A tour of Unison
This document introduces "the big idea" behind Unison and walks through the basics of using the Unison codebase manager to develop and publish your first Unison library. We will introduce bits and pieces of the core Unison language and its syntax as we go. The Unison language reference is a more in-depth resource on this if you have questions or want to learn more.
If you want to follow along with this document (highly recommended), this guide assumes you've already gone through the steps in the quickstart guide.
The source for this document is on GitHub. Feedback and improvements are most welcome!
The most important goal
If there is one motivating idea behind Unison, it's this: the technology for creating software should be thoughtfully crafted in all aspects. Needless complexity and difficulties should be stripped away, leaving only that exhilarating creative essence of programming that made many of us want to learn this subject in the first place. Or at the very least, if we can't have this, let's have programming be reasonable. The fact that things were done a certain way in the 1970s is not a good reason to keep doing them, especially if they make programming worse.
Sure, it's sensible to make compromises regarding when and where to innovate, rather than trying to revolutionize everything right now. But let's be honest that it's a compromise, and not forget to improve things later.
🧠 The big technical idea
Now, if there is one big technical idea behind Unison, explored in pursuit of the overall goal of making programming better, it's this: Unison definitions are identified by content. Each Unison definition is some syntax tree, and by hashing this tree in a way that incorporates the hashes of all that definition's dependencies, we obtain the Unison hash which uniquely identifies that definition. This is the basis for some serious improvements to the programmer experience: it eliminates builds and most dependency conflicts, allows for easy dynamic deployment of code, typed durable storage, and lots more.
When taken to its logical endpoint, this idea of content-addressed code has some striking implications. Consider this: if definitions are identified by their content, there's no such thing as changing a definition, only introducing new definitions. That's interesting. What may change is how definitions are mapped to human-friendly names. For example,
x -> x + 1 (a definition) as opposed to
Nat.increment (a name we associate with it for the purposes of writing and reading other code that references it). An analogy: Unison definitions are like stars in the sky. We can discover the stars in the sky and pick different names for these stars, but the stars exist independently of what we choose to call them.
But the longer you spend with the odd idea of content-addressed code, the more it starts to take hold of you. It's not arbitrary or strange, but a logical and sensible choice with tremendous practical benefits. You start to appreciate the simplicity of the idea and see the need for it everywhere ("this would be a lot easier if the code were content-addressed..."). Is it really feasible, though, to build a programming language around this idea?
Part of the fun in building Unison was in working through the implications of what seemed like a great core idea. A big question that arose: even if definitions themselves are unchanging, we do sometimes want to change which definitions we are interested in and assign nice names to. So how does that work? How do you refactor or upgrade code? Is the codebase still just a mutable bag of text files, or do we need something else?
We do need something else to make it nice to work with content-addressed code. In Unison we call this something else the Unison Codebase Manager.
👋 to the Unison codebase manager
When first launching Unison in a new directory, we get a message like:
No codebase exists here so I'm initializing one in: .unison/v1
What's happening here? This is the Unison Codebase Manager starting up and initializing a fresh codebase. We're used to thinking about our codebase as a bag of text files that's mutated as we make changes to our code, but in Unison the codebase is represented as a collection of serialized syntax trees, identified by a hash of their content and stored in a collection of files inside of that
The Unison codebase format has a few key properties:
- It is append-only: once a file in the
.unisondirectory is created, it is never modified or deleted, and files are always named uniquely and deterministically based on their content.
- As a result, a Unison codebase can be versioned and synchronized with Git or any similar tool and will never generate a conflict in those tools.
If you haven't already worked through the quickstart guide, let's download the Unison base library to the
.> pull https://github.com/unisonweb/base:.releases._latest base
This command uses Git behind the scenes to sync new definitions from the remote Unison codebase to the local codebase.
📁 Here the command is performed in the top-level namespace, represented by
.. The prompt shows us which namespace we are currently in. If we were in a different namespace, we would need to change the
pullcommand from using the relative path
baseto the absolute path
Because of the append-only nature of the codebase format, we can cache all sorts of interesting information about definitions in the codebase and never have to worry about cache invalidation. For instance, Unison is a statically-typed language and we know the type of all definitions in the codebase--the codebase is always in a well-typed state. So one thing that's useful and easy to maintain is an index that lets us search for definitions in the codebase by their type. Try out the following commands (new syntax is explained below):
.> find : [a] -> [a] 1. base.Heap.sort : [a] -> [a] 2. base.List.distinct : [a] -> [a] 3. base.List.reverse : [a] -> [a] 4. base.Heap.sortDescending : [a] -> [a] .> view 3 base.List.reverse : [a] -> [a] base.List.reverse as = use base.List +: base.List.foldl (acc a -> a +: acc)  as
Here, we did a type-based search for functions of type
[a] -> [a], got a list of results, and then used the
view command to look at the nicely formatted source code of one of these results. Let's introduce some Unison syntax:
base.List.reverse : [a] -> [a]is the syntax for giving a type signature to a definition. We pronounce the
:symbol as "has type", as in "reverse has the type
[a] -> [a]".
[Nat]is the syntax for the type consisting of lists of natural numbers (terms like
will have this type), and more generally
[Foo]is the type of lists whose elements have some type
- Any lowercase variable in a type signature is assumed to be universally quantified, so
[a] -> [a]really means and could be written
forall a . [a] -> [a], which is the type of functions that take a list whose elements are some (but any) type, and return a list of elements of that same type.
base.List.reversetakes one parameter, called
as. The stuff after the
=is called the body of the function, and here it's a block, which is demarcated by whitespace.
acc a -> ..is the syntax for an anonymous function.
- Function arguments are separated by spaces and function application binds tighter than any operator, so
f x y + g p qparses as
(f x y) + (g p q). You can always use parentheses to control grouping more explicitly.
- The declaration
use base.List +:lets us reference the function
+:. (This function prepends an element to the front of a list.) Use clauses like this can be placed in any Unison block; they don't need to go at the top of your file.
view base.List.foldlif you're curious to see how it's defined.
Names are stored separately from definitions so renaming is fast and 100% accurate
The Unison codebase, in its definition for
reverse, doesn't store names for the definitions it depends on (like the
foldl function); it references these definitions via their hash. As a result, changing the name(s) associated with a definition is easy.
Let's try this out.
reverse is defined using
l is a needless abbreviation for
left. Let's rename that to
List.foldLeft to make things clearer. Try out the following command (you can use tab completion here if you like):
.> move.term base.List.foldl base.List.foldLeft Done. .> view base.List.reverse base.List.reverse : [a] -> [a] base.List.reverse as = use base.List +: base.List.foldLeft (acc a -> a +: acc)  as
view shows the
foldLeft name now, so the rename has taken effect. Nice!
To make this happen, Unison just changed the name associated with the hash of
foldl in one place. The
view command just looks up the names for the hashes on the fly, right when it's printing out the code.
This is important: Unison isn't doing a bunch of text mutation on your behalf, updating possibly thousands of files, generating a huge textual diff, and also breaking a bunch of downstream library users who are still expecting that definition to be called by the old name. That would be crazy, right?
So rename and move things around as much as you want. Don't worry about picking a perfect name the first time. Give the same definition multiple names if you want. It's all good!
move.termintroduces a new name for a definition without removing the old name(s).
🤓 If you're curious to learn about the guts of the Unison codebase format, you can check out the v1 codebase format specification.
undo to back up a step. (We don't have a
redo yet, though).
.> undo ⏪ Here's the changes I undid: > Moves: Original name New name base.List.foldl base.List.foldLeft .>
Great! OK, go drink some water, and then let's learn more about Unison's interactive way of writing and editing code.
Unison's interactive scratch files
The codebase manager lets you make changes to your codebase and explore the definitions it contains, but it also listens for changes to any file ending in
.u in the current directory. When any such file is saved (which we call a "scratch file"), Unison parses and typechecks that file. Let's try this out.
ucm terminal running and open up a file,
foo.u, or whatever you like) in your preferred text editor (if you want syntax highlighting for Unison files, follow this link for instructions on setting up your editor).
Now put the following in your scratch file:
use .base square : Nat -> Nat square x = x * x
This defines a function called
square. It takes an argument called
x and it returns
x multiplied by itself.
The first line,
use .base, tells Unison that you want to use short names for the base libraries in this file (which allows you to say
Nat instead of having to say
When you save the file, Unison replies:
✅ I found and typechecked these definitions in ~/unisoncode/scratch.u. If you do an `add` or `update` , here's how your codebase would change: ⍟ These new definitions are ok to `add`: square : base.Nat -> base.Nat Now evaluating any watch expressions (lines starting with `>`)... Ctrl+C cancels.
It typechecked the
square function and inferred that it takes a natural number and returns a natural number, so it has the type
Nat -> Nat. It also tells us that
square is "ok to
add". We'll do that shortly, but first, let's try calling our function right in the
scratch.u file, just by starting a line with
use .base square : Nat -> Nat square x = x * x > square 4
And Unison replies:
6 | > square 4 ⧩ 16
6 | is the line number from the file. The
> square 4 on line 6 of the file, starting with a
> is called a "watch expression", and Unison uses these watch expressions instead of having a separate read-eval-print-loop (REPL). The code you are editing can be run interactively, right in the same spot as you are doing the editing, with a full text editor at your disposal, with the same definitions all in scope, without needing to switch to a separate tool.
use .base is a wildcard use clause. This lets us use anything from the
base namespace under the root unqualified. For example we refer to
base.Nat as simply
Question: do we really want to reevaluate all watch expressions on every file save? What if they're expensive? Luckily, Unison keeps a cache of results for expressions it evaluates, keyed by the hash of the expression, and you can clear this cache at any time without ill effects. If a result for a hash is in the cache, Unison returns that instead of evaluating the expression again. So you can think of and use your
.u scratch files a bit like spreadsheets, which only recompute the minimal amount when dependencies change.
🤓 There's one more ingredient that makes this work effectively, and that's functional programming. When an expression has no side effects, its result is deterministic and you can cache it as long as you have a good key to use for the cache, like the Unison content-based hash. Unison's type system won't let you do I/O inside one of these watch expressions or anything else that would make the result change from one evaluation to the next.
Let's try out a few more examples:
-- A comment, ignored by Unison > base.List.reverse [1,2,3,4] > 4 + 6 > 5.0 / 2.0 > not true
✅ ~/unisoncode/scratch.u changed. Now evaluating any watch expressions (lines starting with `>`)... Ctrl+C cancels. 6 | > base.List.reverse [1,2,3,4] ⧩ [4, 3, 2, 1] 7 | > 4 + 6 ⧩ 10 8 | > 5.0 / 2.0 ⧩ 2.5 9 | > not true ⧩ false
Testing your code
Let's add a test for our
use .base square : Nat -> Nat square x = x * x use test test> square.tests.ex1 = check (square 4 == 16)
Save the file, and Unison comes back with:
7 | test> square.tests.ex1 = check (square 4 == 16) ✅ Passed : Passed 1 tests.
Some syntax notes:
test>prefix tells Unison that what follows is a test watch expression. Note that we're also giving a name to this expression,
- Note: there's nothing special about the name
square.tests.ex1; we could call those bindings anything we wanted. Here we use the convention that tests for a definition
check function has type
Boolean -> Test.Result. It takes a
Boolean expression and gives back a list of test results, of type
view Test.Result). In this case there was only one result, and it was a passed test.
A property-based test
Let's test this a bit more thoroughly.
square should have the property that
square a * square b == square (a * b) for all choices of
b. The testing library supports writing property-based tests like this. There's some new syntax here, explained afterwards:
use .base square : Nat -> Nat square x = x * x use test test> square.tests.ex1 = check (square 4 == 16) test> square.tests.prop1 = go _ = a = !nat b = !nat expect (square a * square b == square (a * b)) runs 100 go
8 | go _ = a = !nat ✅ Passed : Passed 100 tests. (cached)
This will test our function with a bunch of different inputs.
- The Unison block which begins after an
=begins a Unison block, which can have any number of bindings (like
a = ...) all at the same indentation level, terminated by a single expression (here
expect (square ..)), which is the result of the block.
- You can call a function parameter
_if you just plan to ignore it. Here,
goignores its argument; its purpose is just to make
golazily evaluated so it can be run multiple times by the
!exprmeans the same thing as
expr (), we say that
!exprforces the delayed computation
nat comes from
test.nat. It's a generator of natural numbers.
!nat generates one of these numbers.
Adding code to the codebase
square function and the tests we've written for it are not yet part of the codebase. So far they only exists in our scratch file. Let's add it now. Switch to the Unison console and type
add. You should get something like:
.> add ⍟ I've added these definitions: square.tests.ex1 : [base.Test.Result] square.tests.prop1 : [base.Test.Result] square : base.Nat -> base.Nat
You've just added a new function and some tests to your Unison codebase. Try typing
view square or
view square.tests.prop1. Notice that Unison inserts precise
use statements when rendering your code.
use statements aren't part of your code once it's in the codebase. When rendering code, a minimal set of
use statements is inserted automatically by the code printer, so you don't have to be precise with your
If you type
test at the Unison prompt, it will "run" your test suite:
.> test Cached test results (`help testcache` to learn more) ◉ square.tests.ex1 : Passed 1 tests. ◉ square.tests.prop1 : Passed 100 tests. ✅ 2 test(s) passing Tip: Use view square.tests.ex1 to view the source of a test.
But actually, it didn't need to run anything! All the tests had been run previously and cached according to their Unison hash. In a purely functional language like Unison, tests like these are deterministic and can be cached and never run again. No more running the same tests over and over again!
Unison namespaces and use clauses
Now that we've added our
square function to the codebase, how do we reference it elsewhere?
The Unison namespace is the mapping from names to definitions. Names in Unison look like this:
foo. That is: an optional
., followed by one or more segments separated by a
., with the last segment allowed to be an operator name like
We often think of these names as forming a tree, much like a directory of files, and names are like file paths in this tree. Absolute names (like
.base.Int) start with a
. and are paths from the root of this tree and relative names (like
math.sqrt) are paths starting from the current namespace, which you can set using the
namespace (or equivalently
.> namespace mylibrary ☝️ The namespace .mylibrary is empty. .mylibrary>
Notice the prompt changes to
.mylibrary>, indicating your current namespace is now
.mylibrary. When editing scratch files, any relative names not locally bound in your file will be resolved by prefixing them with the current namespace of
.mylibrary. And when you issue an
add command, the definitions are put directly into this namespace. For instance, if we added
x = 42 to our scratch file and then did
.mylibrary> add, that would create the definition
You can use
namespace .to move back to the root.
When we added
square, we were at the root, so
square and its tests are directly under the root. To keep our root namespace a bit tidier, let's go ahead and move our definitions into the
.mylibrary> move.term .square square Done. .mylibrary> find 1. square : .base.Nat -> .base.Nat .mylibrary> move.namespace .square.tests square.tests Done.
.square to refer to the
square definition directly under the root, and then moving it to the relative name
square. When you're done shuffling some things around, you can use
find with no arguments to view all the definitions under the current namespace:
.mylibrary> find 1. square.tests.ex1 : [.base.Test.Result] 2. square.tests.prop1 : [.base.Test.Result] 3. square : .base.Nat -> .base.Nat
Also notice that we don't need to rerun our tests after this reshuffling. The tests are still cached:
.mylibrary> test Cached test results (`help testcache` to learn more) ◉ square.tests.ex1 : Passed 1 tests. ◉ square.tests.prop1 : Passed 100 tests. ✅ 2 test(s) passing Tip: Use view square.tests.ex1 to view the source of a test.
We get this for free because the test cache is keyed by the hash of the test, not by what the test is called.
usestatement can do absolute names as well, for instance
use .base.List map.
When you're starting out writing some code, it can be nice to just put it in a temporary namespace, perhaps called
scratch. Later, without breaking anything, you can move that namespace or bits and pieces of it elsewhere, using the
Modifying existing definitions
Instead of starting a function from scratch, often you just want to slightly modify something that already exists. Here we'll make a change to the implementation of our
edit square from your prompt (note you can use tab completion):
.mylibrary> edit square ☝️ I added these definitions to the top of ~/unisoncode/scratch.u square : .base.Nat -> .base.Nat square x = use .base.Nat * x * x You can edit them there, then do `update` to replace the definitions currently in this branch.
This copies the pretty-printed definition of
square into you scratch file "above the fold". That is, it adds a line starting with
--- and puts whatever was already in the file below this line. Unison ignores any file contents below the fold.
Notice that Unison has put the correct type signature on
square. The absolute names
.base.Natlook a bit funny. We will often do
use .baseat the top of our file to refer to all the basic functions and types in
.basewithout a fully qualified name.
square and instead define
square x (just for fun) as the sum of the first
x odd numbers (here's a nice geometric illustration of why this gives the same results):
use .base square : Nat -> Nat square x = sum (map (x -> x * 2 + 1) (range 0 x)) sum : [Nat] -> Nat sum = foldl (+) 0
✅ I found and typechecked these definitions in ~/unisoncode/scratch.u. If you do an `add` or `update` , here's how your codebase would change: ⍟ These new definitions will replace existing ones of the same name and are ok to `update`: square : .base.Nat -> .base.Nat Now evaluating any watch expressions (lines starting with `>`)... Ctrl+C cancels.
Adding an updated definition to the codebase
Notice the message says that
square is "ok to
update". Let's try that:
.mylibrary> update ⍟ I've added these definitions: sum : [.base.Nat] -> .base.Nat ⍟ I've updated to these definitions: square : .base.Nat -> .base.Nat
Only affected tests are rerun on
If we rerun the tests, the tests won't be cached this time, since one of their dependencies has actually changed:
.mylibrary> test New test results: ◉ square.tests.prop1 : Passed 100 tests. ◉ square.tests.ex1 : Passed 1 tests. ✅ 2 test(s) passing Tip: Use view square.tests.prop1 to view the source of a test.
Notice the message indicates that the tests weren't cached. If we do
test again, we'll get the newly cached results.
The dependency tracking for determining whether a test needs rerunning is 100% accurate and is tracked at the level of individual definitions. You'll only rerun a test if one of the individual definitions it depends on has changed.
Publishing code and installing Unison libraries
Code is published using the
push command and libraries are installed just via the
pull command (recall how in the quickstart guide, we installed the base libraries with a
pull). There's no separate tooling needed for managing dependencies or publishing code and you'll never encounter dependency conflicts in Unison.
This document covers the details of how to organize your codebase, issue and review pull requests, install libraries, and make releases.
Before getting going writing Unison code, you can configure UCM to set author and license information. Also see the guide on organizing your codebase and day-to-day workflows.
- Libraries has a catalog of open source Unison projects. If you're working on something in Unison, feel free to open a PR to add it here.
- Abilities in Unison covers a unique aspect of Unison's type system.
- Writing documentation is a tutorial on Unison's approach to documentation.
- The core language reference describes Unison's core language and current syntax in more detail.