Scope of Variables

Each module introduces a new global scope, separate from the global scope of all other modules—​there is no all-encompassing global scope. Modules can introduce variables of other modules into their scope through the using or import statements or through qualified access using the dot-notation, i.e. each module is a so-called namespace as well as a first-class data structure associating names with values. Note that while variable bindings can be read externally, they can only be changed within the module to which they belong. As an escape hatch, you can always evaluate code inside that module to modify a variable; this guarantees, in particular, that module bindings cannot be modified externally by code that never calls eval.

If a top-level expression contains a variable declaration with keyword local, then that variable is not accessible outside that expression. The variable inside the expression does not affect global variables of the same name. An example is to declare local x in a begin or if block at the top-level:

Note that the interactive prompt (aka REPL) is in the global scope of the module Main.

A new local scope is introduced by most code blocks (see above table for a complete list). If such a block is syntactically nested inside of another local scope, the scope it creates is nested inside of all the local scopes that it appears within, which are all ultimately nested inside of the global scope of the module in which the code is evaluated. Variables in outer scopes are visible from any scope they contain — meaning that they can be read and written in inner scopes — unless there is a local variable with the same name that "shadows" the outer variable of the same name. This is true even if the outer local is declared after (in the sense of textually below) an inner block. When we say that a variable "exists" in a given scope, this means that a variable by that name exists in any of the scopes that the current scope is nested inside of, including the current one.

Some programming languages require explicitly declaring new variables before using them. Explicit declaration works in Julia too: in any local scope, writing local x declares a new local variable in that scope, regardless of whether there is already a variable named x in an outer scope or not. Declaring each new variable like this is somewhat verbose and tedious, however, so Julia, like many other languages, considers assignment to a variable name that doesn’t already exist to implicitly declare that variable. If the current scope is global, the new variable is global; if the current scope is local, the new variable is local to the innermost local scope and will be visible inside of that scope but not outside of it. If you assign to an existing local, it always updates that existing local: you can only shadow a local by explicitly declaring a new local in a nested scope with the local keyword. In particular, this applies to variables assigned in inner functions, which may surprise users coming from Python where assignment in an inner function creates a new local unless the variable is explicitly declared to be non-local.

Mostly this is pretty intuitive, but as with many things that behave intuitively, the details are more subtle than one might naïvely imagine.

When x = <value> occurs in a local scope, Julia applies the following rules to decide what the expression means based on where the assignment expression occurs and what x already refers to at that location:

You may note that in non-interactive contexts the hard and soft scope behaviors are identical except that a warning is printed when an implicitly local variable (i.e. not declared with local x) shadows a global. In interactive contexts, the rules follow a more complex heuristic for the sake of convenience. This is covered in depth in examples that follow.

Now that you know the rules, let’s look at some examples. Each example is assumed to be evaluated in a fresh REPL session so that the only globals in each snippet are the ones that are assigned in that block of code.

We’ll begin with a nice and clear-cut situation—​assignment inside of a hard scope, in this case a function body, when no local variable by that name already exists:

Inside of the greet function, the assignment x = "hello" causes x to be a new local variable in the function’s scope. There are two relevant facts: the assignment occurs in local scope and there is no existing local x variable. Since x is local, it doesn’t matter if there is a global named x or not. Here for example we define x = 123 before defining and calling greet:

Since the x in greet is local, the value (or lack thereof) of the global x is unaffected by calling greet. The hard scope rule doesn’t care whether a global named x exists or not: assignment to x in a hard scope is local (unless x is declared global).

The next clear cut situation we’ll consider is when there is already a local variable named x, in which case x = <value> always assigns to this existing local x. This is true whether the assignment occurs in the same local scope, an inner local scope in the same function body, or in the body of a function nested inside of another function, also known as a closure.

We’ll use the sum_to function, which computes the sum of integers from one up to n, as an example:

As in the previous example, the first assignment to s at the top of sum_to causes s to be a new local variable in the body of the function. The for loop has its own inner local scope within the function scope. At the point where s = s + i occurs, s is already a local variable, so the assignment updates the existing s instead of creating a new local. We can test this out by calling sum_to in the REPL:

Since s is local to the function sum_to, calling the function has no effect on the global variable s. We can also see that the update s = s + i in the for loop must have updated the same s created by the initialization s = 0 since we get the correct sum of 55 for the integers 1 through 10.

Let’s dig into the fact that the for loop body has its own scope for a second by writing a slightly more verbose variation which we’ll call sum_to_def, in which we save the sum s + i in a variable t before updating s:

This version returns s as before but it also uses the @isdefined macro to return a boolean indicating whether there is a local variable named t defined in the function’s outermost local scope. As you can see, there is no t defined outside of the for loop body. This is because of the hard scope rule again: since the assignment to t occurs inside of a function, which introduces a hard scope, the assignment causes t to become a new local variable in the local scope where it appears, i.e. inside of the loop body. Even if there were a global named t, it would make no difference—​the hard scope rule isn’t affected by anything in global scope.

Note that the local scope of a for loop body is no different from the local scope of an inner function. This means that we could rewrite this example so that the loop body is implemented as a call to an inner helper function and it behaves the same way:

This example illustrates a couple of key points:

This design means that you can generally move code in or out of an inner function without changing its meaning, which facilitates a number of common idioms in the language using closures (see do blocks).

Let’s move onto some more ambiguous cases covered by the soft scope rule. We’ll explore this by extracting the bodies of the greet and sum_to_def functions into soft scope contexts. First, let’s put the body of greet in a for loop—​which is soft, rather than hard—​and evaluate it in the REPL:

Since the global x is not defined when the for loop is evaluated, the first clause of the soft scope rule applies and x is created as local to the for loop and therefore global x remains undefined after the loop executes. Next, let’s consider the body of sum_to_def extracted into global scope, fixing its argument to n = 10

What does this code do? Hint: it’s a trick question. The answer is "it depends." If this code is entered interactively, it behaves the same way it does in a function body. But if the code appears in a file, it prints an ambiguity warning and throws an undefined variable error. Let’s see it working in the REPL first:

The REPL approximates being in the body of a function by deciding whether assignment inside the loop assigns to a global or creates new local based on whether a global variable by that name is defined or not. If a global by the name exists, then the assignment updates it. If no global exists, then the assignment creates a new local variable. In this example we see both cases in action:

The second fact is why execution of the loop changes the global value of s and the first fact is why t is still undefined after the loop executes. Now, let’s try evaluating this same code as though it were in a file instead:

Here we use include_string, to evaluate code as though it were the contents of a file. We could also save code to a file and then call include on that file—​the result would be the same. As you can see, this behaves quite different from evaluating the same code in the REPL. Let’s break down what’s happening here:

This demonstrates some important aspects of scope: in a scope, each variable can only have one meaning, and that meaning is determined regardless of the order of expressions. The presence of the expression s = t in the loop causes s to be local to the loop, which means that it is also local when it appears on the right hand side of t = s + i, even though that expression appears first and is evaluated first. One might imagine that the s on the first line of the loop could be global while the s on the second line of the loop is local, but that’s not possible since the two lines are in the same scope block and each variable can only mean one thing in a given scope.

We have now covered all the local scope rules, but before wrapping up this section, perhaps a few words should be said about why the ambiguous soft scope case is handled differently in interactive and non-interactive contexts. There are two obvious questions one could ask:

In Julia ≤ 0.6, all global scopes did work like the current REPL: when x = <value> occurred in a loop (or try/catch, or struct body) but outside of a function body (or let block or comprehension), it was decided based on whether a global named x was defined or not whether x should be local to the loop. This behavior has the advantage of being intuitive and convenient since it approximates the behavior inside of a function body as closely as possible. In particular, it makes it easy to move code back and forth between a function body and the REPL when trying to debug the behavior of a function. However, it has some downsides. First, it’s quite a complex behavior: many people over the years were confused about this behavior and complained that it was complicated and hard both to explain and understand. Fair point. Second, and arguably worse, is that it’s bad for programming "at scale." When you see a small piece of code in one place like this, it’s quite clear what’s going on:

Obviously the intention is to modify the existing global variable s. What else could it mean? However, not all real world code is so short or so clear. We found that code like the following often occurs in the wild:

It’s far less clear what should happen here. Since x + "hello" is a method error, it seems probable that the intention is for x to be local to the for loop. But runtime values and what methods happen to exist cannot be used to determine the scopes of variables. With the Julia ≤ 0.6 behavior, it’s especially concerning that someone might have written the for loop first, had it working just fine, but later when someone else adds a new global far away—​possibly in a different file—​the code suddenly changes meaning and either breaks noisily or, worse still, silently does the wrong thing. This kind of "spooky action at a distance" is something that good programming language designs should prevent.

So in Julia 1.0, we simplified the rules for scope: in any local scope, assignment to a name that wasn’t already a local variable created a new local variable. This eliminated the notion of soft scope entirely as well as removing the potential for spooky action. We uncovered and fixed a significant number of bugs due to the removal of soft scope, vindicating the choice to get rid of it. And there was much rejoicing! Well, no, not really. Because some people were angry that they now had to write:

Do you see that global annotation in there? Hideous. Obviously this situation could not be tolerated. But seriously, there are two main issues with requiring global for this kind of top-level code:

As of Julia 1.5, this code works without the global annotation in interactive contexts like the REPL or Jupyter notebooks (just like Julia 0.6) and in files and other non-interactive contexts, it prints this very direct warning:

This addresses both issues while preserving the "programming at scale" benefits of the 1.0 behavior: global variables have no spooky effect on the meaning of code that may be far away; in the REPL copy-and-paste debugging works and beginners don’t have any issues; any time someone either forgets a global annotation or accidentally shadows an existing global with a local in a soft scope, which would be confusing anyway, they get a nice clear warning.

An important property of this design is that any code that executes in a file without a warning will behave the same way in a fresh REPL. And on the flip side, if you take a REPL session and save it to file, if it behaves differently than it did in the REPL, then you will get a warning.

A common use of variables is giving names to specific, unchanging values. Such variables are only assigned once. This intent can be conveyed to the compiler using the const keyword:

Multiple variables can be declared in a single const statement:

The const declaration should only be used in global scope on globals. It is difficult for the compiler to optimize code involving global variables, since their values (or even their types) might change at almost any time. If a global variable will not change, adding a const declaration solves this performance problem.

Local constants are quite different. The compiler is able to determine automatically when a local variable is constant, so local constant declarations are not necessary, and in fact are currently not supported.

Special top-level assignments, such as those performed by the function and struct keywords, are constant by default.

Note that const only affects the variable binding; the variable may be bound to a mutable object (such as an array), and that object may still be modified. Additionally when one tries to assign a value to a variable that is declared constant the following scenarios are possible:

The last rule applies to immutable objects even if the variable binding would change, e.g.:

However, for mutable objects the warning is printed as expected:

Note that although sometimes possible, changing the value of a const variable is strongly discouraged, and is intended only for convenience during interactive use. Changing constants can cause various problems or unexpected behaviors. For instance, if a method references a constant and is already compiled before the constant is changed, then it might keep using the old value:

Similar to being declared as constants, global bindings can also be declared to always be of a constant type. This can either be done without assigning an actual value using the syntax global x::T or upon assignment as x::T = 123.

For any assignment to a global, Julia will first try to convert it to the appropriate type using convert:

The type does not need to be concrete, but annotations with abstract types typically have little performance benefit.

Once a global has either been assigned to or its type has been set, the binding type is not allowed to change: