The Embedded Compiler¶

Hobbes is a simple language with a rich type system, whose driving aim is to allow efficient reimplementation of functional logic at runtime.

As such, it comes packaged with a highly efficient compiler and type marshalling system.

Example: Hi, the Hobbes REPL¶

We’ve already seen one place where the Hobbes compiler is used. In its simplest form, the Hobbes REPL is a loop which takes Hobbes code, runs it line-by-line through the compiler, interprets the results and prints the output:

int main() {
  hobbes::cc c;

  while (std::cin) {
    std::cout << "> " << std::flush;
    std::string line;
    std::getline(std::cin, line);

    try {
      c.compileFn<void()>("print(" + line + ")")();
    } catch (std::exception& ex) {
      std::cout << "*** " << ex.what();
    }

    std::cout << std::endl;
    hobbes::resetMemoryPool();
  }
}

Binding Functions¶

The Hobbes compiler allows us to bind a C++ function; to make it available by name in the Hobbes environment:

int addTwo(int i){
  return 2 + i;
}

...
c.bind("addTwo", &addTwo);
...

Note

Function Pointers

A key aspect of functional programming is that functions are just special kinds of data, and that whilst there may be operations which can be performed on data, the application which can be performed on a function is that of parameter application. Once a function has had all its parameters applied the code is executed and the return value given back to the callsite.

This has a two major implications: we can assign functions to values (as we’ve already seen), and we can write higher-order functions: those which take another function as a parameter, or return a function. In C++, the type of a function is denoted entirely by its signature, the abstract format for which is the function pointer:

int sum(int a, int b){
  return a + b;
}

int (*twoIntOperation)(int, int) = sum;

We declare sum as the concrete example of a function which takes two ints and returns an int, and assign it to the variable twoIntOperation. The type of twoIntOperation is exactly that: a function which takes two ints and returns one.

The syntax is a little hairy. In Hobbes, the function type syntax is a lot clearer, as we’ll see very soon.

Once I’ve bound addTwo to the Hobbes compiler c it becomes available in the REPL environment. By invoking addTwo, I can see that Hobbes knows the type signature and knows how to invoke the function:

> addTwo
*** stdin:1,1-13: Failed to compile expression due to unresolved type constraint:
  Print (int) -> int
> addTwo(3)
5
>

Hobbes has a rich binding environment, with bi-directional support for marshalling common C++ types (as we’ve already seen, with int), as well as C++ structs, and the hobbes types such as record and variant.

Simple scalar types¶

As we’ve already seen, the Tuple is an important and common basic type in Functional Programming, used to keep elements of data together in a small lexical scope. Over time many of these functional types have ‘leaked’ into C++, where we now have std::tuple and the “two-tuple” special case, std::pair.

Hobbes is aware of the parametric std::pair and std::tuple types and composes them appropriately:

typedef std::pair<int, const hobbes::array<char>*> Writer;

Writer* getWriter(){
  return hobbes::make<Writer>(34, hobbes::makeString("Sam"));
}

...
c.bind("getWriter", &getWriter);
...

Then, in the Hobbes REPL:

> getWriter()
(34, "Sam")

Structs¶

There’s usually a point at which our pairs or tuples grow in importance in our domain, and we want to give names to the members. In C++ we might use a struct for this purpose. Hobbes allows us to expose our C++ structs with the DEFINE_STRUCT macro:

DEFINE_STRUCT(Writer,
 (size_t, age),
 (const hobbes::array<char>*, name)
);

hobbes::array<Writer>* getWriters(){
  auto writers = hobbes::makeArray<Writer>(2);

  writers->data[0].age = 21;
  writers->data[0].name = hobbes::makeString("John");

  writers->data[1].age = 22;
  writers->data[1].name = hobbes::makeString("Paul");

  return writers;
}

...
c.bind("getWriters", $getWriters);
...

…and in the REPL:

> getWriters
age   name
___ ______
21    John
22    Paul

Note

Hobbes has been able to determine appropriate column names from the struct definition, just as it does with our hobbes-native record type!

Variant¶

Slightly more complex, our ‘OR’ type, the variant:

typedef hobbes::variant<int, const hobbes::array<char>*> CountOrMessage;

CountOrMessage* classify(int i){
  if(i<22){
    return hobbes::make<CountOrMessage>(i);
  }else{
    return hobbes::make<CountOrMessage(hobbes::makeString("haha"));
  }
}

...
c.bind("classify", $classify);
...

In our example we define our variant type in C++, and then create an instance depending on the value of some function parameter. Then in the Hobbes REPL we are able to call the bound function classify and deal with the result in a functional manner:

> classify(12)
|0=12|
> classify(42)
|1="haha!"|

Completing the round trip¶

In Hobbes we can expose higher-order functions simply. This allows us to expose Hobbes functionality in C++, completing the round trip!

As a Haskell-like language, the syntax is elegant. In the following example we declare a function binaryIntFn which takes two ints and returns an int. By compiling this function and binding it to the runtime Hobbes environment, we’re able to “plug in” behaviour based on elements of the runtime environment:

int binaryIntFn(int (*pf)(int, int), int x){
  return pf(x, x);
}

>  binaryIntFn(\x y.x+y, 3)
6
> binaryIntFn(\x y.x*y, 4)
16