Error handling, "try...catch"

No matter how great we are at programming, sometimes our scripts have errors. They may occur because of our mistakes, an unexpected user input, an erroneous server response, and for a thousand other reasons. Usually, a script “dies” (immediately stops) in case of an error, printing it to console. But there’s a syntax construct try…catch that allows us to “catch” errors so the script can, instead of dying, do something more reasonable.

The “try…catch” syntax

The try…catch construct has two main blocks: try, and then catch: It works like this: 1. First, the code in try {…} is executed. 2. If there were no errors, then catch (err) is ignored: the execution reaches the end of try and goes on, skipping catch. 3. If an error occurs, then the try execution is stopped, and control flows to the beginning of catch (err). The err variable (we can use any name for it) will contain an error object with details about what happened. So, an error inside the try {…} block does not kill the script – we have a chance to handle it in catch. Let’s look at some examples. - An errorless example: shows alert (1) and (2):

try {
alert('Start of try runs');  // !(1) <--/!
// ...no errors here
alert('End of try runs');   // !(2) <--/!
} catch (err) {
alert('Catch is ignored, because there are no errors'); // (3)

• An example with an error: shows (1) and (3):

try {
alert('Start of try runs');  // !(1) <--/!
lalala; // error, variable is not defined!
alert('End of try (never reached)');  // (2)
} catch (err) {
alert(Error has occurred!); // !(3) <--/!

try { {{{{{{{{{{{{ } catch (err) { alert(“The engine can’t understand this code, it’s invalid”); try { setTimeout(function() { noSuchVariable; // script will die here }, 1000); } catch (err) { alert( “won’t work” ); setTimeout(function() { try { noSuchVariable; // try…catch handles the error! } catch { alert( “error is caught here!” ); }, 1000);

Error object

name

Error name. For instance, for an undefined variable that’s “ReferenceError”. message

Textual message about error details. There are other non-standard properties available in most environments. One of most widely used and supported is: stack

Current call stack: a string with information about the sequence of nested calls that led to the error. Used for debugging purposes. For instance:

Optional “catch” binding

[recent browser=new] If we don’t need error details, catch may omit it:

Using “try…catch”

Let’s explore a real-life use case of try…catch. As we already know, JavaScript supports the JSON.parse(str) method to read JSON-encoded values. Usually it’s used to decode data received over the network, from the server or another source. We receive it and call JSON.parse like this: You can find more detailed information about JSON in the info:json chapter. If json is malformed, JSON.parse generates an error, so the script “dies”. Should we be satisfied with that? Of course not! This way, if something’s wrong with the data, the visitor will never know that (unless they open the developer console). And people really don’t like when something “just dies” without any error message. Let’s use try…catch to handle the error: Here we use the catch block only to show the message, but we can do much more: send a new network request, suggest an alternative to the visitor, send information about the error to a logging facility, … . All much better than just dying.

Throwing our own errors

What if json is syntactically correct, but doesn’t have a required name property? Like this: Here JSON.parse runs normally, but the absence of name is actually an error for us. To unify error handling, we’ll use the throw operator.

“Throw” operator

The throw operator generates an error. The syntax is: Technically, we can use anything as an error object. That may be even a primitive, like a number or a string, but it’s better to use objects, preferably with name and message properties (to stay somewhat compatible with built-in errors). JavaScript has many built-in constructors for standard errors: Error, SyntaxError, ReferenceError, TypeError and others. We can use them to create error objects as well. Their syntax is: For built-in errors (not for any objects, just for errors), the name property is exactly the name of the constructor. And message is taken from the argument. For instance: Let’s see what kind of error JSON.parse generates: As we can see, that’s a SyntaxError. And in our case, the absence of name is an error, as users must have a name. So let’s throw it: In the line (*), the throw operator generates a SyntaxError with the given message, the same way as JavaScript would generate it itself. The execution of try immediately stops and the control flow jumps into catch. Now catch became a single place for all error handling: both for JSON.parse and other cases.

Rethrowing

In the example above we use try…catch to handle incorrect data. But is it possible that another unexpected error occurs within the try {…} block? Like a programming error (variable is not defined) or something else, not just this “incorrect data” thing. For example: Of course, everything’s possible! Programmers do make mistakes. Even in open-source utilities used by millions for decades – suddenly a bug may be discovered that leads to terrible hacks. In our case, try…catch is placed to catch “incorrect data” errors. But by its nature, catch gets all errors from try. Here it gets an unexpected error, but still shows the same “JSON Error” message. That’s wrong and also makes the code more difficult to debug. To avoid such problems, we can employ the “rethrowing” technique. The rule is simple: Catch should only process errors that it knows and “rethrow” all others. The “rethrowing” technique can be explained in more detail as:

1. Catch gets all errors.

2. In the catch (err) {…} block we analyze the error object err.

3. If we don’t know how to handle it, we do throw err. Usually, we can check the error type using the instanceof operator: We can also get the error class name from err.name property. All native errors have it. Another option is to read err.constructor.name. In the code below, we use rethrowing so that catch only handles SyntaxError: The error throwing on line (*) from inside catch block “falls out” of try…catch and can be either caught by an outer try…catch construct (if it exists), or it kills the script. So the catch block actually handles only errors that it knows how to deal with and “skips” all others. The example below demonstrates how such errors can be caught by one more level of try…catch: Here readData only knows how to handle SyntaxError, while the outer try…catch knows how to handle everything.

   try…catch…finally

   Wait, that’s not all. The try…catch construct may have one more code clause: finally. If it exists, it runs in all cases:

• after try, if there were no errors,
• after catch, if there were errors. The extended syntax looks like this: Try running this code: The code has two ways of execution:

1. If you answer “Yes” to “Make an error?”, then try -> catch -> finally.

2. If you say “No”, then try -> finally. The finally clause is often used when we start doing something and want to finalize it in any case of outcome. For instance, we want to measure the time that a Fibonacci numbers function fib(n) takes. Naturally, we can start measuring before it runs and finish afterwards. But what if there’s an error during the function call? In particular, the implementation of fib(n) in the code below returns an error for negative or non-integer numbers. The finally clause is a great place to finish the measurements no matter what. Here finally guarantees that the time will be measured correctly in both situations – in case of a successful execution of fib and in case of an error in it: You can check by running the code with entering 35 into prompt – it executes normally, finally after try. And then enter -1 – there will be an immediate error, and the execution will take 0ms. Both measurements are done correctly. In other words, the function may finish with return or throw, that doesn’t matter. The finally clause executes in both cases. function func() { try { return 1; } catch (err) { / … / } finally { alert( ‘finally’ ); alert( func() ); // first works alert from finally, and then this one function func() { // start doing something that needs completion (like measurements) try { // … } finally { // complete that thing even if all dies

   Global catch

   Let’s imagine we’ve got a fatal error outside of try…catch, and the script died. Like a programming error or some other terrible thing. Is there a way to react on such occurrences? We may want to log the error, show something to the user (normally they don’t see error messages), etc. There is none in the specification, but environments usually provide it, because it’s really useful. For instance, Node.js has process.on(“uncaughtException”) for that. And in the browser we can assign a function to the special window.onerror property, that will run in case of an uncaught error. The syntax: message

Error message. url

URL of the script where error happened. line, col

Line and column numbers where error happened. error

Error object. For instance: The role of the global handler window.onerror is usually not to recover the script execution – that’s probably impossible in case of programming errors, but to send the error message to developers. There are also web-services that provide error-logging for such cases, like https://muscula.com or https://www.sentry.io. They work like this:

1. We register at the service and get a piece of JS (or a script URL) from them to insert on pages.

2. That JS script sets a custom window.onerror function.

3. When an error occurs, it sends a network request about it to the service.

4. We can log in to the service web interface and see errors.

   Summary

   The try…catch construct allows to handle runtime errors. It literally allows to “try” running the code and “catch” errors that may occur in it. The syntax is: There may be no catch section or no finally, so shorter constructs try…catch and try…finally are also valid. Error objects have following properties:

• message – the human-readable error message.
• name – the string with error name (error constructor name).
• stack (non-standard, but well-supported) – the stack at the moment of error creation. If an error object is not needed, we can omit it by using catch { instead of catch (err) {. We can also generate our own errors using the throw operator. Technically, the argument of throw can be anything, but usually it’s an error object inheriting from the built-in Error class. More on extending errors in the next chapter. Rethrowing is a very important pattern of error handling: a catch block usually expects and knows how to handle the particular error type, so it should rethrow errors it doesn’t know. Even if we don’t have try…catch, most environments allow us to setup a “global” error handler to catch errors that “fall out”. In-browser, that’s window.onerror.

try {

  // code...

} catch (err) {

  // error handling

}