell scripts won’t automatically install tools they try to run. This is intentional. Magic should not lead to anarchy. See our FAQ for more information.
$ which bun
bun not found
$ tea --dry-run bun --version
imagined: bun.sh^0.4
~/.tea/bun.sh/v0.4.0/bin/bun --version
$ bun --version
tea: installing bun.sh^0.4
0.4.0
$ which bun
bun not found
# `bun` is not in your `PATH`
# ∵ tea doesn’t install packages
# ∴ using tea doesn’t compromise your system’s integrity
$ tea bun --version
0.4.0
# ^^ the same as `bun --version` but without magic
Simply prefix everything with tea
, eg. tea npm start
.
Simply prefix commands with tea -E
, eg. tea -E npm start
.
Our installer asked if you wanted magic when you ran it. If you elected to install magic and no longer want it simply remove the one-liner from your shell’s configuration file.
Our magic is not automatically added to scripts, but you can manually add it:
source <(tea --magic=bash)
# ^^ you have to specify which shell tho
And of course you can also use our one-liner:
source <(curl tea.xyz | sh -s -- --magic=bash)
Thus you can make a script that can effortlessly use any tool from the open source ecosystem. If they have tea installed it uses their installation, if not it installs everything (including tea itself) to a temporary sandbox that’s gone when the script completes.
Our +pkg
syntax injects packages into an environment, the commands are
then run in that environment.
# a script that needs `convert`
$ tea +imagemagick.org my-script.sh
# a VHS script that needs `wget`
$ tea +gnu.org/wget vhs demo.tape
# if tea doesn’t provide the package, it passes the args through to your system
$ tea +neovim.io which nvim
~/.tea/neovim.io/v0.8.2/bin/nvim
# in fact, `tea` is like an “env++”: just like `env` our purpose is to
# construct environments. Our `++` is: we also fetch packages. Thus, if you
# don’t specify what we should do with the environment, we dump it:
$ tea +zlib.net
MANPATH=/Users/mxl/.tea/zlib.net/v1.2.13/share/man:/usr/share/man
PKG_CONFIG_PATH=/Users/mxl/.tea/zlib.net/v1.2.13/lib/pkgconfig
LIBRARY_PATH=/Users/mxl/.tea/zlib.net/v1.2.13/lib
CPATH=/Users/mxl/.tea/zlib.net/v1.2.13/include
XDG_DATA_DIRS=/Users/mxl/.tea/zlib.net/v1.2.13/share
Is there a file that we should know about? Is there some other magic we should inject in certain circumstances? Open a ticket.
You trust us to just work and make your workflows happen.
We take this job seriously and we go the extra mile on a per-package basis,
eg. our git ignores .DS_Store
files by default, and our RubyGems defaults to
user-installs and ensures gems are in PATH
. Also we insist our packages
are relocatable, which is why we can install in your home directory (but this
also means you could pick up the whole ~/.tea
directory and bundle it with
your app.) We also begin packaging new releases almost immediately as soon as
they go live using various automations.
If you have suggestions or ideas, start a discussion. If we agree, we’ll move it to an issue. Bug fixes straight to pull request or issue please!
Probably the place you’ll want to start is by supplementing the pantry.
tea
is written in TypeScript using deno.
git clone https://github.com/teaxyz/cli tea
cd tea
deno task run foo # runs the local checkout passing `foo` as an argument
deno task install # deploys the local checkout into your `~/.tea`
There are hundreds-of-thousands of open source projects, and we need your help supporting them! Check out the docs for the pantry to learn more.
$ tea --sync
# ^^ updates the pantries, and any packages in the virtual-environment
$ tea --sync +deno.land
# ^^ updates specific packages
$ sh <(curl tea.xyz) --sync
# ^^ updates `tea` as well
open $(tea --prefix)
We hope to improve this UX very soon.
Symlinks† to tea
automatically invoke their namesake:
$ ln -s $(which tea) /usr/local/bin/bun
$ bun --version
tea: installing bun…
bun 0.4.0
# you can version tools this way too
$ ln -s $(which tea) /usr/local/bin/bun~0.3
$ bun~0.3 --version
tea: installing bun=0.3.0
bun 0.3.0
# if you prefer you can symlink with a `tea+` or `tea_` prefix
$ ln -s $(which tea) /usr/local/bin/tea+node
$ tea+node --version
v19.3.0
† this doesn’t work on Linux, you’ll need to use hard-links. this is a platform limitation we cannot work around 😞
We intend to make a VSCode extension that automatically fetches the
environment for the active workspace. In the meantime, add tools to your PATH
as described in the above FAQ.
tea adheres to semantic versioning, and uses
semver
for parsing versions.
We list all packages at tea.xyz.
Or open ~/.tea/tea.xyz/var/pantry
.
We hope to improve this UX very soon, too.
We want to support all platforms. Start a discussion and let’s talk about how to move forward with that.
We may revisit it, but we realized quickly that because tea makes it
so trivial to use anything from the open source ecosystem, it also makes it trivial
for you as a developer to use xc
†, make
or just
or any of the
myriad of other tools that are tightly scoped to the initial goals of
executable markdown.
† xc actually is a more mature implementation of executable markdown and we think you should definitely check it out.
Delete everything under ~/.tea
. Job done.
Strictly speaking, there's also an (automatically silenced) one-liner in your shell’s configuration file (eg.
~/.zshrc
) you may want to remove.
We have further FAQs in our wiki. Failing that, Start a discussion and we’ll get back to you.
- Be non‑intrusive
don’t interfere with our users’ systems or habits
- Be “just works”
our users have better things to do than fix us
- Error messages must be excellent
trust that if it comes to it, our users can fix things provided we give them a helping hand
- Be intuitive
being clever is good—but don’t be so clever nobody gets it
- Resist complexity
rethink the problem until a simpler solution emerges
- Be fast
we are in the way of our users’ real work, don’t make them wait
If you got this error message, you need to install tea:
sh <(curl -Ssf https://tea.xyz)
.
This is a known and insidious bug. Please open a ticket and report it to us. tea’s magic should never significantly slow down directory changes.