Functions

In Julia, a function is an object that maps a tuple of argument values to a return value. Julia functions are not pure mathematical functions, because they can alter and be affected by the global state of the program. The basic syntax for defining functions in Julia is:

  1. julia> function f(x,y)
  2. x + y
  3. end
  4. f (generic function with 1 method)

This function accepts two arguments x and y and returns the value of the last expression evaluated, which is x + y.

There is a second, more terse syntax for defining a function in Julia. The traditional function declaration syntax demonstrated above is equivalent to the following compact “assignment form”:

  1. julia> f(x,y) = x + y
  2. f (generic function with 1 method)

In the assignment form, the body of the function must be a single expression, although it can be a compound expression (see Compound Expressions). Short, simple function definitions are common in Julia. The short function syntax is accordingly quite idiomatic, considerably reducing both typing and visual noise.

A function is called using the traditional parenthesis syntax:

  1. julia> f(2,3)
  2. 5

Without parentheses, the expression f refers to the function object, and can be passed around like any other value:

  1. julia> g = f;
  2. julia> g(2,3)
  3. 5

As with variables, Unicode can also be used for function names:

  1. julia> ∑(x,y) = x + y
  2. (generic function with 1 method)
  3. julia> ∑(2, 3)
  4. 5

Argument Passing Behavior

Julia function arguments follow a convention sometimes called “pass-by-sharing”, which means that values are not copied when they are passed to functions. Function arguments themselves act as new variable bindings (new locations that can refer to values), but the values they refer to are identical to the passed values. Modifications to mutable values (such as Arrays) made within a function will be visible to the caller. This is the same behavior found in Scheme, most Lisps, Python, Ruby and Perl, among other dynamic languages.

Argument-type declarations

You can declare the types of function arguments by appending ::TypeName to the argument name, as usual for Type Declarations in Julia. For example, the following function computes Fibonacci numbers recursively:

  1. fib(n::Integer) = n 2 ? one(n) : fib(n-1) + fib(n-2)

and the ::Integer specification means that it will only be callable when n is a subtype of the abstract Integer type.

Argument-type declarations normally have no impact on performance: regardless of what argument types (if any) are declared, Julia compiles a specialized version of the function for the actual argument types passed by the caller. For example, calling fib(1) will trigger the compilation of specialized version of fib optimized specifically for Int arguments, which is then re-used if fib(7) or fib(15) are called. (There are rare exceptions when an argument-type declaration can trigger additional compiler specializations; see: Be aware of when Julia avoids specializing.) The most common reasons to declare argument types in Julia are, instead:

  • Dispatch: As explained in Methods, you can have different versions (“methods”) of a function for different argument types, in which case the argument types are used to determine which implementation is called for which arguments. For example, you might implement a completely different algorithm fib(x::Number) = ... that works for any Number type by using Binet’s formula to extend it to non-integer values.
  • Correctness: Type declarations can be useful if your function only returns correct results for certain argument types. For example, if we omitted argument types and wrote fib(n) = n ≤ 2 ? one(n) : fib(n-1) + fib(n-2), then fib(1.5) would silently give us the nonsensical answer 1.0.
  • Clarity: Type declarations can serve as a form of documentation about the expected arguments.

However, it is a common mistake to overly restrict the argument types, which can unnecessarily limit the applicability of the function and prevent it from being re-used in circumstances you did not anticipate. For example, the fib(n::Integer) function above works equally well for Int arguments (machine integers) and BigInt arbitrary-precision integers (see BigFloats and BigInts), which is especially useful because Fibonacci numbers grow exponentially rapidly and will quickly overflow any fixed-precision type like Int (see Overflow behavior). If we had declared our function as fib(n::Int), however, the application to BigInt would have been prevented for no reason. In general, you should use the most general applicable abstract types for arguments, and when in doubt, omit the argument types. You can always add argument-type specifications later if they become necessary, and you don’t sacrifice performance or functionality by omitting them.

The return Keyword

The value returned by a function is the value of the last expression evaluated, which, by default, is the last expression in the body of the function definition. In the example function, f, from the previous section this is the value of the expression x + y. As an alternative, as in many other languages, the return keyword causes a function to return immediately, providing an expression whose value is returned:

  1. function g(x,y)
  2. return x * y
  3. x + y
  4. end

Since function definitions can be entered into interactive sessions, it is easy to compare these definitions:

  1. julia> f(x,y) = x + y
  2. f (generic function with 1 method)
  3. julia> function g(x,y)
  4. return x * y
  5. x + y
  6. end
  7. g (generic function with 1 method)
  8. julia> f(2,3)
  9. 5
  10. julia> g(2,3)
  11. 6

Of course, in a purely linear function body like g, the usage of return is pointless since the expression x + y is never evaluated and we could simply make x * y the last expression in the function and omit the return. In conjunction with other control flow, however, return is of real use. Here, for example, is a function that computes the hypotenuse length of a right triangle with sides of length x and y, avoiding overflow:

  1. julia> function hypot(x,y)
  2. x = abs(x)
  3. y = abs(y)
  4. if x > y
  5. r = y/x
  6. return x*sqrt(1+r*r)
  7. end
  8. if y == 0
  9. return zero(x)
  10. end
  11. r = x/y
  12. return y*sqrt(1+r*r)
  13. end
  14. hypot (generic function with 1 method)
  15. julia> hypot(3, 4)
  16. 5.0

There are three possible points of return from this function, returning the values of three different expressions, depending on the values of x and y. The return on the last line could be omitted since it is the last expression.

Return type

A return type can be specified in the function declaration using the :: operator. This converts the return value to the specified type.

  1. julia> function g(x, y)::Int8
  2. return x * y
  3. end;
  4. julia> typeof(g(1, 2))
  5. Int8

This function will always return an Int8 regardless of the types of x and y. See Type Declarations for more on return types.

Return type declarations are rarely used in Julia: in general, you should instead write “type-stable” functions in which Julia’s compiler can automatically infer the return type. For more information, see the Performance Tips chapter.

Returning nothing

For functions that do not need to return a value (functions used only for some side effects), the Julia convention is to return the value nothing:

  1. function printx(x)
  2. println("x = $x")
  3. return nothing
  4. end

This is a convention in the sense that nothing is not a Julia keyword but only a singleton object of type Nothing. Also, you may notice that the printx function example above is contrived, because println already returns nothing, so that the return line is redundant.

There are two possible shortened forms for the return nothing expression. On the one hand, the return keyword implicitly returns nothing, so it can be used alone. On the other hand, since functions implicitly return their last expression evaluated, nothing can be used alone when it’s the last expression. The preference for the expression return nothing as opposed to return or nothing alone is a matter of coding style.

Operators Are Functions

In Julia, most operators are just functions with support for special syntax. (The exceptions are operators with special evaluation semantics like && and ||. These operators cannot be functions since Short-Circuit Evaluation requires that their operands are not evaluated before evaluation of the operator.) Accordingly, you can also apply them using parenthesized argument lists, just as you would any other function:

  1. julia> 1 + 2 + 3
  2. 6
  3. julia> +(1,2,3)
  4. 6

The infix form is exactly equivalent to the function application form – in fact the former is parsed to produce the function call internally. This also means that you can assign and pass around operators such as + and * just like you would with other function values:

  1. julia> f = +;
  2. julia> f(1,2,3)
  3. 6

Under the name f, the function does not support infix notation, however.

Operators With Special Names

A few special expressions correspond to calls to functions with non-obvious names. These are:

ExpressionCalls
[A B C …]hcat
[A; B; C; …]vcat
[A B; C D; …]hvcat
A’adjoint
A[i]getindex
A[i] = xsetindex!
A.ngetproperty
A.n = xsetproperty!

Anonymous Functions

Functions in Julia are first-class objects: they can be assigned to variables, and called using the standard function call syntax from the variable they have been assigned to. They can be used as arguments, and they can be returned as values. They can also be created anonymously, without being given a name, using either of these syntaxes:

  1. julia> x -> x^2 + 2x - 1
  2. #1 (generic function with 1 method)
  3. julia> function (x)
  4. x^2 + 2x - 1
  5. end
  6. #3 (generic function with 1 method)

This creates a function taking one argument x and returning the value of the polynomial x^2 + 2x - 1 at that value. Notice that the result is a generic function, but with a compiler-generated name based on consecutive numbering.

The primary use for anonymous functions is passing them to functions which take other functions as arguments. A classic example is map, which applies a function to each value of an array and returns a new array containing the resulting values:

  1. julia> map(round, [1.2, 3.5, 1.7])
  2. 3-element Vector{Float64}:
  3. 1.0
  4. 4.0
  5. 2.0

This is fine if a named function effecting the transform already exists to pass as the first argument to map. Often, however, a ready-to-use, named function does not exist. In these situations, the anonymous function construct allows easy creation of a single-use function object without needing a name:

  1. julia> map(x -> x^2 + 2x - 1, [1, 3, -1])
  2. 3-element Vector{Int64}:
  3. 2
  4. 14
  5. -2

An anonymous function accepting multiple arguments can be written using the syntax (x,y,z)->2x+y-z. A zero-argument anonymous function is written as ()->3. The idea of a function with no arguments may seem strange, but is useful for “delaying” a computation. In this usage, a block of code is wrapped in a zero-argument function, which is later invoked by calling it as f.

As an example, consider this call to get:

  1. get(dict, key) do
  2. # default value calculated here
  3. time()
  4. end

The code above is equivalent to calling get with an anonymous function containing the code enclosed between do and end, like so:

  1. get(()->time(), dict, key)

The call to time is delayed by wrapping it in a 0-argument anonymous function that is called only when the requested key is absent from dict.

Tuples

Julia has a built-in data structure called a tuple that is closely related to function arguments and return values. A tuple is a fixed-length container that can hold any values, but cannot be modified (it is immutable). Tuples are constructed with commas and parentheses, and can be accessed via indexing:

  1. julia> (1, 1+1)
  2. (1, 2)
  3. julia> (1,)
  4. (1,)
  5. julia> x = (0.0, "hello", 6*7)
  6. (0.0, "hello", 42)
  7. julia> x[2]
  8. "hello"

Notice that a length-1 tuple must be written with a comma, (1,), since (1) would just be a parenthesized value. () represents the empty (length-0) tuple.

Named Tuples

The components of tuples can optionally be named, in which case a named tuple is constructed:

  1. julia> x = (a=2, b=1+2)
  2. (a = 2, b = 3)
  3. julia> x[1]
  4. 2
  5. julia> x.a
  6. 2

Named tuples are very similar to tuples, except that fields can additionally be accessed by name using dot syntax (x.a) in addition to the regular indexing syntax (x[1]).

Destructuring Assignment and Multiple Return Values

A comma-separated list of variables (optionally wrapped in parentheses) can appear on the left side of an assignment: the value on the right side is destructured by iterating over and assigning to each variable in turn:

  1. julia> (a,b,c) = 1:3
  2. 1:3
  3. julia> b
  4. 2

The value on the right should be an iterator (see Iteration interface) at least as long as the number of variables on the left (any excess elements of the iterator are ignored).

This can be used to return multiple values from functions by returning a tuple or other iterable value. For example, the following function returns two values:

  1. julia> function foo(a,b)
  2. a+b, a*b
  3. end
  4. foo (generic function with 1 method)

If you call it in an interactive session without assigning the return value anywhere, you will see the tuple returned:

  1. julia> foo(2,3)
  2. (5, 6)

Destructuring assignment extracts each value into a variable:

  1. julia> x, y = foo(2,3)
  2. (5, 6)
  3. julia> x
  4. 5
  5. julia> y
  6. 6

Another common use is for swapping variables:

  1. julia> y, x = x, y
  2. (5, 6)
  3. julia> x
  4. 6
  5. julia> y
  6. 5

If only a subset of the elements of the iterator are required, a common convention is to assign ignored elements to a variable consisting of only underscores _ (which is an otherwise invalid variable name, see Allowed Variable Names):

  1. julia> _, _, _, d = 1:10
  2. 1:10
  3. julia> d
  4. 4

Other valid left-hand side expressions can be used as elements of the assignment list, which will call setindex! or setproperty!, or recursively destructure individual elements of the iterator:

  1. julia> X = zeros(3);
  2. julia> X[1], (a,b) = (1, (2, 3))
  3. (1, (2, 3))
  4. julia> X
  5. 3-element Vector{Float64}:
  6. 1.0
  7. 0.0
  8. 0.0
  9. julia> a
  10. 2
  11. julia> b
  12. 3

Julia 1.6

... with assignment requires Julia 1.6

If the last symbol in the assignment list is suffixed by ... (known as slurping), then it will be assigned a collection or lazy iterator of the remaining elements of the right-hand side iterator:

  1. julia> a, b... = "hello"
  2. "hello"
  3. julia> a
  4. 'h': ASCII/Unicode U+0068 (category Ll: Letter, lowercase)
  5. julia> b
  6. "ello"
  7. julia> a, b... = Iterators.map(abs2, 1:4)
  8. Base.Generator{UnitRange{Int64}, typeof(abs2)}(abs2, 1:4)
  9. julia> a
  10. 1
  11. julia> b
  12. Base.Iterators.Rest{Base.Generator{UnitRange{Int64}, typeof(abs2)}, Int64}(Base.Generator{UnitRange{Int64}, typeof(abs2)}(abs2, 1:4), 1)

See Base.rest for details on the precise handling and customization for specific iterators.

Property destructuring

Instead of destructuring based on iteration, the right side of assignments can also be destructured using property names. This follows the syntax for NamedTuples, and works by assigning to each variable on the left a property of the right side of the assignment with the same name using getproperty:

  1. julia> (; b, a) = (a=1, b=2, c=3)
  2. (a = 1, b = 2, c = 3)
  3. julia> a
  4. 1
  5. julia> b
  6. 2

Argument destructuring

The destructuring feature can also be used within a function argument. If a function argument name is written as a tuple (e.g. (x, y)) instead of just a symbol, then an assignment (x, y) = argument will be inserted for you:

  1. julia> minmax(x, y) = (y < x) ? (y, x) : (x, y)
  2. julia> gap((min, max)) = max - min
  3. julia> gap(minmax(10, 2))
  4. 8

Notice the extra set of parentheses in the definition of gap. Without those, gap would be a two-argument function, and this example would not work.

Similarly, property destructuring can also be used for function arguments:

  1. julia> foo((; x, y)) = x + y
  2. foo (generic function with 1 method)
  3. julia> foo((x=1, y=2))
  4. 3
  5. julia> struct A
  6. x
  7. y
  8. end
  9. julia> foo(A(3, 4))
  10. 7

For anonymous functions, destructuring a single argument requires an extra comma:

  1. julia> map(((x,y),) -> x + y, [(1,2), (3,4)])
  2. 2-element Array{Int64,1}:
  3. 3
  4. 7

Varargs Functions

It is often convenient to be able to write functions taking an arbitrary number of arguments. Such functions are traditionally known as “varargs” functions, which is short for “variable number of arguments”. You can define a varargs function by following the last positional argument with an ellipsis:

  1. julia> bar(a,b,x...) = (a,b,x)
  2. bar (generic function with 1 method)

The variables a and b are bound to the first two argument values as usual, and the variable x is bound to an iterable collection of the zero or more values passed to bar after its first two arguments:

  1. julia> bar(1,2)
  2. (1, 2, ())
  3. julia> bar(1,2,3)
  4. (1, 2, (3,))
  5. julia> bar(1, 2, 3, 4)
  6. (1, 2, (3, 4))
  7. julia> bar(1,2,3,4,5,6)
  8. (1, 2, (3, 4, 5, 6))

In all these cases, x is bound to a tuple of the trailing values passed to bar.

It is possible to constrain the number of values passed as a variable argument; this will be discussed later in Parametrically-constrained Varargs methods.

On the flip side, it is often handy to “splat” the values contained in an iterable collection into a function call as individual arguments. To do this, one also uses ... but in the function call instead:

  1. julia> x = (3, 4)
  2. (3, 4)
  3. julia> bar(1,2,x...)
  4. (1, 2, (3, 4))

In this case a tuple of values is spliced into a varargs call precisely where the variable number of arguments go. This need not be the case, however:

  1. julia> x = (2, 3, 4)
  2. (2, 3, 4)
  3. julia> bar(1,x...)
  4. (1, 2, (3, 4))
  5. julia> x = (1, 2, 3, 4)
  6. (1, 2, 3, 4)
  7. julia> bar(x...)
  8. (1, 2, (3, 4))

Furthermore, the iterable object splatted into a function call need not be a tuple:

  1. julia> x = [3,4]
  2. 2-element Vector{Int64}:
  3. 3
  4. 4
  5. julia> bar(1,2,x...)
  6. (1, 2, (3, 4))
  7. julia> x = [1,2,3,4]
  8. 4-element Vector{Int64}:
  9. 1
  10. 2
  11. 3
  12. 4
  13. julia> bar(x...)
  14. (1, 2, (3, 4))

Also, the function that arguments are splatted into need not be a varargs function (although it often is):

  1. julia> baz(a,b) = a + b;
  2. julia> args = [1,2]
  3. 2-element Vector{Int64}:
  4. 1
  5. 2
  6. julia> baz(args...)
  7. 3
  8. julia> args = [1,2,3]
  9. 3-element Vector{Int64}:
  10. 1
  11. 2
  12. 3
  13. julia> baz(args...)
  14. ERROR: MethodError: no method matching baz(::Int64, ::Int64, ::Int64)
  15. Closest candidates are:
  16. baz(::Any, ::Any) at none:1

As you can see, if the wrong number of elements are in the splatted container, then the function call will fail, just as it would if too many arguments were given explicitly.

Optional Arguments

It is often possible to provide sensible default values for function arguments. This can save users from having to pass every argument on every call. For example, the function Date(y, [m, d]) from Dates module constructs a Date type for a given year y, month m and day d. However, m and d arguments are optional and their default value is 1. This behavior can be expressed concisely as:

  1. function Date(y::Int64, m::Int64=1, d::Int64=1)
  2. err = validargs(Date, y, m, d)
  3. err === nothing || throw(err)
  4. return Date(UTD(totaldays(y, m, d)))
  5. end

Observe, that this definition calls another method of the Date function that takes one argument of type UTInstant{Day}.

With this definition, the function can be called with either one, two or three arguments, and 1 is automatically passed when only one or two of the arguments are specified:

  1. julia> using Dates
  2. julia> Date(2000, 12, 12)
  3. 2000-12-12
  4. julia> Date(2000, 12)
  5. 2000-12-01
  6. julia> Date(2000)
  7. 2000-01-01

Optional arguments are actually just a convenient syntax for writing multiple method definitions with different numbers of arguments (see Note on Optional and keyword Arguments). This can be checked for our Date function example by calling methods function.

Keyword Arguments

Some functions need a large number of arguments, or have a large number of behaviors. Remembering how to call such functions can be difficult. Keyword arguments can make these complex interfaces easier to use and extend by allowing arguments to be identified by name instead of only by position.

For example, consider a function plot that plots a line. This function might have many options, for controlling line style, width, color, and so on. If it accepts keyword arguments, a possible call might look like plot(x, y, width=2), where we have chosen to specify only line width. Notice that this serves two purposes. The call is easier to read, since we can label an argument with its meaning. It also becomes possible to pass any subset of a large number of arguments, in any order.

Functions with keyword arguments are defined using a semicolon in the signature:

  1. function plot(x, y; style="solid", width=1, color="black")
  2. ###
  3. end

When the function is called, the semicolon is optional: one can either call plot(x, y, width=2) or plot(x, y; width=2), but the former style is more common. An explicit semicolon is required only for passing varargs or computed keywords as described below.

Keyword argument default values are evaluated only when necessary (when a corresponding keyword argument is not passed), and in left-to-right order. Therefore default expressions may refer to prior keyword arguments.

The types of keyword arguments can be made explicit as follows:

  1. function f(;x::Int=1)
  2. ###
  3. end

Keyword arguments can also be used in varargs functions:

  1. function plot(x...; style="solid")
  2. ###
  3. end

Extra keyword arguments can be collected using ..., as in varargs functions:

  1. function f(x; y=0, kwargs...)
  2. ###
  3. end

Inside f, kwargs will be an immutable key-value iterator over a named tuple. Named tuples (as well as dictionaries with keys of Symbol) can be passed as keyword arguments using a semicolon in a call, e.g. f(x, z=1; kwargs...).

If a keyword argument is not assigned a default value in the method definition, then it is required: an UndefKeywordError exception will be thrown if the caller does not assign it a value:

  1. function f(x; y)
  2. ###
  3. end
  4. f(3, y=5) # ok, y is assigned
  5. f(3) # throws UndefKeywordError(:y)

One can also pass key => value expressions after a semicolon. For example, plot(x, y; :width => 2) is equivalent to plot(x, y, width=2). This is useful in situations where the keyword name is computed at runtime.

When a bare identifier or dot expression occurs after a semicolon, the keyword argument name is implied by the identifier or field name. For example plot(x, y; width) is equivalent to plot(x, y; width=width) and plot(x, y; options.width) is equivalent to plot(x, y; width=options.width).

The nature of keyword arguments makes it possible to specify the same argument more than once. For example, in the call plot(x, y; options..., width=2) it is possible that the options structure also contains a value for width. In such a case the rightmost occurrence takes precedence; in this example, width is certain to have the value 2. However, explicitly specifying the same keyword argument multiple times, for example plot(x, y, width=2, width=3), is not allowed and results in a syntax error.

Evaluation Scope of Default Values

When optional and keyword argument default expressions are evaluated, only previous arguments are in scope. For example, given this definition:

  1. function f(x, a=b, b=1)
  2. ###
  3. end

the b in a=b refers to a b in an outer scope, not the subsequent argument b.

Do-Block Syntax for Function Arguments

Passing functions as arguments to other functions is a powerful technique, but the syntax for it is not always convenient. Such calls are especially awkward to write when the function argument requires multiple lines. As an example, consider calling map on a function with several cases:

  1. map(x->begin
  2. if x < 0 && iseven(x)
  3. return 0
  4. elseif x == 0
  5. return 1
  6. else
  7. return x
  8. end
  9. end,
  10. [A, B, C])

Julia provides a reserved word do for rewriting this code more clearly:

  1. map([A, B, C]) do x
  2. if x < 0 && iseven(x)
  3. return 0
  4. elseif x == 0
  5. return 1
  6. else
  7. return x
  8. end
  9. end

The do x syntax creates an anonymous function with argument x and passes it as the first argument to map. Similarly, do a,b would create a two-argument anonymous function. Note that do (a,b) would create a one-argument anonymous function, whose argument is a tuple to be deconstructed. A plain do would declare that what follows is an anonymous function of the form () -> ....

How these arguments are initialized depends on the “outer” function; here, map will sequentially set x to A, B, C, calling the anonymous function on each, just as would happen in the syntax map(func, [A, B, C]).

This syntax makes it easier to use functions to effectively extend the language, since calls look like normal code blocks. There are many possible uses quite different from map, such as managing system state. For example, there is a version of open that runs code ensuring that the opened file is eventually closed:

  1. open("outfile", "w") do io
  2. write(io, data)
  3. end

This is accomplished by the following definition:

  1. function open(f::Function, args...)
  2. io = open(args...)
  3. try
  4. f(io)
  5. finally
  6. close(io)
  7. end
  8. end

Here, open first opens the file for writing and then passes the resulting output stream to the anonymous function you defined in the do ... end block. After your function exits, open will make sure that the stream is properly closed, regardless of whether your function exited normally or threw an exception. (The try/finally construct will be described in Control Flow.)

With the do block syntax, it helps to check the documentation or implementation to know how the arguments of the user function are initialized.

A do block, like any other inner function, can “capture” variables from its enclosing scope. For example, the variable data in the above example of open...do is captured from the outer scope. Captured variables can create performance challenges as discussed in performance tips.

Function composition and piping

Functions in Julia can be combined by composing or piping (chaining) them together.

Function composition is when you combine functions together and apply the resulting composition to arguments. You use the function composition operator () to compose the functions, so (f ∘ g)(args...) is the same as f(g(args...)).

You can type the composition operator at the REPL and suitably-configured editors using \circ<tab>.

For example, the sqrt and + functions can be composed like this:

  1. julia> (sqrt +)(3, 6)
  2. 3.0

This adds the numbers first, then finds the square root of the result.

The next example composes three functions and maps the result over an array of strings:

  1. julia> map(first reverse uppercase, split("you can compose functions like this"))
  2. 6-element Vector{Char}:
  3. 'U': ASCII/Unicode U+0055 (category Lu: Letter, uppercase)
  4. 'N': ASCII/Unicode U+004E (category Lu: Letter, uppercase)
  5. 'E': ASCII/Unicode U+0045 (category Lu: Letter, uppercase)
  6. 'S': ASCII/Unicode U+0053 (category Lu: Letter, uppercase)
  7. 'E': ASCII/Unicode U+0045 (category Lu: Letter, uppercase)
  8. 'S': ASCII/Unicode U+0053 (category Lu: Letter, uppercase)

Function chaining (sometimes called “piping” or “using a pipe” to send data to a subsequent function) is when you apply a function to the previous function’s output:

  1. julia> 1:10 |> sum |> sqrt
  2. 7.416198487095663

Here, the total produced by sum is passed to the sqrt function. The equivalent composition would be:

  1. julia> (sqrt sum)(1:10)
  2. 7.416198487095663

The pipe operator can also be used with broadcasting, as .|>, to provide a useful combination of the chaining/piping and dot vectorization syntax (described below).

  1. julia> ["a", "list", "of", "strings"] .|> [uppercase, reverse, titlecase, length]
  2. 4-element Vector{Any}:
  3. "A"
  4. "tsil"
  5. "Of"
  6. 7

When combining pipes with anonymous functions, parentheses must be used if subsequent pipes are not to parsed as part of the anonymous function’s body. Compare:

  1. julia> 1:3 .|> (x -> x^2) |> sum |> sqrt
  2. 3.7416573867739413
  3. julia> 1:3 .|> x -> x^2 |> sum |> sqrt
  4. 3-element Vector{Float64}:
  5. 1.0
  6. 2.0
  7. 3.0

Dot Syntax for Vectorizing Functions

In technical-computing languages, it is common to have “vectorized” versions of functions, which simply apply a given function f(x) to each element of an array A to yield a new array via f(A). This kind of syntax is convenient for data processing, but in other languages vectorization is also often required for performance: if loops are slow, the “vectorized” version of a function can call fast library code written in a low-level language. In Julia, vectorized functions are not required for performance, and indeed it is often beneficial to write your own loops (see Performance Tips), but they can still be convenient. Therefore, any Julia function f can be applied elementwise to any array (or other collection) with the syntax f.(A). For example, sin can be applied to all elements in the vector A like so:

  1. julia> A = [1.0, 2.0, 3.0]
  2. 3-element Vector{Float64}:
  3. 1.0
  4. 2.0
  5. 3.0
  6. julia> sin.(A)
  7. 3-element Vector{Float64}:
  8. 0.8414709848078965
  9. 0.9092974268256817
  10. 0.1411200080598672

Of course, you can omit the dot if you write a specialized “vector” method of f, e.g. via f(A::AbstractArray) = map(f, A), and this is just as efficient as f.(A). The advantage of the f.(A) syntax is that which functions are vectorizable need not be decided upon in advance by the library writer.

More generally, f.(args...) is actually equivalent to broadcast(f, args...), which allows you to operate on multiple arrays (even of different shapes), or a mix of arrays and scalars (see Broadcasting). For example, if you have f(x,y) = 3x + 4y, then f.(pi,A) will return a new array consisting of f(pi,a) for each a in A, and f.(vector1,vector2) will return a new vector consisting of f(vector1[i],vector2[i]) for each index i (throwing an exception if the vectors have different length).

  1. julia> f(x,y) = 3x + 4y;
  2. julia> A = [1.0, 2.0, 3.0];
  3. julia> B = [4.0, 5.0, 6.0];
  4. julia> f.(pi, A)
  5. 3-element Vector{Float64}:
  6. 13.42477796076938
  7. 17.42477796076938
  8. 21.42477796076938
  9. julia> f.(A, B)
  10. 3-element Vector{Float64}:
  11. 19.0
  12. 26.0
  13. 33.0

Keyword arguments are not broadcasted over, but are simply passed through to each call of the function. For example, round.(x, digits=3) is equivalent to broadcast(x -> round(x, digits=3), x).

Moreover, nested f.(args...) calls are fused into a single broadcast loop. For example, sin.(cos.(X)) is equivalent to broadcast(x -> sin(cos(x)), X), similar to [sin(cos(x)) for x in X]: there is only a single loop over X, and a single array is allocated for the result. [In contrast, sin(cos(X)) in a typical “vectorized” language would first allocate one temporary array for tmp=cos(X), and then compute sin(tmp) in a separate loop, allocating a second array.] This loop fusion is not a compiler optimization that may or may not occur, it is a syntactic guarantee whenever nested f.(args...) calls are encountered. Technically, the fusion stops as soon as a “non-dot” function call is encountered; for example, in sin.(sort(cos.(X))) the sin and cos loops cannot be merged because of the intervening sort function.

Finally, the maximum efficiency is typically achieved when the output array of a vectorized operation is pre-allocated, so that repeated calls do not allocate new arrays over and over again for the results (see Pre-allocating outputs). A convenient syntax for this is X .= ..., which is equivalent to broadcast!(identity, X, ...) except that, as above, the broadcast! loop is fused with any nested “dot” calls. For example, X .= sin.(Y) is equivalent to broadcast!(sin, X, Y), overwriting X with sin.(Y) in-place. If the left-hand side is an array-indexing expression, e.g. X[begin+1:end] .= sin.(Y), then it translates to broadcast! on a view, e.g. broadcast!(sin, view(X, firstindex(X)+1:lastindex(X)), Y), so that the left-hand side is updated in-place.

Since adding dots to many operations and function calls in an expression can be tedious and lead to code that is difficult to read, the macro @. is provided to convert every function call, operation, and assignment in an expression into the “dotted” version.

  1. julia> Y = [1.0, 2.0, 3.0, 4.0];
  2. julia> X = similar(Y); # pre-allocate output array
  3. julia> @. X = sin(cos(Y)) # equivalent to X .= sin.(cos.(Y))
  4. 4-element Vector{Float64}:
  5. 0.5143952585235492
  6. -0.4042391538522658
  7. -0.8360218615377305
  8. -0.6080830096407656

Binary (or unary) operators like .+ are handled with the same mechanism: they are equivalent to broadcast calls and are fused with other nested “dot” calls. X .+= Y etcetera is equivalent to X .= X .+ Y and results in a fused in-place assignment; see also dot operators.

You can also combine dot operations with function chaining using |>, as in this example:

  1. julia> [1:5;] .|> [x->x^2, inv, x->2*x, -, isodd]
  2. 5-element Vector{Real}:
  3. 1
  4. 0.5
  5. 6
  6. -4
  7. true

Further Reading

We should mention here that this is far from a complete picture of defining functions. Julia has a sophisticated type system and allows multiple dispatch on argument types. None of the examples given here provide any type annotations on their arguments, meaning that they are applicable to all types of arguments. The type system is described in Types and defining a function in terms of methods chosen by multiple dispatch on run-time argument types is described in Methods.