Type safety with native JavaScript

Why everybody wants type safety and how to get it.

Typecast example

1. Overview

Many developer tools like IDEs, frameworks, libraries, and linters try to provide some level of type safety to JavaScript. This article explains what type safety is, why we want it, and how we can get it using native JavaScript.

The good news is most type errors in JavaScript can be resolved using typecasting. In this article we demonstrate how to typecast using simple functions, and then we show how typecasting works with other best practices to improve productivity and quality.

2. What is type safety?

Type safety is the extent a programming language discourages or prevents type errors. A type error occurs when an unintended or incompatible type value is provided to a function or expression, usually as an argument or context. Dividing a number by an array is a perfect example. Usually this doesn’t make sense, yet JavaScript support this very operation, often with surprising and unwanted results.

Let’s see how easy it is to create a type error in Javascript using the semantic style all the cool kids are using these days. This is awful code, so please don’t copy it. We’ll improve it as we go along.

  function repeats(counts, run)
  {
      while(counts < 0)
      {
          run(counts)
          counts+=1
      }
  }

  function reports(info)
  {
      console.log(info)
  }

  repeats('-3', reports)

That didn’t take long! The first type error occurs when we provide the ‘-3’ argument to the repeats invocation. This value should be an integer, but we instead provide a string. Then all hell breaks loose.

When repeats is invoked it creates an ‘endless’ loop condition that will rather quickly consume all available resources within its execution environment. This results in a force-kill of a NodeJS process, or a browser tab, or the entire browser process, or the host OS.

If we watch the progression of the value of counts in the while loop we see the following series: '-3', '-31', '-311', '-3111', .... The test condition counts < 0 does convert the counts string into a number, but by then it is too late: the value is always less than 0 and it only gets more negative with each iteration. That’s because the expression counts+=1 always converts the number 1 into a string '1' and appends it to the counts variable.

3. Why do we want type safety?

We want type safety with JavaScript because type errors can be quite troublesome:

  1. They are easy to create
  2. They are challenging to resolve
  3. They are often serious

Let’s look at each of these issues in turn.

3.1. JS type errors are easy to create

JavaScript is missing almost all type controls because it was not originally intended to run large-scale applications. This makes creating type errors exceedingly easy.

3.1.1 No variable declarations

JavaScript has no means to formally declare a variable type. A variable may contain any type which may be changed at any time during execution. There are no sigils to identify type, and a variable name is unconstrained. We probably all have seen a JavaScript application or two where a single symbol like watches is used as an object, a map, a list, a boolean flag, a string, an integer, and a floating point number all in different contexts. Yet in practice most variables used in JavaScript are not intended to change type within a context because doing so is needlessly confusing.

Other languages provide a ‘wild card’ variables to reference any type of value. These are usually easy to identify because they have a unique syntax. Developers know they these ‘wild cards’ need to be handled with care. In JavaScript, all variables are ‘wild cards.’

3.1.2 Coercion and polymorphism

JavaScript expressions often intermingle type coercion and polymorphic operators which result in a large number of behaviors that provide surprising and undesirable results.

Type coercion occurs when the language “automatically” converts one type to another. Polymorphic operators are symbols like the plus sign (+) which in JavaScript either concatenates strings or adds two numbers or converts a string to a number.

Let’s look at a few examples. Those playing along at home might want to cover the right-hand side of the table below to see if you can guess what the returned value will be. How’d you do?

  var x, y;
  // expression   // |     returned value | Coercion       | Op
  x = 3;          // |                 3  | -              |
  x = 3 + 1;      // |                 4  | -              | +num
  x = 3 + '1';    // |               '31' | 3      => '3'  | +str
  x = 3 + [];     // |                '3' | []     => ''   | +str
  x = 3 + [ 21 ]; // |              '321' | [ 21 ] => '21' | +str
  x = 3 + {};     // | '3[object Object]' | {}     => str  | +str
  x = '3' - 2;    // |                 1  | '3'    =>  3   | -num
  x = '3' + 2;    // |               '32' | 2      => '2'  | +str
  x = + '3';      // |                 3  | '3'    =>  3   | cast_num
  x = 0 + '3';    // |               '03' | 0      => '0'  | +str
  x = y + 3;      // |                NaN | -              | NaN
  x = y + '3';    // |    'undefined3'    | undef  => str  | +str

Remember that there at least 4 major JavaScript engines on the market (V8, IonMonkey, Nitro, Chakra) and at least another dozen minor players. We wouldn’t be surprised to find that the behaviors show here (using V8) vary from engine to engine and generation to generation.

Other languages have less complex behaviors because they usually have stricter type checking, stricter coercion rules, fewer polymorphic operators, and fewer vendors. Perl, for example, uses the dot (.) operator to join strings and the plus (+) operator to add values. Perl also uses sigals (prefixes) like $, @, and % to denote variable types and a special syntax to identify ‘wild-card’ references.

3.1.3 No static checking

Many languages provide some level of static type checking. Java, for example, resolves most variable types during compilation. If JavaScript had a similar mechanism we wouldn’t be able to run our application until we resolved the compile errors. In this imaginary world, our JavaScript compile output might look like this:

00: ok                           | x = 3;          |
01: ok                           | x = 3 + 1;      |
02: compile_error: type_mismatch | x = 3 + '1';    |
03: compile_error: type_mismatch | x = 3 + [];     |
04: compile_error: type_mismatch | x = 3 + [ 21 ]; |
06: compile_error: type_mismatch | x = 3 + {};     |
07: compile_error: type_mismatch | x = '3' - 2;    |
08: compile_error: type_mismatch | x = '3' + 2;    |
09: compile_error: type_mismatch | x = + '3';      |
10: compile_error: type_mismatch | x = 0 + '3';    |
11: compile_error: type_mismatch | x = y + 3;      |
12: compile_error: type_mismatch | x = y + '3';    |

Perhaps the greatest advantage of static (compile-time) type checking is that it can improve performance: every type check that can be resolved once during a compile removes a type check that would need to be invoked on every call of a function or method. This can remove a large number of calls when the application is run and thus improves performance.

3.1.4 Roll-your-own dynamic checking

Static type checking does not work in all situations, especially when dealing with data from unknown or untrusted sources. In those cases the application must resort to dynamic type checking at run-time. With native JavaScript, run-time checks are pretty much our only option, yet the language only has limited tools to help here. Even the typeof built-in function fails to distinguish between an object and an array.

3.2. Type errors are challenging to resolve

Type errors can be hard to identify and debug. When one routine fails to check for type an incorrect result can propagate up the call stack resulting in a cascade of errors. The originating flaw can be hard to spot if variable aren’t named to indicate their intended type, like so:

  // Cool-kid, no-consistency convention:
  var total = watches / in_use;

However, if we name our variable by intended type the mismatches become obvious:

  // Real-developer naming convention:
  var total_str = watch_list / use_bool;

Yes, the intended variable type is that important. We’ve had to maintain plenty of third-party modules which used the cool-kid, no-consistency convention where variable names provide no hint of type or purpose, or worse, are patently misleading. For example, what genius decided to call a boolean flag item_array? We’d rather name our variables sensibly and use the time saved to focus on new challenges.

3.3. Type errors are often serious

As we have shown, type errors can result in severe application failures and security holes. Imagine some NodeJS code that doesn’t properly type-check its JSON API. One could implement a Denial Of Service (DOS) attack and shut down an entire web farm by simply sending strings instead of numbers in API requests. This stuff happens.

4. How do we get type safety?

Getting type safety in native JavaScript isn’t particularly hard. First recognize that the values we need to worry about are inputs to public methods like arguments, external data, and context. We can use typecasting to guarantee these value types.

4.1 Typecasting

Typecasting, for the purposes of this article, is the process of converting a value into the desired data type using a very strict set of rules. Our typecasting functions either return the requested value type or a failure value which is undefined by default. Consider the examples below:

  // return_data = castInt( <value-to-cast> [, <failure-value>] );

  return_data = castInt( 0      ); // 0
  return_data = castInt( '0'    ); // 0
  return_data = castInt( 'a'    ); // undefined
  return_data = castInt( []     ); // undefined
  return_data = castInt( 'a', 0 ); // 0
  return_data = castInt( [],  0 ); // 0

castInt:

4.1.1 Use typecasting

Let’s adjust our example function to use castInt and castFn to ensure the provided arguments are the correct type. If they are not, the function will return without any further processing:

  function repeats(arg_counts, arg_run)
  {
      var counts = castInt(arg_counts)
      var run = castFn(arg_run)
      if (!(counts && run)){return}

      while (counts < 0)
      {
          run(counts)
          counts+=1
      }
  }

  function reports(info)
  {
      console.log(info)
  }

  repeats('-3', reports)

The function is now impervious to most type errors.

4.1.2 Get typecast methods

We can get typecast methods from the hi_score project which is easy to install (type npm install hi_score into a terminal). If you edit the example application you can use all the cast methods from xhi.util.js.

  npm install hi_score
  cd npm_modules/hi_score
  npm install
  npm run setup
  npm test
  google-chrome ./index.html
  # Open the JavaScript console to access xhi._util_ functions.

You don’t have to use the whole library; you can just crib the methods from xhi.util.js if you want. Go ahead, you won’t hurt anyone’s feelings.

4.1.2 More typecasting

The xhi utility library provides the following typecast methods:

  castBool, castFn,  castInt, castJQ,
  castList, castMap, castNum, castObj, castStr

All these methods take either one or two arguments. Only numbers, strings, and integers are converted between types and only when the conversion is unambiguous.

The first argument is always the value to cast; the second argument is the value to use if the cast fails. If a second argument is omitted, undefined is returned instead. Let’s update our example again to use an argument map and a failure value for the counts variable:

  function repeats(arg_map) {
      var map = castMap(arg_map, {})
      var counts = castInt(map.counts, 0)
      var run = castFn(map.run)

      if (run)
      {
          while (counts < 0)
          {
              run(counts)
              counts+=1
          }
      }
  }

  function reports (info)
  {
      console.log(info)
  }
  repeats({counts:'-3',run:reports});

This is run-time (dynamic) type checking. There is no native static checking in JavaScript in any real sense (one could make arguments about the JIT compiler, but at best such checking is very incomplete). One could use a type-safety tool like Flow or a framework like TypeScript which in theory can perform static type analysis. However, it is not clear that this provides a performance benefit worth the compile overhead. We will compare native typecasting with these solutions in an upcoming article.

4.1.3 The Zen of typecasting

The cast methods have three benefits over native JavaScript coercion: they are predictable, explicit, and self-documenting. Let’s compare pedictability first:

4.1.3.1 JavaScript type coercion examples

JavaScript tries hard to coerce types, and the results are often undesirable. Blank cells are conditions where a type error exception is usually thrown.

Value Bool Fn Num Ary Obj Str
’’ FALSE   0     ’’
‘0’ t   0     ‘0’
‘1’ t   1     ‘1’
‘20’ t   20     ‘20’
‘ten’ t   NaN     ‘ten’
0 FALSE   0     ‘0’
1 t   1     ‘1’
NaN FALSE   NaN     ‘NaN’
[] t   0 [] {} ’’
[‘ten’] t   NaN [‘ten’] { 0: ‘ten’ } ‘ten’
[‘ten’,’t’] t   NaN [‘ten’,’t’] { 0: ‘ten’, 1: ‘t’} ‘ten,t’
[10] t   10 [10] { 0 : 10 } ‘10’
[10,20] t   NaN [10,20] { 0: 10, 1:20 } ‘10,20’
false FALSE   0     ‘false’
function(){} t function(){} NaN [] {} ‘function(){}’
null FALSE   0     ‘null’
true t   1     ‘true’
undefined FALSE   NaN     ‘undefined’
{} t   NaN [] {} ‘0’
-Infinity t -Infinity       ‘-Infinity’
Infinity t Infinity       ‘Infinity’

4.1.4.1 hi_score cast method examples

The cast methods are very strict, and only convert the most unambigous values. Blank cells are conditions where the failure value (undefined by default) will be returned. No exceptions are thrown by these methods.

Value Bool Fn Num Ary Obj Str
’’           ’’
‘0’     0     ‘0’
‘1’     1     ‘1’
‘20’     20     ‘20’
‘ten’           ‘ten’
0     0     ‘0’
1     1     ‘1’
NaN            
[]       []    
[‘ten’]       [‘ten]    
[‘ten’,’t’]       [‘ten’,’t’]    
[10]       [10]    
[10,20]       [10,20]    
false FALSE          
function(){}   function(){}        
null            
true true          
undefined            
{}         {}  
-Infinity     -Infinity      
Infinity     Infinity      

The cast methods are explicit and visible calls unlike JavaScript’s native type coercion, and they are named with the intent to illustrate exactly what is being accomplished without the need for comments. We hope that it is obvious, for example, that the castFn call is intended validate a function.

4.2 Best practices with typecasting

Typecasting works best when we adopt a few additional good habits:

  1. Name variables to indicate type
  2. Write consistent API definition
  3. Test the APIs

Let’s look at each of these and update our sample code as we go along.

4.3 Name variables to indicate type

Let’s improve our code again by changing the variable names. While they might make sense to one brain, they are inconsistent and misleading to another. We’ve purposely used plural variable names to illustrate how bad a practice that can be. We use our handy JS Code standard cheat sheet to light the way:

  function repeatFn ( arg_map ) {
    var
      map = castMap( arg_map,  {} ),
      int = castInt( map._int_, 0 ),
      fn  = castFn(  map._fn_     ),
      idx
      ;

    if ( ! fn ) { return; }

    for ( idx = int; idx < 0; idx++ ) {
      fn( idx );
    }
  }

  function printToConsole ( idx ) { console.log( idx ); }

  repeatFn({ _int_ : '-3', _fn_ : printToConsole });

We employed additional best practices suggested by the guide such as formatting (spacing, alignment, K&R indenting) and replaced the dangerous while loop with a for loop. This code will now pass JSLint. Thanks to the naming convention we can tell that tell that fn should be a function and and idx should be an integer regardless if any other code is visible. Think about how much time can be saved by this alone.

The full JS Code standard discusses why a simple naming convention can vastly reduce the need for comments. We think it’s an interesting and compelling read if you’re into that kind of thing.

4.4. Write consistent API definitions

Now that we have consistent named-by-type variables and better formatting, we can easily read the code to create in-line API documentation. Using the guide from the code standard we get the following:

  // BEGIN utility method /repeatFn/
  // Summary   : repeatFn({ _int_ : <integer>, _fn_ : <function> )
  // Purpose   : Repeatedly call a function 'fn' as long as the
  //             counter 'int' is < 0.  After each call, 'int' is
  //             incremented by 1.  If the initial value of 'int'
  //             is not < 0 the function 'fn' is not called.
  // Example   : repeatFn({
  //               _int_ : -3,
  //               _fn_ : function (idx ) { console.log( idx ) }
  //             });
  // Arguments  : ( named )
  //   _fn_     : The function to execute. The current value of the
  //              index (idx) is provided as its sole argument.
  //   _int_    : The initial value of idx. Idx is incremented after
  //              _fn_ is executed. Thus a value of '-1' will result in a
  //              single execution of _fn_( idx );
  // Returns    : undefined
  // Throws     : none
  //
  function repeatFn ( arg_map ) {
    var
      map = castMap( arg_map,  {} ),
      int = castInt( map._int_, 0 ),
      fn  = castFn(  map._fn_     ),
      idx
      ;

    if ( ! fn ) { return; }
    for ( idx = int; idx < 0; idx++ ) {
      fn( idx );
    }
  }
  // END utility method /repeatFn/

  function printToConsole ( idx ) { console.log( idx ); }
  repeatFn({ _int_ : '-3', _fn_ : printToConsole });

Remember where we begged you not to copy our first code example? Now the code is impervious to most type errors, readable, testable, maintainable, and well documented. If you insist on copying some code, we recommend this version!

4.5 Test the Code

We can use tools like Istanbul and nodeunit along with the in-line API documentation to test our our shiny new type-safe repeatFn function. Check out the test suite for hi_score to see how this is done:

  npm install hi_score
  cd npm_modules/hi_score
  npm install
  npm run setup
  npm test
  npm run coverage
  google-chrome index.html
  google-chrome coverage/lcov-info/index.html
  vi test/xhi_level_0.js # Or preferred editor

5. What about frameworks and libraries?

An excellent question is how does this native typecasting technique compare with Flow and TypeScript? We intend to publish a sequel that will answer this question. In theory both these transpiled languages should be able to provide static type checking and eliminate many dynamic type check calls. However, in practice the results may be surprising since the real-world overhead of cast methods can be actually quite low, and these languages introduce their own overhead.

One thing to remember is don’t go typcasting crazy. Use it when processing external data and inputs of public methods. Private methods often don’t benefit from typecasting since the the callers and data types are already known, at least if we name our variables to indicate type.

We hope you found this useful! Please share your thoughts and experiences in the comments below.

Cheers, Mike

Written on November 14, 2016