Handling Panics in Go

Introduction

Errors that a program encounters fall into two broad categories: those the programmer has anticipated and those the programmer has not. The error interface that we have covered in our previous two articles on error handling largely deal with errors that we expect as we are writing Go programs. The error interface even allows us to acknowledge the rare possibility of an error occurring from function calls, so we can respond appropriately in those situations.

Panics fall into the second category of errors, which are unanticipated by the programmer. These unforeseen errors lead a program to spontaneously terminate and exit the running Go program. Common mistakes are often responsible for creating panics. In this tutorial, we’ll examine a few ways that common operations can produce panics in Go, and we’ll also see ways to avoid those panics. We’ll also use defer statements along with the recover function to capture panics before they have a chance to unexpectedly terminate our running Go programs.

Understanding Panics

There are certain operations in Go that automatically return panics and stop the program. Common operations include indexing an array beyond its capacity, performing type assertions, calling methods on nil pointers, incorrectly using mutexes, and attempting to work with closed channels. Most of these situations result from mistakes made while programming that the compiler has no ability to detect while compiling your program.

Since panics include detail that is useful for resolving an issue, developers commonly use panics as an indication that they have made a mistake during a program’s development.

Out of Bounds Panics

When you attempt to access an index beyond the length of a slice or the capacity of an array, the Go runtime will generate a panic.

The following example makes the common mistake of attempting to access the last element of a slice using the length of the slice returned by the len builtin. Try running this code to see why this might produce a panic:

package main

import (
	"fmt"
)

func main() {
	names := []string{
		"lobster",
		"sea urchin",
		"sea cucumber",
	}
	fmt.Println("My favorite sea creature is:", names[len(names)])
}

This will have the following output:

Output
panic: runtime error: index out of range [3] with length 3

goroutine 1 [running]:
main.main()
	/tmp/sandbox879828148/prog.go:13 +0x20

The name of the panic’s output provides a hint: panic: runtime error: index out of range. We created a slice with three sea creatures. We then tried to get the last element of the slice by indexing that slice with the length of the slice using the len builtin function. Remember that slices and arrays are zero-based; so the first element is zero and the last element in this slice is at index 2. Since we try to access the slice at the third index, 3, there is no element in the slice to return because it is beyond the bounds of the slice. The runtime has no option but to terminate and exit since we have asked it to do something impossible. Go also can’t prove during compilation that this code will try to do this, so the compiler cannot catch this.

Notice also that the subsequent code did not run. This is because a panic is an event that completely halts the execution of your Go program. The message produced contains multiple pieces of information helpful for diagnosing the cause of the panic.

Anatomy of a Panic

Panics are composed of a message indicating the cause of the panic and a stack trace that helps you locate where in your code the panic was produced.

The first part of any panic is the message. It will always begin with the string panic:, which will be followed with a string that varies depending on the cause of the panic. The panic from the previous exercise has the message:

panic: runtime error: index out of range [3] with length 3

The string runtime error: following the panic: prefix tells us that the panic was generated by the language runtime. This panic is telling us that we attempted to use an index [3] that was out of range of the slice’s length 3.

Following this message is the stack trace. Stack traces form a map that we can follow to locate exactly what line of code was executing when the panic was generated, and how that code was invoked by earlier code.

goroutine 1 [running]:
main.main()
	/tmp/sandbox879828148/prog.go:13 +0x20

This stack trace, from the previous example, shows that our program generated the panic from the file /tmp/sandbox879828148/prog.go at line number 13. It also tells us that this panic was generated in the main() function from the main package.

The stack trace is broken into separate blocks—one for each goroutine in your program. Every Go program’s execution is accomplished by one or more goroutines that can each independently and simultaneously execute parts of your Go code. Each block begins with the header goroutine X [state]:. The header gives the ID number of the goroutine along with the state that it was in when the panic occurred. After the header, the stack trace shows the function that the program was executing when the panic happened, along with the filename and line number where the function executed.

The panic in the previous example was generated by an out-of-bounds access to a slice. Panics can also be generated when methods are called on pointers that are unset.

Nil Receivers

The Go programming language has pointers to refer to a specific instance of some type existing in the computer’s memory at runtime. Pointers can assume the value nil indicating that they are not pointing at anything. When we attempt to call methods on a pointer that is nil, the Go runtime will generate a panic. Similarly, variables that are interface types will also produce panics when methods are called on them. To see the panics generated in these cases, try the following example:

package main

import (
	"fmt"
)

type Shark struct {
	Name string
}

func (s *Shark) SayHello() {
	fmt.Println("Hi! My name is", s.Name)
}

func main() {
	s := &Shark{"Sammy"}
	s = nil
	s.SayHello()
}

The panics produced will look like this:

Output
panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0xffffffff addr=0x0 pc=0xdfeba]

goroutine 1 [running]:
main.(*Shark).SayHello(...)
	/tmp/sandbox160713813/prog.go:12
main.main()
	/tmp/sandbox160713813/prog.go:18 +0x1a

In this example, we defined a struct called Shark. Shark has one method defined on its pointer receiver called SayHello that will print a greeting to standard out when called. Within the body of our main function, we create a new instance of this Shark struct and request a pointer to it using the & operator. This pointer is assigned to the s variable. We then reassign the s variable to the value nil with the statement s = nil. Finally we attempt to call the SayHello method on the variable s. Instead of receiving a friendly message from Sammy, we receive a panic that we have attempted to access an invalid memory address. Because the s variable is nil, when the SayHello function is called, it tries to access the field Name on the *Shark type. Because this is a pointer receiver, and the receiver in this case is nil, it panics because it can’t dereference a nil pointer.

While we have set s to nil explicitly in this example, in practice this happens less obviously. When you see panics involving nil pointer dereference, be sure that you have properly assigned any pointer variables that you may have created.

Panics generated from nil pointers and out-of-bounds accesses are two commonly occurring panics generated by the runtime. It is also possible to manually generate a panic using a builtin function.

Using the panic Builtin Function

We can also generate panics of our own using the panic built-in function. It takes a single string as an argument, which is the message the panic will produce. Typically this message is less verbose than rewriting our code to return an error. Furthermore, we can use this within our own packages to indicate to developers that they may have made a mistake when using our package’s code. Whenever possible, best practice is to try to return error values to consumers of our package.

Run this code to see a panic generated from a function called from another function:

package main

func main() {
	foo()
}

func foo() {
	panic("oh no!")
}

The panic output produced looks like:

Output
panic: oh no!

goroutine 1 [running]:
main.foo(...)
	/tmp/sandbox494710869/prog.go:8
main.main()
	/tmp/sandbox494710869/prog.go:4 +0x40

Here we define a function foo that calls the panic builtin with the string "oh no!". This function is called by our main function. Notice how the output has the message panic: oh no! and the stack trace shows a single goroutine with two lines in the stack trace: one for the main() function and one for our foo() function.

We’ve seen that panics appear to terminate our program where they are generated. This can create problems when there are open resources that need to be properly closed. Go provides a mechanism to execute some code always, even in the presence of a panic.

Deferred Functions

Your program may have resources that it must clean up properly, even while a panic is being processed by the runtime. Go allows you to defer the execution of a function call until its calling function has completed execution. Deferred functions run even in the presence of a panic, and are used as a safety mechanism to guard against the chaotic nature of panics. Functions are deferred by calling them as usual, then prefixing the entire statement with the defer keyword, as in defer sayHello(). Run this example to see how a message can be printed even though a panic was produced:

package main

import "fmt"

func main() {
	defer func() {
		fmt.Println("hello from the deferred function!")
	}()

	panic("oh no!")
}

The output produced from this example will look like:

Output
hello from the deferred function!
panic: oh no!

goroutine 1 [running]:
main.main()
	/Users/gopherguides/learn/src/github.com/gopherguides/learn//handle-panics/src/main.go:10 +0x55

Within the main function of this example, we first defer a call to an anonymous function that prints the message "hello from the deferred function!". The main function then immediately produces a panic using the panic function. In the output from this program, we first see that the deferred function is executed and prints its message. Following this is the panic we generated in main.

Deferred functions provide protection against the surprising nature of panics. Within deferred functions, Go also provides us the opportunity to stop a panic from terminating our Go program using another built-in function.

Handling Panics

Panics have a single recovery mechanism—the recover builtin function. This function allows you to intercept a panic on its way up through the call stack and prevent it from unexpectedly terminating your program. It has strict rules for its use, but can be invaluable in a production application.

Since it is part of the builtin package, recover can be called without importing any additional packages:

package main

import (
	"fmt"
	"log"
)

func main() {
	divideByZero()
	fmt.Println("we survived dividing by zero!")

}

func divideByZero() {
	defer func() {
		if err := recover(); err != nil {
			log.Println("panic occurred:", err)
		}
	}()
	fmt.Println(divide(1, 0))
}

func divide(a, b int) int {
	return a / b
}

This example will output:

Output
2009/11/10 23:00:00 panic occurred: runtime error: integer divide by zero
we survived dividing by zero!

Our main function in this example calls a function we define, divideByZero. Within this function, we defer a call to an anonymous function responsible for dealing with any panics that may arise while executing divideByZero. Within this deferred anonymous function, we call the recover builtin function and assign the error it returns to a variable. If divideByZero is panicking, this error value will be set, otherwise it will be nil. By comparing the err variable against nil, we can detect if a panic occurred, and in this case we log the panic using the log.Println function, as though it were any other error.

Following this deferred anonymous function, we call another function that we defined, divide, and attempt to print its results using fmt.Println. The arguments provided will cause divide to perform a division by zero, which will produce a panic.

In the output to this example, we first see the log message from the anonymous function that recovers the panic, followed by the message we survived dividing by zero!. We have indeed done this, thanks to the recover builtin function stopping an otherwise catastrophic panic that would terminate our Go program.

The err value returned from recover() is exactly the value that was provided to the call to panic(). It’s therefore critical to ensure that the err value is only nil when a panic has not occurred.

Detecting Panics with recover

The recover function relies on the value of the error to make determinations as to whether a panic occurred or not. Since the argument to the panic function is an empty interface, it can be any type. The zero value for any interface type, including the empty interface, is nil. Care must be taken to avoid nil as an argument to panic as demonstrated by this example:

package main

import (
	"fmt"
	"log"
)

func main() {
	divideByZero()
	fmt.Println("we survived dividing by zero!")

}

func divideByZero() {
	defer func() {
		if err := recover(); err != nil {
			log.Println("panic occurred:", err)
		}
	}()
	fmt.Println(divide(1, 0))
}

func divide(a, b int) int {
	if b == 0 {
		panic(nil)
	}
	return a / b
}

This will output:

Output
we survived dividing by zero!

This example is the same as the previous example involving recover with some slight modifications. The divide function has been altered to check if its divisor, b, is equal to 0. If it is, it will generate a panic using the panic builtin with an argument of nil. The output, this time, does not include the log message showing that a panic occurred even though one was created by divide. This silent behavior is why it is very important to ensure that the argument to the panic builtin function is not nil.

Conclusion

We have seen a number of ways that panics can be created in Go and how they can be recovered from using the recover builtin. While you may not necessarily use panic yourself, proper recovery from panics is an important step of making Go applications production-ready.

You can also explore our entire How To Code in Go series.