Yet another tutorial on borrowed pointers

19 July 2012

Here is my latest stab at a tutorial on borrowed pointers. I know, I know, enough with the borrowed pointer tutorials already! Hopefully this will be my last post in this vein for a while. I am much happier with this version. It is still too long to serve as a chapter in the general Rust tutorial, but I think it’s more approachable than the previous attempt, which was more of a reference document. As always, feedback welcome! I have tried to incorporate what people wrote in the comments into this version.

UPDATE: Added discussion on purity.

Borrowed pointers

Introduction

Borrowed pointers are one of the more flexible and powerful tools available in Rust. A borrowed pointer can be used to point anywhere: into the shared and exchange heaps, into the stack, and even into the interior of another data structure. With regard to flexibility, it is comparable to a C pointer or C++ reference. However, unlike C and C++, the Rust compiler includes special checks that ensure that borrowed pointers are being used safely. Another advantage of borrowed pointers is that they are invisible to the garbage collector, so working with borrowed pointers helps keep things efficient.

Despite the fact that they are completely safe, at runtime, a borrowed pointer is “just a pointer”. They introduce zero overhead. All safety checks are done at compilation time.

Although borrowed pointers have rather elaborate theoretical underpinnings (region pointers), the core concepts will be familiar to anyone who worked with C or C++. Therefore, the best way to explain how they are used—and their limitations—is probably just to work through several examples.

By example

Borrowed pointers are called borrowed because they are only valid for a limit duration. Borrowed pointers never claim any kind of ownership over the data that they point at: instead, they are used for cases where you like to make use of data for a short time.

As an example, consider a simple record type point:

type point = {x: float, y: float};

We can use this simple definition to allocate points in many ways. For example, in this code, each of these three local variables contains a point, but allocated in a different place:

let on_the_stack : point  =  {x: 3.0, y: 4.0};
let shared_box   : @point = @{x: 5.0, y: 1.0};
let unique_box   : ~point = ~{x: 7.0, y: 9.0};

Suppose we wanted to write a procedure that computed the distance between any two points, no matter where they were stored. For example, we might like to compute the distance between on_the_stack and shared_box, or between shared_box and unique_box. One option is to define a function that takes two arguments of type point—that is, it takes the points by value. But this will cause the points to be copied when we call the function. For points, this is probably not so bad, but often copies are expensive or, worse, if there are mutable fields, they can change the semantics of your program. So we’d like to define a function that takes the points by pointer. We can use borrowed pointers to do this:

fn compute_distance(p1: &point, p2: &point) -> float {
    let x_d = p1.x - p2.x;
    let y_d = p1.y - p2.y;
    sqrt(x_d * x_d + y_d * y_d)
}

Now we can call compute_distance() in various ways:

compute_distance(&on_the_stack, shared_box)
compute_distance(shared_box, unique_box)

Here the & operator is used to take the address of the variable on_the_stack; this is because on_the_stack has the type point (that is, a record value) and we have to take its address to get a value. We also call this borrowing the local variable on_the_stack, because we are created an alias: that is, another route to the same data.

In the case of the boxes shared_box and unique_box, however, no explicit action is necessary. The compiler will automatically convert a box like @point or ~point to a borrowed pointer like &point. This is another form of borrowing; in this case, the contents of the shared/unique box is being lent out.

Whenever a value is borrowed, there are some limitations on what you can do with the original. For example, if the contents of a variable have been lent out, you cannot send that variable to another task, nor will you be permitted to take actions that might cause the borrowed value to be freed or to change its type (I’ll get into what kinds of actions those are shortly). This rule should make intuitive sense: you must wait for a borrowed value to be returned (that is, for the borrowed pointer to go out of scope) before you can make full use of it again.

Other uses for the & operator

In the previous example, the value on_the_stack was defined like so:

let on_the_stack : point = {x: 3.0, y: 4.0};

This results in a by-value variable. As a consequence, we had to explicitly take the address of on_the_stack to get a borrowed pointer. Sometimes however it is more convenient to move the & operator into the definition of on_the_stack:

let on_the_stack2 : &point = &{x: 3.0, y: 4.0};

Applying & to an rvalue (non-assignable location) is just a convenient shorthand for creating a temporary and taking its address:

let tmp = {x: 3.0, y: 4.0};
let on_the_stack2 : &point = &tmp;

Taking the address of fields

As in C, the & operator is not limited to taking the address of local variables. It can also be used to take the address of fields or individual array elements. For example, consider this type definition for rectangle:

type point = {x: float, y: float}; // as before
type size = {w: float, h: float}; // as before
type rectangle = {origin: point, size: size};

Now again I can define rectangles in a few different ways:

let rect_stack  = &{origin: {x: 1, y: 2}, size: {w: 3, h: 4}};
let rect_shared = @{origin: {x: 3, y: 4}, size: {w: 3, h: 4}};
let rect_unique = ~{origin: {x: 5, y: 6}, size: {w: 3, h: 4}};

In each case I can use the & operator to extact out individual subcomponents. For example, I could write:

compute_distance(&rect_stack.origin, &rect_shared.origin);

which would borrow the field origin from the rectangle on the stack from the shared box and then compute the distance between them.

Borrowing shared boxes and rooting

We’ve seen a few examples so far where heap boxes (both shared and unique) are borrowed. Up till this point, we’ve glossed over issues of safety. As stated in the introduction, at runtime a borrowed pointer is simply a pointer, nothing more. Therefore, if we wish to avoid the issues that C has with dangling pointers (and we do!), a compile-time safety check is required.

The basis for the check is the notion of lifetimes. A lifetime is basically a static approximation of the period in which the pointer is valid: it always corresponds to some expression or block within the program. Within that expression, the pointer can be used freely, but if the pointer somehow leaks outside of that expression, the compiler will report an error. We’ll be discussing lifetimes more in the examples to come, and a more thorough introduction is also available.

When a borrowed pointer is created, the compiler must ensure that it will remain valid for its entire lifetime. Sometimes this is relatively easy, such as when taking the address of a local variable or a field that is stored on the stack:

fn example1() {
    let mut x = {f: 3};
    let y = &mut x.f;  // -+ L
    ...                //  |
}                      // -+

Here, the lifetime of the borrowed pointer is simply L, the remainder of the function body. No extra work is required to ensure that x.f will not be freed. This is true even if x is mutated.

The situation gets more complex when borrowing data that resides in heap boxes:

fn example2() {
    let mut x = @{f: 3};
    let y = &x.f;      // -+ L
    ...                //  |
}                      // -+

In this example, the value x is in fact a heap box, and y is therefore a pointer into that heap box. Again the lifetime of y will be L, the remainder of the function body. But there is a crucial difference: suppose x were reassigned during the lifetime L? If we’re not careful, that could mean that the shared box would become unrooted and therefore be subject to garbage collection (an aside: in our current implementation, the garbage collector is implemented using reference counting and cycle detection).

For this reason, whenever the interior of a shared box stored in a mutable location is borrowed, the compiler will insert a temporary that ensures that the shared box remains live for the entire lifetime. So, the above example would be compiled as:

fn example2() {
    let mut x = @{f: 3};
    let x1 = x;
    let y = &x1.f;     // -+ L
    ...                //  |
}                      // -+

Now if x is reassigned, the pointer y will still remain valid. This process is called “rooting”.

Borrowing unique boxes

The previous example demonstrated rooting, the process by which the compiler ensures that shared boxes remain live for the duration of a borrow. Unfortunately, rooting does not work if the data being borrowed is a unique box, as it is not possible to have two references to a unique box.

For unique boxes, therefore, the compiler will only allow a borrow if the compiler can guarantee that the unique box will not be reassigned or moved for the lifetime of the pointer. This does not necessarily mean that the unique box is stored in immutable memory. For example, the following function is legal:

fn example3() -> int {
    let mut x = ~{f: 3};
    if some_condition {
        let y = &x.f;      // -+ L
        ret *y;            //  |
    }                      // -+
    x = ~{f: 4};
    ...
}

Here, as before, the interior of the variable x is being borrowed and x is declared as mutable. However, the compiler can clearly see that x is not assigned anywhere in the lifetime L of the variable y. Therefore, it accepts the function, even though x is mutable and in fact is mutated later in the function.

It may not be clear why we are so concerned about the variable which was borrowed being mutated. The reason is that unique boxes are freed as soon as their owning reference is changed or goes out of scope. Therefore, a program like this is illegal (and would be rejected by the compiler):

fn example3() -> int {
    let mut x = ~{f: 3};
    let y = &x.f;
    x = ~{f: 4};  // Error reported here.
    *y
}

To make this clearer, consider this diagram showing the state of memory immediately before the re-assignment of x:

    Stack               Exchange Heap

  x +----------+
    | ~{f:int} | ----+
  y +----------+     |
    | &int     | ----+
    +----------+     |    +---------+
                     +--> |  f: 3   |
                          +---------+

Once the reassignment occurs, the memory will look like this:

    Stack               Exchange Heap

  x +----------+          +---------+
    | ~{f:int} | -------> |  f: 4   |
  y +----------+          +---------+
    | &int     | ----+
    +----------+     |    +---------+
                     +--> | (freed) |
                          +---------+

Here you can see that the variable y still points at the old box, which has been freed.

In fact, the compiler can apply this same kind of reasoning can be applied to any memory which is (uniquely) owned by the stack frame. So we could modify the previous example to introduce additional unique pointers and records, and the compiler will still be able to detect possible mutations:

fn example3() -> int {
    let mut x = ~{mut f: ~{g: 3}};
    let y = &x.f.g;
    x = ~{mut f: ~{g: 4}}; // Error reported here.
    x.f = ~{g: 5};         // Error reported here.
    *y
}

In this case, two errors are reported, one when the variable x is modified and another when x.f is modified. Either modification would cause the pointer y to be invalidated.

Things get tricker when the unique box is not uniquely owned by the stack frame (or when the compiler doesn’t know who the owner is). Consider a program like this:

fn example5a(x: @{mut f: ~{g: int}}, ...) -> int {
    let y = &x.f.g;   // Error reported here.
    ...
}

Here the heap looks something like:

     Stack            Shared Heap       Exchange Heap

  x +------+        +-------------+       +------+
    | @... | ---->  | mut f: ~... | --+-> | g: 3 |
  y +------+        +-------------+   |   +------+
    | &int | -------------------------+
    +------+

In this case, the owning reference to the value being borrowed is in fact x.f. Moreover, x.f is both mutable and aliasable. Aliasable means that it is possible that there are other pointers to that same shared box, so even if the compiler were to prevent x.f from being mutated, the field might still be changed through some alias of x. Therefore, to be safe, the compiler only accepts pure actions during the lifetime of y. We’ll have a final example on purity but in short it means “actions that cannot modify heap data”.

Besides ensuring purity, the only way to borrow the interior of a unique found in aliasable memory is to ensure that it is stored within unique fields, as in the following example:

fn example5b(x: @{f: ~{g: int}}, ...) -> int {
    let y = &x.f.g;
    ...
}

Here, the field f is not declared as mutable. But that is enough for the compiler to know that, even if aliases to x exist, the field f cannot be changed and hence the unique box g will remain valid.

If you do have a unique box in a mutable field, and you wish to borrow it, one option is to use the swap operator to bring that unique box onto your stack:

fn example5c(x: @{mut f: ~int}, ...) -> int {
    let mut v = ~0;
    v <-> x.f;         // Swap v and x.f
    let y = &v;
    ...
    x.f <- v;          // Replace x.f
}

Of course, this has the side effect of modifying your shared box for the duration of the borrow, so it works best when you know that you won’t be accessing that same box again.

Borrowing and enums

The previous example showed that borrowing unique boxes found in aliasable, mutable memory is not permitted, so as to prevent pointers into freed memory. There is one other case where the compiler must be very careful to ensure that pointers remain valid: pointers into the interior of an enum.

As an example, let’s look at the following shape type that can represent both rectangles and circles:

type point = {x: float, y: float}; // as before
type size = {w: float, h: float}; // as before
enum shape {
    circle(point, float),   // origin, radius
    rectangle(point, size)  // upper-left, dimensions
}

Now I might write a function to compute the area of a shape. This function takes a borrowed pointer to a shape to avoid the need of copying them.

fn compute_area(shape: &shape) -> float {
    alt *shape {
        circle(_, radius) => 0.5 * tau * radius * radius,
        rectangle(_, ref size) => size.w * size.h
    }
}

The first case matches against circles. Here the radius is extracted from the shape variant and used to compute the area of the circle (Like any up-to-date engineer, we use the tau circle constant and not that dreadfully outdated notion of pi).

The second match is more interesting. Here we match against a rectangle and extract its size: but rather than copy the size struct, we use a by-reference binding to create a pointer to it. In other words, a pattern binding like ref size in fact creates a pointer of type &size into the interior of the enum.

To make this more clear, let’s look at a diagram of how things are laid out in memory in the case where shape points at a rectangle:

Stack             Memory

+-------+         +---------------+
| shape | ------> | rectangle(    |
+-------+         |   {x: float,  |
| size  | -+      |    y: float}, |
+-------+  +----> |   {w: float,  |
                  |    h: float}) |
                  +---------------+

Here you can see that rectangular shapes are composed of five words of memory. The first is a tag indicating which variant this enum is (rectangle, in this case). The next two words are the x and y fields for the point and the remaining two are the w and h fields for the size. The binding size is then a pointer into the inside of the shape.

Perhaps you can see where the danger lies: if the shape were somehow to be reassigned, perhaps to a circle, then although the memory used to store that shape value would still be valid, it would have a different type! This is shown in the following diagram, depicting what the state of memory would be if shape were overwritten with a circle:

Stack             Memory

+-------+         +---------------+
| shape | ------> | circle(       |
+-------+         |   {x: float,  |
| size  | -+      |    y: float}, |
+-------+  +----> |   float)      |
                  |               |
                  +---------------+

As you can see, the size pointer would not be pointing at a float and not a record. This is not good.

So, in fact, for every ref binding, the compiler will impose the same rules as the ones we saw for borrowing the interior of a unique box: it must be able to guarantee that the enum will not be overwritten for the duration of the borrow. In fact, the example I gave earlier would be considered safe. This is because the shape pointer has type &shape, which means “borrowed pointer to immutable memory containing a shape”. If however the type of that pointer were &const shape or &mut shape, then the ref binding would not be permitted. Just as with unique boxes, the compiler will permit ref bindings into data owned by the stack frame even if it is mutable, but otherwise it requires that the data reside in immutable memory.

Note: the ref notation is due to be implemented as part of #2855. Instead, right now, all pattern bindings are by-reference. We expect this to change so that copies are the default and references must be noted explicitly.

Returning borrowed pointers

So far, all of the examples we’ve looked at use borrowed pointers in a “downward” direction. That is, the borrowed pointer is created and then used during the method or code block which created it. In some cases, it is also possible to return borrowed pointers to the caller, but as we’ll see this is more limited.

For example, we could write a subroutine like this:

type point = {x: float, y: float};
fn get_x(p: &point) -> &float { &p.x }

Here, the function get_x() returns a pointer into the structure it was given. You’ll note that both the parameter and the return value are borrowed pointers; this is important. In general, it is only possible to return borrowed pointers if they are derived from a borrowed pointer which was given as input to the procedure.

In the example, get_x() took a borrowed pointer to a point as input. In general, for all borrowed pointers that appear in the signature of a function (such as the parameter and return types), the compiler assigns the same symbolic lifetime L (we will see later that there are ways to differentiate the lifetimes of different parameters if that should be necessary). This means that, from the compiler’s point of view, get_x() takes and returns two pointers with the same lifetime. Now, unlike other lifetimes, this lifetime is a bit abstract: it doesn’t refer to a specific expression within get_x(), but rather to some expression within the caller. This is called a lifetime parameter, because the lifetime L is effectively defined by the caller to get_x(), just as the value for the parameter p is defined by the caller.

In any case, whatever the lifetime L is, the pointer produced by &p.x always has the same lifetime as p itself, as a pointer to a field of a record is valid as long as the record is valid. Therefore, the compiler is satisfied with the function get_x().

To drill in this point, let’s look at a variation on the example, this time one which does not compile:

type point = {x: float, y: float};
fn get_x_sh(p: @point) -> &float {
    &p.x // Error reported here
}

Here, the function get_x_sh() takes a shared box as input and returns a borrowed pointer. As before, the lifetime of the borrowed pointer that will be returned is a parameter (specified by the caller). That means that effectively get_x_sh() is promising to return a borrowed pointer that is valid for as long as the caller would like: this is subtly different from the first example, which promised to return a pointer that was valid for as long as the pointer it was given.

Within get_x_sh(), we see the expression &p.x which takes the address of a field of a shared box. This implies that the compiler must guarantee that, so long as the resulting pointer is valid, the shared box will not be reclaimed by the garbage collector. But recall that get_x_sh() also promised to return a pointer that was valid for as long as the caller wanted it to be. Clearly, get_x_sh() is not in a position to make both of these guarantees; in fact, it cannot guarantee that the pointer will remain valid at all once it returns, as the parameter p may or may not be live in the caller. Therefore, the compiler will report an error here.

In general, if you borrow a shared (or unique) box to create a borrowed pointer, the pointer will only be valid within the function and cannot be returned. Generally, the only way to return borrowed pointers is to take borrowed pointers as input.

Named lifetimes

So far we have always used the notation &T for a borrowed pointer. However, sometimes if a function takes many parameters, it is useful to be able to group those parameters by lifetime. For example, consider this function:

fn select<T>(shape: &shape, threshold: float,
             a: &T, b: &T) -> &T {
    if compute_area(shape) > threshold {a} else {b}
}

This function takes three borrowed pointers. Because of the way that the system works, each will be assigned the same lifetime: the default lifetime parameter. In practice, this means that, in the caller, the lifetime of the returned value will be the intersection of the lifetime of the three region parameters. This may be overloy conservative, as in this example:

                                              // -+ L
fn select_based_on_unit_circle<T>(            //  |-+ B
    threshold: float, a: &T, b: &T) -> &T {   //  | |
                                              //  | |
    let shape = circle({x: 0, y: 0}, 1);      //  | |
    select(&shape, threshold, a, b)           //  | |
}                                             //  |-+
                                              // -+

In this call to select(), the lifetime of the first parameter shape is B, the function body. Both of the second two parameters a and b share the same lifetime, L, which is the lifetime parameter of select_based_on_unit_circle(). The caller will infer the intersection of these three lifetimes as the lifetime of the returned value, and hence the return value of shape() will be assigned a return value of B. This will in turn lead to a compilation error, because select_based_on_unit_circle() is supposed to return a value with the lifetime L.

To address this, we could modify the definition of select() to distinguish the lifetime of the first parameter from the lifetime of the latter two. After all, the first parameter is not being returned. To do so, we make use of the notation &lt/T, which is a borrowed pointer with an explicit lifetime. This effectively creates a second lifetime parameter for the function; named lifetime parameters do not need to be declared, you just use them. Here is how the new select() might look:

fn select<T>(shape: &tmp/shape, threshold: float,
             a: &T, b: &T) -> &T {
    if compute_area(shape) > threshold {a} else {b}
}

Here you can see the lifetime of shape is now being called tmp. The parameters a, b, and the return value all remain with the default lifetime parameter.

You could also write select() using all named lifetime parameters, which might look like:

fn select<T>(shape: &tmp/shape, threshold: float,
             a: &r/T, b: &r/T) -> &r/T {
    if compute_area(shape) > threshold {a} else {b}
}

This is equivalent to the previous definition.

Purity

As mentioned before, the Rust compiler offers a kind of escape hatch that permits borrowing of any data, but only if the actions that occur during the lifetime of the borrow are pure. Pure actions are those which only modify data owned by the current stack frame. The compiler can therefore permit arbitrary pointers into the heap, secure in the knowledge that no pure action will ever cause them to become invalidated (the compiler must still track data on the stack which is borrowed and enforce those rules normally, of course).

Let’s revisit a previous example and show how purity can affect the compiler’s result. Here is example5a(), which borrows the interior of a unique box found in an aliasable, mutable location, only now we’ve replaced the ... with some specific code:

fn example5a(x: @{mut f: ~{g: int}}, ...) -> int {
    let y = &x.f.g;   // Unsafe
    *y + 1        
}

The new code simply returns an incremented version of y. This clearly doesn’t do mutate anything in the heap, so the compiler is satisfied.

But suppose we wanted to pull the increment code into a helper, like this:

fn add_one(x: &int) -> int { *x + 1 }

We can now update example5a() to use add_one():

fn example5a(x: @{mut f: ~{g: int}}, ...) -> int {
    let y = &x.f.g;
    add_one(y)        // Error reported here
}

But now the compiler will report an error again. The reason is that it only considers one function at a time (like most type checkers), and so it does not know that add_one() only takes pure actions. We can help the compiler by labeling add_one() as pure:

pure fn add_one(x: &int) -> int { *x + 1 }

With this change, the modified version of example5a() will again compile.

Conclusion

So there you have it. A (relatively) brief tour of borrowed pointer system. For more details, I refer to the (yet to be written) reference document on borrowed pointers, which will explain the full notation and give more examples.