The language was born in the mind of a learner. It was good. The learner wanted to know about programming languages. He learned deeply. He learned about lambda calculus and type theory. He learned about System-F and typeclasses. He learned about unification and compilers. He built lexers and parsers with his hands. He generated code with LLVM. He learned many things. The work was hard but good. It ended in 2020.

The language was like Haskell and C together. It had garbage collection. It was eager. It checked types. It was pure but could change when needed. It had operators you could extend. It had typeclasses. It managed resources well. It matched patterns and checked branches. It knew about types without being told.

It was a good language. It was called Ace.

Note: these instructions are likely no longer functional.

To play with Ace in Docker, try this.

git clone https://github.com/wbbradley/ace.git
cd ace

# Get a docker image set up ready to run a build (assumes Docker is running).
./docker-build.sh && ./docker-run.sh bash

# The prior command should open up a bash prompt within a new docker container.
# Build and install Ace inside this container.
make install

# The prior command should have installed Ace to /usr/local. Set up the
# $ACE_ROOT environment variable.
export ACE_ROOT="/usr/local/share/ace"

# Build and run a simple test program
cd
echo 'fn main() { print("Hello world.") }' > hello_world.ace
ace hello_world

# Read more
man ace

Ace is a statically typed procedural/functional language. It is a work in progress. Please reach out if you'd like to get involved.

fn main() {
  print("Hello world.")
}

The syntax resembles C or Python (with braces.) The type system is based on System F with extensions for Type Classes, newtypes and pattern matching. There is no macro system but there is a rich syntax available via reader macros within the parser.

fn main() {
  let filename = "some-file.txt"
  # 'with' gives guarantees that the value can clean itself up. See std.WithElseResource.
  with let f = open(filename) {
    for line in readlines(f) {
      print(strip(line))
    }
  } else errno {
    print("Failed to open ${filename}: ${errno}")
  }
}

import itertools {zip}

fn main() {
  # Multiply some zipped Ints and put them into a Vector
  print([x * y for (x, y) in zip([1..3], [4..])])
  # prints [4, 10, 18]...
}

fn main() {
  let double = fn (x) => x * 2
  assert(double(25) == 50)
}

fn main() {
  let double = |x| => x * 2
  assert(double(25) == 50)
}

The evaluation of Ace is strict, not lazy. The call-by-value method of passing arguments is used.

There is no explicit notion of immutability, however it is implicit unless var is used. var declarations wrap initialization values in a mutable reference cell. Under the covers, this is the std.Ref type. The primary way to maintain mutable state is to use var.

fn main() {
  # Create a value with let. By default it is immutable.
  let y = 4
  # Try to change it...
  y = 5          // error: type error

  # Create a variable with var. It is mutable.
  var x = 5
  print("${x}")  // Prints "5"

  # Change what is in the memory cell described by x...
  x = 7
  print("${x}")  // Prints "7"

  # Try putting some other type of thing in there...
  x = "hey!"     // error: type error. Int != string.String
}

There is no class-based encapsulation in Ace. Encapsulation can be achieved by

1. using modules to implement Abstract Data Types, exposing only the functions relevant to the creation, use, and lifetime of a type.
2. not letting local variables escape from functions (or blocks), or by using module-local functions.

Ace lacks support for shared libraries or any shareable intermediate representation. Code complexity and leaky abstractions can still be avoided by limiting which symbols are exposed from source modules.

Types are inferred but type annotations are also allowed/encouraged as documentation and sometimes necessary when types cannot be inferred. Ace rejects intermediate type defaulting by design. Although, if a good design for that comes along, it might happen.

Polymorphism comes in two flavors.

Type-based polymorphism exists at compile time in the ability to use type variables which are re-bound per function invocation. At run-time, there are a couple different notions of polymorphism. First, Ace supports sum types by allowing the declaration of types with multiple data constructors. This then relies on match statements (pattern matching) to branch on the run-time value. This form of polymorphism may feel unfamiliar to folks coming from "OOP" languages that rely on inheritance and/or abstract classes with any number of derived implementations.

Since Ace treats functions as values and allows closure over function definitions (fn), you can return new behaviors as functions. Users of those functions will get statically checked run-time varying behavior (aka run-time polymorphism). For example, the Iterable type-class requires the definition of a single function which will itself return a function which can be called repeatedly to iterate. It has a signature like

class Iterable collection item {
  fn iter(collection) fn () Maybe item
}

So, on top of the type being returned by the iterator being compile-time polymorphic, the usage of such Iterables at run-time also involves a run-time closure that may have any number of behaviors or "shapes" in how it operates. Thus, it is polymorphic, but conforms to the specification that it must return a Maybe type. In this case, the Maybe type has two data constructors, Just and Nothing. If an iterator returns Nothing, it indicates that it is done iterating.

All code that is reachable from main is specialized and monomorphized prior to the final code generation phase. Code generation creates LLVM IR, which is passed through clang to perform static linking, optimization, and lowering to the target host.

The best way to learn more at this time is to read through the tests/test_*.ace code.

TODO: struct types do not support pattern matching. proposed solution: eliminate structs, but add names to newtypes.