Category: JavaScript

Microvium has Classes!

Microvium has Classes!

TL;DR: Microvium now has support for classes. Here are some details for the curious.


I’ve been putting off adding support for classes in Microvium because you can do a lot without them, and other features have been more important in my opinion. I’m also personally more of a fan of functional-style programming rather than object-orientated programming, so I seldom write classes myself.

But I kept running into little things that are difficult to do without support for classes. With the recent addition of try-catch, for example, you could now throw, but you couldn’t throw new Error(...) because Error is a class. Another example is the ECMA-419 API for embedded systems, which is an API standard for things like GPIO, but which relies heavily on support for classes.

Microvium now has support for classes, which I think makes it the only JavaScript engine under 100kB of flash that has all the features required to implement ECMA-419 (and Microvium is currently under 10kB at the moment).

Microvium functions are not objects

Classes in JavaScript are more or less just syntactic sugar for old-style prototype-based code. For example, the following are more or less equivalent in normal JavaScript (but the prototype style doesn’t work in Microvium):

// ES6 class style
class Foo {
  bar() { console.log('Method bar called') } 
}

// Old prototype-style
function Foo() {}
Foo.prototype.bar = function() { console.log('Method bar called') }

Both styles are setting up the Foo.prototype object, but with the class syntax, it is implicit. Then if you evaluate new Foo(), the engine will create a new object whose __protype__ is Foo.prototype.

This is all possible because functions in JavaScript are also objects, and can hold properties like any other object.

But Microvium functions do not support properties, so the function Foo cannot have a prototype property (or any other static class properties like Foo.myStaticProperty).

I made the choice to omit this feature from the engine because this pattern is relatively uncommon outside of classes. It would be a massive waste of RAM to have every function also be an object. A closure function in Microvium currently uses only uses 6 bytes of RAM, and most functions will use no RAM at all if the static analysis determines them not to be closures (e.g. all top-level functions). Whereas a pair of normal objects like { prototype: {} } is already 16 bytes of RAM (6 bytes for each object plus 4 bytes for each property). It would be madness to require that every function declaration uses 16 bytes of RAM just in case someone wants to use it as a constructor.

So how do classes work in Microvium?

The way it works is quite simple: when you use the class syntax, Microvium creates a function that supports properties (i.e. “a class”), but when you use the normal function syntax, it will keep its old behavior.

A class in Microvium is a distinct type. Although typeof MyClass will return "function", as per the JS spec, if you use the C function mvm_typeOf to probe the type, it will tell you VM_T_CLASS not VM_T_FUNCTION.

Internally, a class is a tuple of a props object and constructor function, as shown in the following diagram. When you access the class like an object (e.g. MyClass.myStaticProperty or MyClass.prototype), it delegates the property-access operation to the props object.

Memory layout for a minimal empty class and an instantiation of the class

When you construct an instance of the class using new MyClass(), it creates a new object whose prototype (__proto__) is props.prototype, and then calls the constructor function with the new object as the this parameter.

The constructor function itself could just be a bytecode function in ROM, but it can, in general, be any function because it delegates the call to the normal calling machinery of Microvium. For example, if you have a class declaration nested inside a function then the constructor will be a closure, with access to the locals of the surrounding function. Or the constructor can be an imported host function. But the simplest case is a class declared at the top level of the code, in which case the constructor part of the tuple just points directly to a function in the bytecode ROM image.

The key-value pair in the above diagram (where the key is "prototype") is how Microvium implements object properties. Each property is a 4-byte pair containing a 2-byte property key and a 2-byte value. Here I’m only assuming one property (MyClass.prototype) but other properties would follow contiguously if there were more.

The next field is unimportant for this example. When you add properties to an object, they’re actually added using a linked list rather than resizing the object structure (for CPU efficiency) but then the garbage collector compacts these linked lists into a contiguous form (for memory efficiency).

Classes and objects are expensive

Classes (and objects) are one of the more expensive features of Microvium. The above minimal/empty class is already 22 bytes of RAM. While each instance of the class is only 6 bytes of RAM, every property on an instance is an additional 4 bytes. So an object with just 2 properties is already 14 bytes.

Property lookup is also quite expensive:

  • It is linear-time because it has to search through the property list.
  • Many different things support property lookup, such as arrays, classes, buffers, and of course objects. The lookup code needs to figure out what it is in order to know how to find its properties.
  • Properties use string keys. For hard-coded literal strings, like x.y or x['y'], these strings don’t incur RAM overhead but they do add to the bytecode size.
  • For computed string properties like x['a' + 'b'], there is additional overhead to perform string interning — string interning is the process of consolidating different strings with the same content so that they also have the same reference identity, which makes property lookups more efficient.
  • String interning can potentially trigger a garbage collection cycle because it’s growing the intern table. Apart from the overhead of the collection itself, just the possibility of a garbage collection means that the implementation of property lookup needs to deal with the fact that all the memory may be shuffled around during the property access (e.g. all the temporaries need to be properly GC-reachable), which itself adds overhead.
  • A property lookup instruction in bytecode involves multiple steps consuming at least 5 bytes of bytecode:
    1. Loading the object from which to get the property (a 1-byte instruction, at least)
    2. Loading the string that represents the property (typically a 3-byte instruction because it embeds a reference to the string)
    3. An instruction to actually trigger the property loading (a 1-byte instruction)

Closures, on the other hand, are asymptotically more efficient. A closure that closes over 2 variables is 14 bytes — the same size as an object with 2 properties. But:

  • Each additional variable closed over is another 2 bytes, rather than the 4 bytes of a property.
  • Closure variable access is typically O(1) because closures are random-access structures and the indexes are computed at compile time by static analysis (see my post on closure variable indexing).
  • Up to 15 different closure variables can be accessed by a single 1-byte instruction, compared to a minimum of 5 bytes for the instruction sequence for object property access.

Can objects be more efficient?

The experimental static analysis technique I designed for Microvium Boost a while back computes all the information that would be required to convert an object to a fixed-length array, in cases where it’s possible to infer the closed set of property keys that are ever used to access an object. My hope is that in the future I could implement this optimization (transforming objects and classes into fixed-length arrays) which could have some significant performance benefits:

  • An object with 2 properties would be 6 bytes instead of 14.
  • Each additional object property would take another 2 bytes.
  • Property lookup would be O(1).
  • Up to 15 different properties could be accessed using a single-byte instruction.

Conclusion

As it stands, classes are not a feature that I would want to use very often since they are so expensive. But having classes opens the door to different kinds of libraries and APIs, and in particular, is a stepping-stone towards implementing other standards-compliant features in Microvium.

Inside Microvium Closures

Inside Microvium Closures

TL;DR: Support for closures in Microvium sets it apart from other JS engines of a similar size. Closures simplify state machines and enable functional-style code. Closures in snapshots are a new way of sharing compile-time state with the runtime program. This post goes through some examples and design details.

What is a closure?

MDN has already done a great job of explaining what a closure is, so I’m just going to borrow their explanation:

closure is the combination of a function bundled together (enclosed) with references to its surrounding state (the lexical environment). In other words, a closure gives you access to an outer function’s scope from an inner function. In JavaScript, closures are created every time a function is created, at function creation time.

Here is a simple example:

function makeCounter() {
  let x = 0;
  function incCounter() {
    x++;
    return x;
  }
  return incCounter;
}

const myCounter1 = makeCounter();
const myCounter2 = makeCounter();

console.log(myCounter1()); // 1
console.log(myCounter1()); // 2
console.log(myCounter1()); // 3

// myCounter2 is an independent counter
console.log(myCounter2()); // 1
console.log(myCounter2()); // 2
console.log(myCounter2()); // 3

In the above example, the function named incCounter is a closure because it closes over variable x in its outer lexical scope. A closure is just a function, nested in another function, which accesses variables in the outer function.

Microvium also supports the arrow function syntax, so the following example has the same output but is more concise:

const makeCounter = x => () => ++x;
const myCounter = makeCounter(0);
console.log(myCounter()); // 1
console.log(myCounter()); // 2
console.log(myCounter()); // 3

For more detail, take a look at the above-mentioned MDN article. The rest of this post will assume that you know what a closure is.

Why are closures useful?

Closures in snapshots

Let’s say that we want a script that exports two functions: one that prints “hello” and the other that prints “world”. Without closures, we could implement this in Microvium as follows:

vmExport(0, printHello);
vmExport(1, printWorld);

function printHello() {
  console.log('hello');
}

function printWorld() {
  console.log('world');
}

(Recall that vmExport is a function that the script calls to export a function to the host).

In this example, printHello and printWorld are each functions that take no arguments and will print the corresponding string to the console1.

With the introduction of closures, we could factor out the commonality between printHello and printWorld and just have a printX that can print either one:

const printHello = makePrinter('hello');
const printWorld = makePrinter('world');
vmExport(0, printHello);
vmExport(1, printWorld);

function makePrinter(thingToPrint) {
  return printX;
  function printX() {
    console.log(thingToPrint);
  }
}

This refactors the code so that console.log only appears once but is shared by both printHello and printWorld. For the simple case of console.log this doesn’t add much benefit, but you can imagine cases where printX is a lot more complicated and so this refactoring may be beneficial.

Another thing to note in this example is that makePrinter is called at compile time since it’s in the top-level code. The resulting closures printHello and printWorld instantiated at compile time are carried to runtime via the snapshot, along with their state (the value of thingToPrint). The closures feature here plays nicely with snapshotting as a new way to share compile-time state to runtime.

Depending on the style of code you’re familiar with, we can also write the same example more concisely as:

const makePrinter = s => () => console.log(x);
vmExport(0, makePrinter('hello'));
vmExport(1, makePrinter('world'));

If you’re not comfortable with the idea of functions returning other functions, here’s another variant of the example that does the same thing:

function exportPrinter(id, textToPrint) {
  vmExport(id, () => console.log(textToPrint));  
}
exportPrinter(0, 'hello');
exportPrinter(1, 'world');

We could also get the list of things to export from a dynamic source, such as an array (or even something read from a file at compile time using fs.readFileSync):

const printers = [
  { id: 0, textToPrint: 'hello' },
  { id: 1, textToPrint: 'world' },
];
// OR:
// printers = JSON.parse(fs.readFileSync('printers.json', 'utf8'));

for (let i = 0; i < printers.length; i++) {
  const id = printers[i].id;
  const textToPrint = printers[i].textToPrint; 
  vmExport(id, () => console.log(textToPrint));  
}

Side note: the above example also demonstrates the usefulness of vmExport being a normal function, rather than some special syntax. Think about your favorite language or engine and how you would implement the above in that language. You can’t define an extern void foo() {} inside a for-loop in C, or public static void foo() {} inside a for-loop in C#. The only solution in these environments to the objective of exporting a programmatically-defined set of functions would be to use a code generator and the result would be much more complicated.

Closures for state machines

It’s much easier and better performance to implement a finite state machine using closures. Consider the following two-state state machine:

The above state machine has 2 states, which I’ve called stateA and stateB. When event 1 is received while in stateA, the machine will transition to stateB, but if any other event (e.g. event 2) is received while in stateA, the machine will not transition to stateB. See Wikipedia for a more detailed description of FSMs.

The ability to use closures allows us to implement this state machine using a function for each state. In the following example, stateA is a function that receives events by its event parameter. We “make” stateA only when we need it, by calling enterStateA(). The example includes an eventCount as part of state A to show how states can have their own variables that are persisted across multiple events.

function enterStateA() {
  console.log('Transitioned to State A!');
  let eventCount = 0; // Some internal state only used by stateA

  // A function that handles events while we're in stateA
  function stateA(event) {
    if (event === 1) {
      currentState = enterStateB();
    } else {
      eventCount++;
      console.log(`Received ${eventCount} events while in state A`);
    }
  }

  return stateA;
}

function enterStateB() {
  console.log('Transitioned to State B!');
  return event => {
    if (event === 2) {
      currentState = enterStateA();
    }
  }
}

// We'll start in stateA
let currentState = enterStateA();

// Every time an event is received, we send it to the current state
const processEvent = event => currentState(event);

// Allow the host firmware to events to the state machine
vmExport(0, processEvent);

// Some example events:
processEvent(5); // Received 1 events while in state A
processEvent(5); // Received 2 events while in state A
processEvent(5); // Received 3 events while in state A
processEvent(1); // Transitioned to State B!
processEvent(1); //
processEvent(2); // Transitioned to State A!
processEvent(2); // Received 1 events while in state A

In the above example, state A has the eventCount counter which is part of its closure. When the system transitions to state B, the counter can be garbage collected. This might not be very useful when only considering a single counter variable, but the pattern generalizes nicely to systems that have more expensive states that may hold buffers and other resources.

Once you understand closures and higher-order functions, this is a very natural way to represent state machines.

Closures under the hood

Let’s go back to the simple “counter” example for illustration:

function makeCounter() {
  let x = 0;
  function incCounter() {
    return ++x;
  }
  return incCounter;
}

const myCounter = makeCounter();
console.log(myCounter()); // 1
console.log(myCounter()); // 2
console.log(myCounter()); // 3

The runtime memory layout for this example is as follows:

The global variable myCounter is a 2-byte slot, as are all variables in Microvium. The slot contains a pointer to the closure, which is an immutable tuple containing a reference to the function code (incCounter, in this example) and the enclosing lexical environment which in Microvium is called the scope.

The closure and scope are allocated on the garbage-collected heap — if and when the closure is no longer reachable, it will be freed.

When the closure myCounter is called, the VM sees that the callee is a closure and sets a special scope register in the VM to the closure’s scope before running the target bytecode. The bytecode can then interact with the scope variables through special machine instructions that leverage the scope register.

For a look into how variables are accessed by closure code, see my post on Closure Variable Indexing.

More efficient than objects

Closures are much more efficient than objects in Microvium:

  • Every closure variable is one word (2 bytes) while an object property is 2 words (4 bytes) because a property is stored as a key-value pair. Variables aren’t stored with their name because the static analysis can determine an index for them.
  • Closure variables can be accessed with single-byte instructions (a 4-bit opcode and 4-bit literal variable index) whereas typical object properties take at least 5 bytes of bytecode instructions to access. This is in part because the “current” closure scope is implicit, while there is no such thing as the “current” object (the object needs to be specified explicitly), and also because object property access requires specifying the property key which is a reference to a string.
  • All the object key strings take up memory in the bytecode image.
  • Most closure variables are accessed in O(1) time — adding more variables does not slow down the access time, but adding more properties to an object slows it down by O(n).

Conclusion

Closures are useful and form the backbone of a lot of JavaScript programming. I’ve talked before about how closures are useful for callback-style asynchronous code, and in this post, I’ve also shown how they are useful for modeling state machines and make certain kinds of refactoring possible.

Other tiny JavaScript engines such as Elk and mJS don’t support closures, so this feature sets Microvium apart from the crowd2.

The closures feature has been the single most complicated and time-consuming feature to implement in Microvium because the static analysis requires multiple passes and completely changed how the compiler front-end works. It’s really not as simple as one would first imagine. Consider, as just one example, that the let i binding in for (let i = 0 ...) has multiple instances of i relative to the containing function, and that on each iteration the previous value is copied to the new instance.

But after a long journey, I can say that I’m happy with the result.


  1. This example assumes that you’ve added `console.log` to the global scope 

  2. Larger engines such as Espruino and XS support closures along with many other features, but come at a serious size premium. 

Microvium has try-catch!

Microvium has try-catch!

TL;DR: Microvium now supports exception handling, with just 4 bytes of overhead per active try block. This post covers some of the details of the design, for those who are interested.


Note: if you’re new to try-catch, see the MDN article on the subject. Note however that Microvium does not yet support finally, but in my experience finally is a less-used feature of JS.

Why try-catch?

One of my objectives with Microvium is to allow users to write JS code that is stylistically similar to the kind of code they would write for a larger engine like V8/Node.js. I want it to be easy to write JS libraries that run on both Microvium and Node, and the public interface of these libraries especially should not be contorted to fit the Microvium feature subset.

A major thing that’s been missing from this vision is the ability to use exceptions to pass errors. A library that can’t use exceptions to pass errors must use some other technique such as error codes, which are not typical JavaScript style and would clutter up the API of such a library.

A look at alternatives

Microvium is implemented in C. How does one implement exceptions in a language like C? In the journey of finding the best way to do exceptions, I thought of a number of different ideas.

The most obvious one to come to mind is something like setjmp/longjmp. This is a common way to implement exception-like behavior in a language like C, since it’s the only way to jump from one C function into another that’s earlier on the stack. At face value, this sounds sensible since both the engine and host are running on C.

Looking at other embedded JavaScript engines, mJS and Elk don’t support exceptions at all, but Moddable’s XS uses this setjmp/longjmp approach. At each try, XS mallocs a jump structure which is populated with the state of current virtual registers and physical registers, as well as a pointer to the catch code that would handle the exception. Upon encountering a throw, all the virtual and physical registers are restored to their saved state, which implicitly unwinds the stack, and then control is moved to the referenced catch block.

In my measurements1, one of these jump structures in XS is 120-140 bytes on an embedded ARM architecture and 256-312 bytes on my x64 machine2. Even the jmp_buf on its own is 92 bytes on Cortex M0 (the architecture to which I’ve targetted all my size tests in Microvium). That’s quite heavy! Not to mention the processing overhead of doing a malloc and populating this structure (each time a try block is entered).

How it works in Microvium

After thinking about it for some time, the following is the solution I settled on. Consider the following code, where we have 2 variables on the stack, and 2 try-catch blocks:

If we were to throw an exception on line 5, what needs to happen?

  1. We need to pop the variable y off the stack (but not x), since we’re exiting its containing scope. This is called unwinding the stack.
  2. We need to pass control to the e1 catch clause (line 7)
  3. We need to now record that the catch(e2) on line 11 is the next outer catch block, in case there is another throw

Microvium does this by keeping a linked list of try blocks on the stack. Each try block in the linked list is 2 words (4 bytes): a pointer to the bytecode address of the corresponding catch bytecode, plus a pointer to the next outer try block in the linked list. For the previous example, there will be 2 try blocks on the stack when we’re at line 5, as shown in the following diagram:

Each of the smaller rectangles here represents a 16-bit word (aka slot). The arrows here represent pointers. The red blocks (with an extra border) each represent an active try block consisting of 2 words (catch and nextTry). They form a linked list because each has a pointer to the next outer try block.

The topTry register points to the inner-most active try block — the head of the linked list. Each time the program enters a try statement, Microvium will push another try block to the stack and record its stack address in the topTry register.

The topTry register serves 2 purposes simultaneously. For one, it points to the head of the linked list, which allows us to find the associated catch block when an exception is thrown. But it also points to exactly the stack level we need to unwind to when an exception is thrown. This is similarly true for each nextTry pointer.

When an exception is thrown, Microvium will:

  1. Take the current topTry and unwind the stack to the level it points to.
  2. Transfer control (the program counter) to the bytecode address of the catch block which is now sitting at the top of the stack (i.e. pop the program counter off the stack).
  3. Save the nextTry pointer as the new topTry (i.e. it pops the topTry register off the stack).

Et voila! We’ve implemented try-catch in Microvium with just 4 bytes of overhead per try.

This also works when throwing from one function to a catch in a caller function. The only difference is that the unwind process in step 1 then needs to restore the registers of the caller (e.g. argument count, closure state, return address, etc). This information is already on the stack since it’s also used by the return instruction — we don’t need to save it separately during a try. This works for any level of nested calls if we just unwind repeatedly until reaching the frame containing the catch target.

Conclusion

I’m pretty satisfied with this design.

  • Only one new virtual register (topTry)
  • Only 4 bytes of RAM per active try3
  • No heap overhead
  • Conceptually simple
  • Only makes the engine 156 bytes larger in ROM

The static analysis to implement this is quite hairy — it’s complicated by the interaction between trycatch and return, break, and closures, among other things. For those interested, see the test cases for examples of the kinds of scenarios that try-catch needs to deal with.

On a meta-level, this is a good reminder of how the first idea that comes to mind is often not the best one, and the benefit of thinking through a problem before jumping right in. I’ve been brainstorming ideas about how exceptions might work in Microvium since July 2020 — almost 2 years ago — and I kept going back to the ideas to revise and reconsider, going back and forth between different options. Sometimes, mulling over an idea over a long period of time can help you get to a much better solution in the end.

Another lesson is that simplicity is deceptively time-consuming. Looking only at this apparently-simple final design, you might not guess at the amount of work required to get there. This is a general tragedy of software engineering: it often takes more time to get to a simpler solution, which gives the appearance of achieving less while taking longer to do so.


  1. Please, anyone correct me if I’ve made a mistake here or misrepresented XS’s design in any way. 

  2. The ranges in sizes here depend on whether I use the builtin jmp_buf type or the one supplied by XS 

  3. And 6 bytes of bytecode ROM per try block in the code. 

Snapshotting is like compiling but better

Snapshotting is like compiling but better

TL;DR: The final output of a traditional compiler like GCC bears a family resemblance to a Microvium snapshot, but the snapshotting paradigm is both easier to use and more powerful because it allows real application code to run at build time and its state to persist until runtime.

What is snapshotting?

My Microvium JavaScript engine is built on the paradigm of creating a VM snapshot as the deployable build artifact rather than creating a traditional compiled binary. As a developer, when you run the Microvium engine on your desktop with a command like microvium main.js it will execute the script until all the top-level code is complete and then output a snapshot file containing the final VM state. The snapshot file can then be “resumed” on an embedded device using Microvium’s embedded C library (for more details, see Getting Started). Although Microvium is designed especially for microcontrollers, the principle of snapshotting goes beyond the embedded space.

Comparing to GCC

For this post, I’ll mostly compare Microvium to GCC:

gcc main.c         # Compile a C program with GCC
microvium main.js  # "Compile" a Microvium program

These two commands are analagous. Both produce a single file as the result, and this file is what you want to deploy to the target environment. In the case of GCC, the output is of course the executable (e.g. a.out), while in the case of Microvium, it’s the snapshot (main.mvm-bc).

Both of these commands do some kind of compilation as part of the process. GCC translates your function code to machine instructions, while Microvium translates to virtual machine instructions (bytecode instructions).

Constants

Both the GCC output and the Microvium output have a section for constants, including function code. You may be familiar with this as the .text section. Among other things, this contains constant values, such as:

// JavaScript
const x = 42;
// C
const int x = 42;

… but better

You can do this in Microvium but not in C:

const x = foo();

function foo() {
  return 42;
}

In C, it’s a compile-time error to call runtime functions for the calculation of constants. But in Microvium, this is perfectly legal since there is no distinction between compile-time and runtime — there is only really runtime before the snapshot and runtime after the snapshot. Although informally we may refer to the former as “compile-time” and the latter as “runtime”.

Apart from just being cleaner and easier to use, the Microvium snapshotting paradigm here allows computationally-intensive constants to be calculated at build time, using arbitrary functions and libraries that might also be useful at runtime.

Variable initializers… but better

For non-constant variables, both GCC and Microvium have an output section for the initial1 value of all the variables, which is copied into RAM at runtime. You may know this traditionally as the .data section.

But a major difference between them is that a Microvium snapshot also contains heap data.

// JavaScript
let arr = [1, 2, 3];

// C
int* arr = malloc(3 * sizeof(int));  // ! Can't do this (in top-level code)

The best you could do in C for the above example would be to have an init function that runs early in the program to set up the initial runtime state of the program. Snapshotting is better because this initial structure can be established at build time.

Modules … but better

Microvium and C both have support for structuring your code in multiple files which get bundled into the same output artifact:

import { foo } from './foo.js'
#include "foo.h"

With the #include here, the dependent module implementation (e.g. foo.c) is not automatically compiled and linked into the program by GCC — you need to separately list foo.c to be compiled by GCC, or orchestrate the dependencies using a makefile.

But in the case of a Microvium import, the import statement itself is executed at build time, performing the module resolution, loading, parsing, and linking at build time, as well as executing the top-level code of the imported module. The top-level code of the imported module may in turn import other modules, transitively importing the whole module graph and executing its top-level initialization code.

Preprocessor… but better

Both C and Microvium support “compile-time” logic:

// C
#if USE_FOO_1
#define foo foo1
#else
#define foo foo2
#endif
// JavaScript
const foo = USE_FOO_1 ? foo1 : foo2;

Of course, you already knew that, because all the examples so far have demonstrated the fact that snapshotting allows you to run JavaScript at compile time. But I want to emphasize some of the key reasons why the snapshotting paradigm is better for this:

  • You don’t need two different programming languages (e.g. the preprocessor language and the C language).
  • Your “compile-time” code has the full power of the main language.
  • Your runtime code carries over the state from your compile-time code.

So in some sense, this unifies the preprocessor language with the main language. This applies similarly to other “compile-time languages”, such as makefiles and linker scripts. Or if you’re coming from the JS world, consider how the snapshotting paradigm obviates the need for a webpack.config (see Snapshotting vs Bundling).

But what about using the preprocessor to conditionally include different runtime logic, such as for different devices? For example, consider the following C code:

int myFunction() {
#if SOME_CONDITION
  doX();
#else
  doY();
#endif 
}

Of course, it’s easy to see how this example might translate to JS:

function myFunction(someCondition) {
  if (someCondition) {
    doX();
  } else {
    doY();
  }
}

Now we have the bonus that our unit tests can inject someCondition to test both cases.

But doesn’t this code mean that now we have both doX and doY branches at runtime, taking up ROM space? (and someCondition)

That’s why I’ve also been developing the experimental Microvium Boost: an optimizer that analyzes a snapshot and removes unused branches of code. For example, if the analysis shows that someCondition is always true in your program, it can remove it as a parameter from myFunction and also remove the call to doY as dead code. This is still experimental but has shown significant success so far.

Host exports… but better

So far we’ve been considering an executable output from GCC, but it would be more accurate to compare a Microvium snapshot with a compiled shared library (e.g. DLL). Like a shared library, Microvium snapshots do not have a single entry point but may contain many exported functions to be resolved at runtime on the device.

A Windows DLL suits the analogy better than a Linux shared library, so in this section the examples will use msbuild rather than GCC.

Both a Microvium snapshot and a DLL binary contain a section for dynamic linking information — a table that associates relevant functions in the DLL with a number2 so that the host program using the DLL/snapshot at runtime can find them.

In the case of a DLL, you can provide the compiler with a DEF file that tells the compiler what to put into the DLL export table. If you wanted to export the functions foo, bar, and baz from the DLL with IDs 1, 2, and 3 respectively, your DEF file might look like this:

LIBRARY   MyLibrary
EXPORTS
   foo   @1
   bar   @2
   baz   @3
// C
int foo() { return 42; }
int bar() { return 43; }
int baz() { return 44; }

The equivalent in Microvium would be as follows:

// JavaScript
function foo() { return 42; }
function bar() { return 43; }
function baz() { return 44; }

vmExport(1, foo);
vmExport(2, bar);
vmExport(3, baz);

You may have noticed a recurring theme in this post: the Microvium snapshotting paradigm doesn’t require a whole new language in order to do different build-time tasks. In this case, Microvium doesn’t require a DEF file (or a special __declspec(dllexport) language extension), since vmExport is just a normal function. This is just simpler and more natural.

Another recurring theme here is that the snapshotting approach is more powerful, allowing you to do things that are impossible or impractical in the traditional paradigm. Take a look at the following example in Microvium:

// JavaScript
for (let i = 1; i <= 3; i++) {
  vmExport(i, () => 41 + i);
}

This has the same overall effect as the previous code, adding 3 functions to the export table of the deployed binary with IDs 1, 2, and 3 and which return 42, 43, and 44 respectively. But having vmExport be a normal function means that now we have the full power of the language for orchestrating these exports, or for writing an abstraction layer over the export system, or outsourcing this logic to a third-party library.

Side note: a more subtle point in this example for advanced readers is its code cohesion. The single line of code mixes both compile-time and runtime code (vmExport(i,...) and ()=> 41 + i respectively), but keeps the related parts of both in close proximity. This is the difference between temporal cohesion (grouping code by when its run) vs functional cohesion (grouping code by what feature it relates to) (see Wikipedia). A common disadvantage of having separate build-time or deploy-time code (e.g. a DEF file, makefile, linker script, webpack.config, dockerfile, terraform file, etc) is that it pushes you into temporal cohesion, which in turn damages modularity and reusability.

Conclusion

The idea of deploying a snapshot rather than a traditional compiled binary opens up a whole new paradigm for software development. The end result is very similar — a binary image with sections for different memory spaces, compiled function code, constants, initial variable values, and export/import tables — but the snapshotting paradigm is both simpler and more powerful.


  1. In the context of Microvium, the word “initial” here refers to the initial state when the snapshot is resumed, not the initial state when the program is started, since the program starts at build time, and variables in JavaScript start with the value undefined. 

  2. DLL exports can be by name or number, but Microvium exports are only by number, for efficiency reasons, so that’s what I’m using in the analogy here. 

Microvium is very small

Microvium is very small

TL;DR: The Microvium JavaScript engine for microcontrollers takes less than 16 kB of ROM and 64 bytes of RAM per VM while idle, making it possibly the smallest JavaScript engine to date with more language features than engines 4x its size.


I’ve designed Microvium from the ground up with the intention for it to be tiny, and it’s been an absolute success in that sense. Microvium may be the smallest JavaScript engine out there, but it still packs a punch in terms of features.

*Edit: this post was originally written when Microvium was around 8.2 kB (rounded up to 8.5 kB). Since then, new features have been added. As of June 2022, Microvium is now 9.04 kB (rounded up to 10 kB). In the TLDR headline, I’ve rounded this up again to “less than 16 kB” since that figure will remain true for a very long time (possibly forever). Similarly, the idle RAM usage for a VM can be as low as 22 B1, but I’ve rounded this up to “less than 64 bytes” to accommodate different situations and future expansion.

Does size matter?

Size often matters in small MCU devices. A large proportion of microcontroller models available on the market still have less than 64 kB of flash and less than 2 kB of RAM. These are still used because they’re smaller, cheaper, and have lower power than their larger counterparts. All the microcontrollers I’ve worked with in my career as a firmware engineer have had ≤ 16 kB RAM.

Some might say that you shouldn’t even want JavaScript on such small devices, and certainly in some cases that would be true. But as I pointed out in my last post, juggling multiple operations in firmware can be both easier and more memory efficient if the high-level logic is described in terms of a language like JavaScript, even if that’s the only thing you’re using it for.

Even on larger devices, do you really want to dedicate a large chunk of it to a JavaScript engine? A smaller engine is a smaller commitment to make — a lower barrier to entry.

How does it compare?

If I Google “smallest JavaScript engine for microcontrollers”, the first one on the list is Elk. Elk is indeed pretty tiny. For me, it compiles to just 11.5 kB of flash2. Microvium compiled with the same settings compiles to about 10 kB — in the same ballpark.

What about RAM?

The amount of RAM Elk uses is not pre-defined — you give it a buffer of RAM of any size you want, but it needs to be at least 96 bytes for the VM kernel state. Microvium takes 54 bytes for the kernel state.

But where there’s a massive difference in memory requirement is that Elk requires all of its memory allocated upfront, and keeps it for the lifetime of the VM. If your script’s peak memory in Elk is 1 kB then you need to give it a 1 kB buffer at startup, so its idle memory usage is 1 kB. Microvium on the other hand uses malloc and free to allocate when needed and free when not needed. Its idle memory usage can be as low as 62 bytes3. In typical firmware, idle memory is much more important than peak memory, as I explained in my last post.

What about the feature set? This is another area where Microvium and Elk diverge significantly. The following table shows the differences:

MicroviumElk
var, const (Elk supports let only)
do, switch, for
Computed member access a[b]
Arrow functions, closures
trycatch
Modules
Snapshotting
Uses intermediate bytecode (better performance)
Parser at runtime
ROM10 kB11.5 kB
Idle RAM36 BLots
Peak kernel RAM56 B96 B
Slot size (size of simple variables)2 B8 B

The only thing that Elk can do that Microvium can’t do is execute strings of JavaScript text at runtime. So if your use case involves having human users directly provide scripts to the device, without any intermediate tools that could pre-process the script, then you can’t use Microvium and you might want to use Elk, mJS, or a larger engine like XS. On the other hand, if your use case has at any point a place where you can preprocess scripts before downloading them to the device then you can use Microvium.

Comparing with mJS

But Cesanta, the maker of Elk, also made a larger JS engine with more features: mJS, which is probably the closest match to Microvium in terms of feature set. mJS lets you write for-loops and switch statements for example.

Since they’re closely matched for intent and features, I did a more detailed comparison of mJS and Microvium here. But here’s a summary:

MicroviummJSElk
var, const (mJS supports let only)
Template strings
Arrow functions and closures
trycatch
ES Modules
(but mJS does support a non-standard load function)
do, switch, for
Computed member access a[b]
Uses intermediate bytecode (better performance)
Some builtin-functions
Parser at runtime
ROM10 kB45.6 kB11.5 kB
Slot size2 B8 B8 B

I’ve lumped “some builtin-functions” into one box because it’s not a language feature as such. mJS has a number of builtin functions that Microvium doesn’t have – most notably print, ffi, s2o, JSON.stringify, JSON.parse and Object.create. You can implement these yourself in Microvium quite easily without modifying the engine (or find implementations online), and it gives you the option of choosing what you want rather than having all that space forced on you4.

In terms of features, mJS is a more “realistic” JavaScript engine, compared to Elk’s minimalistic approach. I wouldn’t want to write any substantial real-world JavaScript without a for-loop for example. Like Microvium, mJS also precompiles the scripts to bytecode and then executes the bytecode, which results in much better performance than trying to parse on the fly. Engines like Elk that parse as they execute also have the unexpected characteristic that comments and whitespace slow them down at runtime.

But the added features in mJS means it costs a lot more in terms of ROM space — about 4x more than Elk and 5x more than Microvium.

Microvium still has more core language features than mJS, making it arguably a more pleasant language to work in. These features are actually quite useful in certain scenarios:

  • Proper ES module support is important for code organization and means that your Microvium modules can also be imported into a node.js or browser environment. You can have the same algorithms shared by your edge devices (microcontrollers), backend servers, and web interfaces, to give your users a unified experience.
  • Closures are fundamental to callback-style asynchronous code, as I explained in my previous post.

Conclusion

I’m obviously somewhat biased since Microvium is my own creation, but the overall picture I get is this:

  • Microvium is the smallest JavaScript engine that I’m aware of5
  • In this tiny size, Microvium actually supports more core language features than engines more than 5x its size. Some of these features are really useful for writing real-world JS apps.
  • Having said that, Microvium has fewer built-in functions — it’s more of a pay-as-go philosophy where your upfront commitment is much less and you bring in support for what you need when you need it.
  • The big trade-off is that Microvium doesn’t have a parser at runtime. In the rare case that you really need a parser at runtime, Microvium simply won’t work for you.

Something that made me smile is this note by one of the authors of mJS in a blog posts:

That makes mJS fit into less than 50k of flash space (!) and less than 1k of RAM (!!). That is hard to beat.

https://mongoose-os.com/blog/mjs-a-new-approach-to-embedded-scripting/

I have great respect for the authors of mJS and what they’ve done, which makes me all the more proud that Microvium is able to knock this out of the ballpark, beating what the seasoned professionals have called “hard to beat”. Of course, this comes with some tradeoffs (no parser and no builtin functions), but I’ve achieved my objective of making a JavaScript engine that has a super-low upfront commitment and will squeeze into the tiniest of free spaces, all while still including most of the language features I consider to be important for real-world JavaScript apps.


  1. When compiled for a 16-bit device and excluding any built-in objects. 

  2. All of the sizes quoted in this post are when targetting the 32-bit ARM Cortex M0 using GCC with optimization for size. I’m measuring these sizes in June 2022, and of course they may change over time. 

  3. This idle memory figure includes the overhead of a single heap bucket with the default array prototype. The earlier quoted “kernel state” does not include this but includes the register bank which is not allocated during idle time. 

  4. The ffi in mJS is something that would need to be a built-in in most engines but Microvium’s unique snapshotting approach makes it possible to implement the ffi as a library just like any of the other functions 

  5. Please let me know if you know of a smaller JS engine than Microvium. 

Single-threading is more memory-efficient

Single-threading is more memory-efficient

TL;DR: Single-threading with super-loops or job queues may make more efficient use of a microcontroller’s memory over time, and Microvium’s closures make single-threading easier with callback-style async code.

Multi-threading

In my last post, I proposed the idea that we should think of the memory on a microcontroller not just as a space but as a space-time, with each memory allocation occupying a certain space for some duration of time. I suggested therefore that we should then measure the cost of an allocation in byte-seconds (the area of the above rectangles as bytes × seconds), so long as we assumed that allocations were each small and occurred randomly over time. Randomness like this is a natural byproduct of a multi-threaded environment, where at any moment you may coincidentally have multiple tasks doing work simultaneously and each taking up memory. In this kind of situation, tasks must be careful to use as little memory as possible because at any moment some other tasks may fire up and want to share the memory space for their own work.

The following diagram was generated by simulating a real malloc algorithm with random allocations over time (code here, and there’s a diagram with more allocations here):

A key thing to note is that the peak memory in this hypothetical chaotic environment can be quite a bit higher than the average, and that these peaks are not easily predictable and repeatable because they correspond to the coincidental execution of multiple tasks competing for the same memory space1.

This leaves a random chance that at any moment you could run out of memory if too many things happen at once. You can guard against this risk by just leaving large margins — lots of free memory — but this is not a very efficient use of the space. There is a better way: single threading.

Single threading

Firmware is sometimes structured in a so-called super-loop design, where the main function has a single while(1) loop that services all the tasks in turn (e.g. calling a function corresponding to each task in the firmware). This structure can have a significant advantage for memory efficiency. In this way of doing things, each task essentially has access to all the free memory while it has its turn, as long as it cleans up before the next task, as depicted in the following diagram. (And there may still be some statically-allocated memory and “long-lived” memory that is dynamic but used beyond the “turn” of a task).

Overall, this is a much more organized use of memory and potentially more space-efficient.

In a multi-threaded architecture, if two memory-heavy tasks require memory around the same time, neither has to wait for the other to be finished — or to put it another way, malloc looks for available space but not available time for that space. On the other hand, in a super-loop architecture, those same tasks will each get a turn at different times. Each will have much more memory available to them during their turn while having much less impact on other tasks the rest of the time. And the overall memory profile is a bit more predictable and repeatable.

An animated diagram you may have seen before on my blog demonstrates the general philosophy here. A task remains idle until its turn, at which point it takes center stage and can use all the resources it likes, as long as it packs up and cleans up before the next task.

So, what counts as expensive in this new memory model?

It’s quite clear from the earlier diagram:

  • Memory that is only used for one turn is clearly very cheap. Tasks won’t be interrupted during their turn, so they have full access to all the free memory without impacting the rest of the system.
  • Statically-allocated memory is clearly the most expensive: it takes away from the available memory for all other tasks across all turns.
  • Long-lived dynamic allocations — or just allocations that live beyond a single turn — are back to the stochastic model we had with multi-threading. Their cost is the amount of space × the number of turns they occupy the space for. Because these are a bit unpredictable, they also have an additional cost because they add to the overall risk of randomly running out of memory, so these kinds of allocations should be kept as small and short as possible.

Microvium is designed this way

Microvium is built from the ground up on this philosophy — keeping the idle memory usage as small as possible so that other operations get a turn to use that memory afterward, but not worrying as much about short spikes in memory that last only a single turn.

  • The idle memory of a Microvium virtual machine is as low as 34 bytes2.
  • Microvium uses a compacting garbage collector — one that consolidates and defragments all the living allocations into a single contiguous block — and releases any unused space back to the host firmware. The GC itself uses quite a bit of memory3 but it does so only for a very short time and only synchronously.
  • The virtual call-stack and registers are deallocated when control returns from the VM back to the host firmware.
  • Arrays grow their capacity geometrically (they double in size each time) but a GC cycle truncates unused space in arrays when it compacts.

See here for some more details.

Better than super-loop: a Job Queue

The trouble with a super-loop architecture is that it services every single task in each cycle. It’s inefficient and doesn’t scale well as the number of tasks grows4. There’s a better approach — one that JavaScript programmers will be well familiar with: the job queue.

A job queue architecture in firmware is still pretty simple. Your main loop is just something like this:

while (1) {
  if (thereIsAJobInTheQueue) 
    doNextJob();
  else
    goToSleep();
}

When I write bare-metal firmware, often the first thing I do is to bring in a simple job queue like this. If you’re using an RTOS, you might implement it using RTOS queues, but I’ve personally found that the job-queue style of architecture often obviates the need for an RTOS at all.

As JavaScript programmers may also be familiar with, working in a cooperative single-threaded environment has other benefits. You don’t need to think about locking, mutexes, race conditions, and deadlocks. There is less unpredictable behavior and fewer heisenbugs. In a microcontroller environment especially, a single-threaded design also means you also save on the cost of having multiple dedicated call stacks being permanently allocated for different RTOS threads.

Advice for using job queues

JavaScript programmers have been working with a single-threaded job-queue-based environment for decades and are well familiar with the need to keep the jobs short. When running JS in a browser, long jobs means that the page becomes unresponsive, and the same is true in firmware: long jobs make the firmware unresponsive — unable to respond to I/O or service accumulated buffers, etc. In a firmware scenario, you may want to keep all jobs below 1ms or 10ms, depending on what kind of responsiveness you need5.

As a rule of thumb, to keep jobs short, they should almost never block or wait for I/O. For example, if a task needs to power-on an external modem chip, it should not block while the modem to boots up. It should probably schedule another job to handle the powered-on event later, allowing other jobs to run in the meantime.

But in a single-threaded environment, how do we implement long-running tasks without blocking the main thread? Do you need to create complicated state machines? JavaScript programmers will again recognize a solution…

Callback-based async

JavaScript programmers will be quite familiar the pattern of using continuation-passing-style (CPS) to implement long-running operations in a non-blocking way. The essence of CPS is that a long-running operation should accept a callback argument to be called when the operation completes.

The recent addition of closures (nested functions) as a feature in Microvium makes this so much easier. Here is a toy example one might use for sending data to a server in a multi-step process that continues across 3 separate turns of the job queue:

function sendToServer(url, data) {
  modem.powerOn(powerOnCallback);

  function powerOnCallback() {
    modem.connectTo(url, connectedCallback);
  }

  function connectedCallback() {
    modem.send(data);
  } 
}

Here, the data parameter is in scope for the inner connectedCallback function (closure) to access, and the garbage collector will automatically free both the closure and the data when they aren’t needed anymore. A closure like this is much more memory-efficient than having to allocate a whole RTOS thread, and much less complicated than manually fiddling with state machines and memory ownership yourself.

Microvium also supports arrow functions, so you could write this same example more succinctly like this:

function sendToServer(url, data) {
  modem.powerOn( () => 
    modem.connectTo(url, () => 
      modem.send(data)));
}

Each of these 3 stages — powerOn, connectTo and send — happen in a separate job in the queue. Between each job, the VM is idle — it does not consume any stack space6 and the heap is in a compacted state7.

If you’re interested in more detail about the mechanics of how modem.powerOn etc. might be implemented in a non-blocking way, take a look at this gist where I go through this example in more detail.

Conclusion

So, we’ve seen that multi-threading can be a little hazardous when it comes to dynamic memory management because memory usage is unpredictable, and this also leads to inefficiencies because you need to leave a wider margin of error to avoid randomly running out of memory.

We’ve also seen how single-threading can help to alleviate this problem by allowing each operation to consume resources while it has control, as long it cleans up before the next operation. The super-loop architecture is a simple way to achieve this but an event-driven job-queue architecture is more modular and efficient.

And lastly, we saw that the Microvium JavaScript engine for embedded devices is well suited to this kind of design, because its idle memory usage is particularly small and because it facilitates callback-style asynchronous programming. Writing code this way avoids the hassle and complexity of writing state machines in C, of manually keeping track of memory ownership across those states, and the pitfalls and overheads of multithreading.


  1. This simulation with random allocations is not a completely fair representation of how most firmware allocates memory during typical operation, but it shows the consequence of having many memory-consuming operations that can be preempted unpredictably or outside the control of the firmware itself. 

  2. Or 22 bytes on a 16-bit platform 

  3. In the worst case, it doubles the size of heap while it’s collecting 

  4. A super-loop also makes it more challenging to know when to put the device to sleep since the main loop doesn’t necessarily know when there aren’t any tasks that need servicing right now, without some extra work. 

  5. There will still be some tasks that need to be real-time and can’t afford to wait even a few ms in a job queue to be serviced. I’ve personally found that interrupts are sufficient for handling this kind of real-time behavior, but your needs may vary. Mixing a job queue with some real-time RTOS threads may be a way to get the best of both worlds — if you need it. 

  6. Closures are stored on the virtual heap. 

  7. It’s in a compacted state if you run a GC collection cycle after each event, which you would do if you cared a lot about idle memory usage. 

Short-lived memory is cheaper

Short-lived memory is cheaper

TL;DR: RAM on a microcontroller should not just be thought of as space but as space-time: a task that occupies the same memory but for a longer time is more expensive.

MCU memory is expensive

If you’re reading this, I probably don’t need to tell you: RAM on a microcontroller is typically very constrained. A 3 GHz desktop computer might have 16 GB of RAM, while a 3 MHz MCU might have 16 kB of RAM — a thousand times less processing power but a million times smaller RAM. So in some sense, RAM on an MCU may be a thousand times more valuable than on a desktop machine. Regardless of the exact number, I’m sure we can agree that RAM is a very constrained resource on an MCU1. This makes it important to think about the cost of various features, especially in terms of their RAM usage.

Statically-allocated memory

It’s common especially in smaller C firmware to just pre-allocate different pieces of memory to different components of the firmware (rather than using malloc and free). For example, at the global level, we may declare a 256-byte buffer for receiving data on the serial port:

uint8_t rxBuffer[256];

If we have 1kB of RAM on a device for example, maybe there are 4 components that each own 256 B. Or more likely, some features are more memory-hogging than others, but you get the idea: in this model, each component in the code owns a piece of the RAM for all time.

Dividing up RAM over time

It’s of course a waste to have a 256-byte buffer allocated forever if it’s only used occasionally. The use of dynamic memory allocation (malloc and free) can help to resolve this by allowing components to share the same physical memory at different times, requesting it when needed, and releasing it when not needed.

This allows us to reconceptualize memory as a mosaic over time, with different pieces of the program occupying different blocks of space-time (occupying memory space for some time).

When visualizing memory like this, it’s easy to start to feel like the cost of a memory allocation is not just the amount of memory it locks, but also the time that it locks it for (e.g. the area of each rectangle in the above diagram). In this sense, if I allocate 256 bytes for 2 seconds then it costs 512 byte-seconds, which is an equivalent cost to allocating 512 bytes for 1 second or 128 bytes for 4 seconds.

Being a little bit more rigorous

Skip this section if you don’t care about the edge cases. I’m just trying to be more complete here.

This measure of memory cost is of course just one way of looking at it, and breaks down in edge cases. For example, on a 64kB device, a task2 that consumes 1B for 64k-seconds seems relatively noninvasive while a task that consumes 64kB for 1s is much more costly. So the analogy breaks down in cases where the size of the allocations are significant compared to the total memory size.

Another way the model breaks down is if many of the tasks need memory around the same time — e.g. if there is some burst of activity that requires collaboration between many different tasks. The typical implementation of malloc will just fail if there is not memory available right now, as opposed to perhaps blocking the thread until the requested memory becomes available, as if the memory was like a mutex to be acquired.

But the model is accurate if we make these assumptions:

  • The size of the individual allocations is small relative to the total memory size
  • Allocations happen randomly over time and space

Under these assumptions, the total memory usage becomes a stochastic random variable whose expected value is exactly:

The expected allocation size × the expected allocation size × the expected allocation frequency

We could also calculate the probability that the device runs out of memory at any given point (I won’t do the calculations here).

Conclusion

In situations where memory allocations can be approximated as being small and random, the duration of a memory allocation is just as important as its size. Much more care must be taken to optimize memory usage for permanent or long-lived operations.

I’m not very happy with this stochastic viewpoint and all these edge cases. It means that at any point, we could randomly exceed the amount of available memory and the program will just die. Is there a better way to organize memory space-time so we don’t need to worry as much? I believe there is… and I’ll cover that in the next post.


  1. Another thing that makes it a constrained resource is the lack of virtual memory and the ability to page memory in and out of physical RAM, so the RAM size is a hard limit 

  2. When talking about this space-time model, I think it’s easier to talk about a “task” than a “component”, where a task here is some activity that needs to be done by the program over a finite stretch of time, and will consume resources over that time. 

Can you parse this?
JavaScript Corners

Can you parse this?
JavaScript Corners

What does the following JavaScript mean:

const x = await / +y; const z = await / +y;

Hint: it’s a trick question.

The answer depends on the context, as is demonstrated by the following snippet:

function foo() {
  const y = 10;
  const await = 5;
  const x = await / +y; const z = await / +y;
  console.log(x);
}
async function bar() {
  const y = 10;
  const x = await / +y; const z = await / +y;
  console.log(x);
}
foo(); // Prints 0.5
bar(); // Prints / +y; const z = await /10

Within the context of an async function, await is like a keyword, and the thing after await is considered to be an expression. In JavaScript, an expression that starts with forward-slash is a Regexp literal, and that literal ends with the next unescaped forward slash. The +y at the end then represents string concatenation, so both the regular expression and y are converted to strings, and the concatenated result string is "/ +y; const z = await /10".

This interpretation is easier to visualize if the syntax highlighting identifies and colorizes the respective parse tokens as follows:

Outside of the context of an async function, await is just a normal identifier and has no special meaning (this is important so that the introduction of the await syntax to the JavaScript language didn’t modify the meaning of existing JavaScript code which might have used await as a variable or parameter name).

If syntax highlighting was correct, as seen in the above images, the difference would be pretty obvious. Unfortunately, I needed to photo-shop the above images, since VS Code highlights both examples the same, and both incorrect:

Global To-String
JavaScript Corners - Part 9

Global To-String
JavaScript Corners - Part 9

(This is Part 9 in my series on JavaScript corner cases).

Here’s another one.

In JavaScript, global variables are properties of the global object. By default, the global object is like any other, and inherits from the Object.prototype  object. Object.prototype comes with a number of its own properties, such as the toString method. So, that means that toString is also a global variable1.

console.log('toString' in global); // prints true
console.log(toString === global.toString); // prints true
console.log(global.toString()); // prints [object global]
console.log(toString()); // prints [object Undefined]. Why is this?

Everything seems expected, except the last line, which might seem a little confusing. The toString() call is clearly invoking a function using a reference to that function, where the base of the reference is the global object, right? (Take a look at my posts on references). So surely toString() and global.toString() mean the same thing?

Wrong.

There’s a subtlety here. The unqualified toString reference actually has a base value2 that is the global environment, which “knows about” the global object, but is not exactly the global object. The base object for the global environment is actually always the value undefined. See here in the spec. This is why it prints “[object Undefined]” .

 


  1. To qualify as a global variable, there is actually an additional criterion. The property of the global object must not be listed in the set of unscopables on the global object. In this case, toString is not listed as an unscopable, since it was introduced into JavaScript before the existence of the unscopables feature, and for backwards compatibility it remains that way. 

  2. Recall that a reference has two components: the thing being referred on, and the name of the thing being referred to. For example, referring to the property named x on the object obj, in the case of obj.x 

JavaScript Corners – Part 9
Node.js With-Statement Bug

JavaScript Corners – Part 9
Node.js With-Statement Bug

What does the following evil code print?

var x = 'before';
var obj = { x };
with (obj) {
  x = (delete x, 'after');
}
console.log(x);

If you’re not sure, don’t worry — neither are current JavaScript engines. Firefox prints “after”, while Edge, IE, and Node.js print “before” (node v7.9.0). I believe that Firefox is correct in this case.

The tricky statement is obviously the following one, which sets a property on an object in the same statement that deletes the property:

x = (delete x, 'after');

(Side note: if you’re not very familiar with JavaScript, the relevant language features that are being used here are the delete operator, comma operator, and the good ol’ evil with statement).

What we expect to happen

The statement var x introduces a new variable at the script scope1.

The { x } expression creates a new object with a single property2 x, where the value of x  is copied from the variable x in the outer scope, so it has the initial value of ‘before’.

The with  statement brings the properties of the object obj into scope in a new lexical environment.

The statement x = (delete x, ‘after’) should perform the following steps:

  1. Evaluate the left hand side
  2. Evaluate the right hand side
  3. Assign the value from the right hand result, to the reference created when evaluating the left hand side

When the left hand side is evaluated, the property x will be found in object obj. The base value of the reference is the object, not the script variable scope.

The right hand side evaluates to ‘after’, but in the process it deletes the property x  from obj. However, the reference on the left hand side should still refer to “the property named ‘x’ on the object obj“, even though the property with that name is now deleted.

When the assignment happens, it should create a new property named ‘x’ on object obj, with value ‘after’. The variable x in the outer scope should be left unaffected.

In this case, I think Node.js gets the wrong answer.


  1. Theoretically, the script scope is the global scope. But in Node.js, scripts are wrapped in a module wrapper that changes the behavior of global vars. This doesn’t affect the outcome of this experiment though 

  2. Bonus fact. Object literals inherit from the global intrinsic object Object.prototype, which has other properties on it, such as toString. So when I say that it has a single property, it would be more accurate to instead say that it has a single own property