Programming in Cairo

Your first function

Let’s start by looking at the following Cairo function which computes the sum of the elements of an array:

# Computes the sum of the memory elements at addresses:
#   arr + 0, arr + 1, ..., arr + (size - 1).
func array_sum(arr : felt*, size) -> (sum):
    if size == 0:
        return (sum=0)

    # size is not zero.
    let (sum_of_rest) = array_sum(arr=arr + 1, size=size - 1)
    return (sum=[arr] + sum_of_rest)

The first two lines are comment lines, and are ignored by the compiler. Comments in Cairo start with # and continue until the end of the line.

The first non-comment line func array_sum(arr : felt*, size) -> (sum): defines a function named array_sum which takes two arguments: arr and size. arr points to an array of size elements. The type of arr is felt* which is a pointer (for more information about felt, see below). The function declares that it returns one value called sum. The scope of the function ends with the word end (although end doesn’t mean that the function returns – you must add the return statement explicitly).

The next line if size == 0: instructs Cairo to execute the code in the if statement’s body if size is zero, and skip to the end of the if statement’s body otherwise.

When size == 0, there are no elements in the array, and so we can return that the sum is zero. Note that the syntax sum= is not mandatory (you could write return (0)), but it is recommended as it increases the readability of the code. On the other hand, the parentheses are required. As with other programming languages, the return statement ends the execution of the function immediately and returns the control to the calling function.

Now to the interesting part. The line let (sum_of_rest) = array_sum(arr=arr + 1, size=size - 1) (which is executed only if size != 0) makes a recursive call to array_sum(), starting from the second element (as arr points to the first memory cell of the array, arr + 1 points to the second cell. Also note that we need to pass size - 1). As with named value in the return statement, you don’t have to write arr= and size=. The expression let (sum_of_rest) = says that the function returns one value, which can be accessed using sum_of_rest.

The scope of the return values of functions is restricted (for example, they may be revoked due to jumps or function calls). Later, we will see how to overcome this by using local variables. You can read more in Revoked references.

Note that you cannot call functions as part of other expressions. For example trying to write something like foo(bar()) will not compile.

Finally, we return the sum of the first element with the sum of the rest of the elements. [...] is the dereference operator, so [arr] is the value of the memory at address arr, or in our case the first element of the array.


Write a function that computes the product of all the even entries of an array ([arr] * [arr + 2] * ...). You may assume that the length is even. (You may want to wait with running it until you read Using array_sum().)

A low-level language with powerful syntactic sugar

Cairo is not a high-level language. It’s a low-level language with some powerful syntactic sugar to allow writing maintainable code. The advantage is that the Cairo language allows you to write very efficient code (you can write in the Cairo language almost anything you can run on the Cairo machine). The main disadvantage is that in some places you’ll have to know what you’re doing in order to avoid some common mistakes. Don’t worry though, this document will guide you through those delicate places.

Recursion instead of loops

You may have noticed that we’ve used recursion in the code above rather than the loop structure you may have expected. The main reason for this is that the Cairo memory is immutable – once you write the value of a memory cell, this cell cannot change in the future. This is similar to pure functional languages, whose objects are also immutable, where you also have to replace loops with recursion for the same reason.

Having said that, we will note that loop structures (in some sense) are possible in Cairo, but they are limited (you cannot call functions inside a loop, for example) and more complicated to implement. Their only advantage is that they tend to be slightly more efficient than recursion.

The assert statement

The assert statement:

assert <expr0> = <expr1>

which we use below, allows us to do two things: verify that two values are the same (as you may have expected), but also to assign a value to a memory cell. For example, assert [ptr] = 0 will set the value of the memory cell at address ptr to 0 (if it was not set before). This has to do with the fact that the Cairo memory is immutable: If the values were previously set it will function as an assert statement. On the other hand, if the value on the left-hand side (in some simple cases it will work with the right-hand side as well) was not set yet, Cairo will set it, thus causing the assert to pass.

So how can I change the value of [ptr] if I already set it before? Won’t the assert statement function as an assert rather than an assignment? The answer is that you can’t – Cairo memory is immutable, which means that once a value was written to a memory cell, it cannot change.

You can read more in Memory model.

Writing a main() function

Before we write a main() function that will call array_sum(), let’s start with something simpler:

%builtins output

from starkware.cairo.common.serialize import serialize_word

func main{output_ptr : felt*}():
    return ()

There are a few new components we see here:

  1. The function main(): The main() function is the starting point of the Cairo program.

  2. The builtin directive and the output builtin: The directive %builtins output instructs the Cairo compiler that our program will use the “output” builtin. You can learn about builtins in general here. For now we will focus on the output builtin we’re using here.

    The output builtin is what allows the program to communicate with the external world. You can think of it as the equivalent of Python’s print() or C++’s std::cout. As with all builtins, we don’t have special instructions in Cairo to use them – the communication with the builtin is done by reading/writing values to the memory.

    The output builtin is quite simple: Declaring it using %builtins turns the signature of main() to main{output_ptr : felt*}(). The syntax {output_ptr : felt*} declares an “implicit argument”, which means that behind the scenes, it adds both a corresponding argument and return value. More information about implicit arguments can be found in Implicit arguments.

    The argument points to the beginning of the memory segment to which the program output should be written. The program should then return a pointer that marks the end of the output. The convention we use in Cairo is that the end of a memory segment always points to the memory cell after the last written cell. And indeed, this is what Cairo expects of the returned value.

  3. The function serialize_word(): To write the value x to the output, we can use the library function serialize_word(x). serialize_word gets one argument (the value we want to write) and one implicit argument output_ptr (which means that behind the scenes it also returns one value). In fact it’s quite simple: it writes x to the memory cell pointed by output_ptr (that is, [output_ptr]) and returns output_ptr + 1. Now the implicit argument mechanism kicks in: in the first call to serialize_word() the Cairo compiler passes the value of output_ptr as the implicit argument. In the second call it uses the value returned by the first call.

  4. Import statements: The line from starkware.cairo.common.serialize import serialize_word instructs the compiler to compile the file starkware/cairo/common/serialize.cairo, and to expose the identifier serialize_word. You can use ... import serialize_word as foo to choose a different name in which you can address serialize_word in the current module. You can learn more about the import statement here.

Running the code

Save the code above (with the main() function) as array_sum.cairo (later we will change it to call array_sum()), and compile and run it using the following commands:

cairo-compile array_sum.cairo --output array_sum_compiled.json

cairo-run --program=array_sum_compiled.json \
    --print_output --layout=small

You should get:

Program output:

The --layout flag is needed becuase we’re using the output builtin, which is not available in the plain layout (see Layouts).

The primitive type - field element (felt)

In Cairo when you don’t specify a type of a variable/argument, its type is a field element (represented by the keyword felt). In the context of Cairo, when we say “a field element” we mean an integer in the range \(-P/2 < x < P/2\) where \(P\) is a very large (prime) number (currently it is a 252-bit number, which is a number with 76 decimal digits). When we add, subtract or multiply and the result is outside the range above, there is an overflow, and the appropriate multiple of \(P\) is added or subtracted to bring the result back into this range (in other words, the result is computed modulo \(P\)).

The most important difference between integers and field elements is division: Division of field elements (and therefore division in Cairo) is not the integer division you have in many programming languages, where the integral part of the quotient is returned (so you get 7 / 3 = 2). As long as the numerator is a multiple of the denominator, it will behave as you expect (6 / 3 = 2). If this is not the case, for example when we divide 7/3, it will result in a field element x that will satisfy 3 * x = 7. It won’t be 2.3333 because x has to be an integer. If this seems impossible remember that if 3 * x is outside the range \(-P/2 < x < P/2\) an overflow will occur which can bring the result down to 7. It’s a well-known mathematical fact that unless the denominator is zero, there will always be a value x satisfying denominator * x = numerator.

Let’s try it! Modify the code in array_sum.cairo as follows:

%builtins output

from starkware.cairo.common.serialize import serialize_word

func main{output_ptr : felt*}():
    serialize_word(6 / 3)
    serialize_word(7 / 3)
    return ()

Use the commands above to run it (don’t forget to compile again, or you’ll get the same output you had before). You should get:

Program output:

Now, edit the code to print the result of multiplying the last number by 3 and verify that you indeed get 7.

You’ll see that in most of your code (unless you intend to write a very algebraic code), you won’t have to deal with the fact that the values in Cairo are field elements and you’ll be able to treat them as if they were normal integers.

Using array_sum()

Now, let’s write a main() function that will use array_sum(). To do this, we will need to allocate space for the array. This is done using the library function alloc():

%builtins output

from starkware.cairo.common.alloc import alloc
from starkware.cairo.common.serialize import serialize_word

func array_sum(arr, size) -> (sum):
    # ...

func main{output_ptr : felt*}():
    const ARRAY_SIZE = 3

    # Allocate an array.
    let (ptr) = alloc()

    # Populate some values in the array.
    assert [ptr] = 9
    assert [ptr + 1] = 16
    assert [ptr + 2] = 25

    # Call array_sum to compute the sum of the elements.
    let (sum) = array_sum(arr=ptr, size=ARRAY_SIZE)

    # Write the sum to the program output.

    return ()

Here we have a few additional new things:

  1. Memory allocation: We use the standard library function alloc() to allocate a new memory segment. In practice the exact location of the allocated memory will be determined only when the program terminates, which allows us to avoid specifying the size of the allocation.

  2. Constants: A constant in Cairo is defined using const CONST_NAME = <expr> where <expr> must be an integer (a field element, to be precise), known at compile time.

Compile and run the code (note that you’ll have to copy the code of array_sum() from the top of the page). You should get:

Program output:


If you haven’t done so already, try to run your product function using the main() above. Don’t forget to adjust the number of elements to an even number.