TL;DR: Microvium closures are now as small as 6 bytes on the heap, down from 12 bytes previously and compared to over 100 B in some other engines.

---

Microvium is all about squeezing things into tiny spaces, and the latest of these is closures — the memory structure used to carry the state of nested functions. If you’re new to closures, have a look at my previous post which explains closures from a C programmer’s perspective.

One of my goals with Microvium has been to make it feasible and realistic to run JavaScript on really tiny devices. More than half of the microcontrollers selling on Digikey have less than 16kB of RAM and 128kB of flash. When you’re working with devices this small, you count every byte you use.

People typically program these kinds of devices in C. However, as I’ve shown before, there can be some real, practical benefits to using a language like JavaScript in these situations. For example, JavaScript-style single-threading can be more memory-efficient than multithreading, and using a garbage collector like Microvium’s can avoid heap fragmentation overhead and help to reduce the memory leaks associated with manually freeing memory. And of course, your code may be less complicated when working in a higher-level language, which itself can mean less development time and fewer bugs.

When you’re working with such small memory spaces, what kind of size would you consider to be acceptable for a closure function, like the myCounter in the following example?

function makeCounter() {
  let x = 0;

  function incCounter() {
    return ++x;
  }

  return incCounter;
}

const myCounter = makeCounter();

As I demonstrated in my previous post, if you code up something like this in C, it might take 20 bytes of memory¹, ignoring fragmentation overhead. So what kind of size would be acceptable to get the convenience of the JavaScript syntax? Maybe 30 bytes? 50 bytes?

Other JavaScript engines that I’ve measured take over 100 bytes for a closure like this! (See here for my measurement methodology and feel free to correct it). That’s not a fault of the engine, it’s a product of the JavaScript spec which requires that all functions are objects. For function declarations (functions declared using function rather than arrow syntax), not only is the function an object, but it also has a fresh prototype object in case you use the function with new.

That’s a heavy price to pay! Maybe that’s ok on a desktop-class machine. But when you’re working on a device with tens of kilobytes of memory, that’s just not an affordable feature anymore. So, in Microvium, closures are stripped down to their bare minimum. Functions in Microvium are not objects — they cannot have properties. And in return for this trade-off, Microvium closures can be really tiny: the closure in the above example is just 8 bytes! (Including a 2-byte reference and 2-byte allocation header).

Another reason why this is so small in Microvium is its 16-bit slot size. It stores all pointers as 16-bit, even on a 32-bit machine. Numbers, such as x in the example, start out as 16-bit but grow as needed. This is great for things like counters which are likely to be small most of the time but which are able to count up to 2⁵³ without overflowing. This is in contrast to C where you typically need to use memory upfront for the largest-possible value that a variable might have.

In general, closures like this in Microvium take 4 + 2n bytes of memory on the heap, where n is the number of variables. This is down from 10 + 2n bytes of memory in the previous design which is a nice improvement.

---

1. Assuming here a 32-bit device and FreeRTOS heap. The size I’ve quoted here includes an 8-byte allocation, with 8-byte allocation header, and a 4-byte pointer to the allocation ↩

TL;DR: Closures are a good way to represent callbacks for non-blocking code. Although C doesn’t support closures, you can achieve something similar using structs on the heap, but you can get even better memory efficiency by using JavaScript running on Microvium.

---

I’ve recently made some improvements to the memory usage of closures in Microvium, which I’ll talk about in more detail in an upcoming post. But before I get there, I want to first talk about how a similar problem can be solved in C, as a point of comparison.

In this post, we’ll look at a scenario where closures are useful, starting with the most basic solution to the problem in C and then progressively improving the design to address its weaknesses until we end up with something closure-like, which I will then contrast with the same solution in JavaScript. This post is aimed at people who are more familiar with C than with JavaScript, and are not necessarily familiar with closures.

Preserving state during long-lived operations

Let’s start by imagining that we have a C program that needs to perform some slow, I/O-bound operation, such as sending data to a server. For simplicity, let’s assume that we just need to send a single integer, and let’s assume that someone has given us a library that does exactly this:

// Send an int to the server, and block until we get confirmation of receipt
void sendDataToServerAndWaitForResponse(int payload);

And let’s say that we have a requirement to log once the payload was sent:

void sendToServerAndLog(int num) {
  sendDataToServerAndWaitForResponse(num);
  printf("This num was successfully received by the server: %i\n", num);
}

So here I’ve constructed an example where the value of num, which is available before we contact the server, is also used after we get a response from the server. I’ve simplified the example by ignoring failure cases. This is a special case of a common situation where some program state needs to be preserved across long-lived operations, such as a state machine or retry counter.

A problem with the above solution is that it blocks the thread. In terms of memory usage, we can think of it as requiring the memory of a whole call stack, which may occupy anything from hundreds of bytes to megabytes, depending on the system we’re running. And more importantly, we can’t use the thread for anything else while it’s blocked.

Using a callback

To avoid blocking the thread, we could instead consider using a callback function. Imagine that our hypothetical library supported this by taking a callback argument, rather than blocking until completion:

typedef void Callback();
void sendDataToServer(int payload, Callback* callback);

Now we can write our application code like the following, where we’ve split the behavior of sendToServerAndLog into two parts — one part before sending the data and one that’s the continuation after we’ve received a response:

int save_num;

void sendToServerAndLog(int num) {
  save_num = num;
  sendDataToServer(num, &continue_sendToServerAndLog);
}

void continue_sendToServerAndLog() {
  int num = save_num;
  printf("This num was successfully received by the server: %i\n", num);
}

In order to have the state of num available in the callback, we needed to persist it somewhere, so we need the save_num variable. This solution is much more memory efficient that the previous — if we’re running on a 32-bit platform, our app code occupies only the 4 bytes of save_num. The library code now also needs to preserve the callback pointer across the call, which uses an additional 4 bytes of memory. So this solution takes a total of 8 bytes of memory during the server operation, compared to using a whole thread and call stack of memory before. But there are still two issues with this design:

1. The save_num variable persists forever, even when we’re not sending data to the server. If the hypothetical library uses the same pattern to store the callback, then together the full 8 bytes are consuming memory forever.
2. We can’t have multiple calls to the server running in parallel here. If we call sendToServerAndLog a second time before the first response is received, the value of save_num is corrupted.

Adding context

A common pattern to get around the above problem is to have the library code accept an additional parameter that it passes back to our callback, which here we might name “context” because it represents the contextual information of the particular caller:

typedef void Callback(void* context);
// Whatever `context` is provided will be passed to the `Callback` untouched.
void sendDataToServer(int payload, Callback* callback, void* context);

Now we can use it like this:

void sendToServerAndLog(int num) {
  int* context = malloc(sizeof(int));
  *context = num;
  sendDataToServer(num, &continue_sendToServerAndLog, context);
}

void continue_sendToServerAndLog(void* context_) {
  int* context = (int*)context_;
  int num = *context;
  printf("This num was successfully received by the server: %i\n", num);
  free(context);
}

Now, if we call sendToServerAndLog multiple times, each call creates a distinct context that is completely independent of any other call.

Side note: Why make the context a void* instead of just an int, since we only need an int here? The reason is that we’re imagining here that sendDataToServer is part of a reusable library. Even if it’s a first-party library, it’s generally better practice to write it in such a way that it’s decoupled from the particular way you use the library. Making it a void* allows the user of the library to decide the size and shape of the preserved state, rather than coupling it to the library itself.

This is the same reason why we use a function pointer at all, rather than just hardcoding the library to invoke our particular callback directly: the library should not be coupled to the particular function that executes after it. And as the program evolves, we might land up calling it from multiple places that each require a different callback.

Applying the same pattern to sendToServerAndLog

We decided that a good interface for the library function sendDataToServer is one that takes a callback and a context, because the operation takes a long time, and a caller may want to continue doing something else after the operation completes. But similarly, sendToServerAndLog is also an operation that takes a long time, and its caller may also want to do something when the operation completes.

If we’re working with highly coupled code, then maybe we already know whether or not the caller of sendToServerAndLog needs to do anything else afterward, and exactly what it needs to do. But if we want sendToServerAndLog to be a reusable code that is decoupled from its caller, then we should probably have it accept its own callback and context from its caller. If we do this, then we need to persist the caller’s callback and context until the whole operation completes, so let’s upgrade our context to a struct that includes these fields:

// Context for sendToServerAndLog
typedef struct sendToServerAndLog_Context {
  int num;
  Callback* caller_callback;
  void* caller_context;
} sendToServerAndLog_Context;

void sendToServerAndLog(int num, Callback* caller_callback, void* caller_context) {
  sendToServerAndLog_Context* context = malloc(sizeof *context);
  context->num = num;
  context->caller_callback = caller_callback;
  context->context = caller_context;
  sendDataToServer(num, &continue_sendToServerAndLog, context);
}

void continue_sendToServerAndLog(void* context_) {
  sendToServerAndLog_Context* context = context_;
  int num = context->num;
  printf("This num was successfully received by the server: %i\n", num);
  context->caller_callback(context->caller_context);
  free(context);
}

Embedding the function pointer

Stylistically, it’s interesting to consider one further modification to this example. You may or may not agree with the following design change, but it leads nicely into the topic of closures in Microvium, which I’ll get to in a moment.

Rather than having the function save and copy around 2 pieces of state on behalf of the caller — the caller_callback and caller_context — we can combine these into one: we can just reference the caller_context and require that the first field in the caller_context is the callback function pointer, as in the following code. Also, rather than calling this a context, let’s now going to call it a closure, since it captures both a function pointer and some general state. The relationship between this and real closures will become more clear later.

typedef void ClosureFunc(Closure* closure);

// A general closure is expected to take this shape
typedef struct Closure {
  // Must be the first field
  ClosureFunc* invoke;

  /* ...other fields may follow...*/
} Closure;

// Here we have a specific closure shape for our function
typedef struct sendToServerAndLog_Closure {
  // Must be the first field. If you're using C++, you might instead 
  // inherit `sendToServerAndLog_Closure` from `Closure`
  ClosureFunc* invoke;

  // Other fields:
  int num;
  // Note now that we don’t need to store a separate caller_context
  // since both the context and the function pointer are combined
  // into the single closure struct.
  Closure* caller_callback;
} sendToServerAndLog_Closure;

void sendToServerAndLog(int num, Closure* caller_callback) {
  sendToServerAndLog_Closure* closure = malloc(sizeof *closure);
  closure->invoke = continue_sendToServerAndLog;
  closure->num = num;
  closure->caller_callback = caller_callback;
  sendDataToServer(num, closure);
}

void continue_sendToServerAndLog(Closure* closure_) {
  sendToServerAndLog_Closure* closure = (sendToServerAndLog_Closure*)closure_;

  int num = closure->num;
  printf("This num was successfully received by the server: %i\n", num);

  Closure* caller_callback = closure->caller_callback;
  caller_callback->invoke(caller_callback);

  free(closure);
}

This final design is quite clean:

• The memory for the transient state (e.g. num) is only allocated while the long-running operation is active. Remember that short-lived memory is cheaper memory.
• It doesn’t block the thread, so it’s easier to parallelize multiple operations if needed, without each operation consuming a whole thread of memory.
• Each layer is decoupled: sendDataToServer doesn’t need to know who its caller is, and similarly sendToServerAndLog doesn’t need to know who its caller is.
• The callback is neatly encapsulated into a single pointer value that can be passed around as a first-class value. If you’re familiar with C++, this is a similar benefit to using the std::Function<> type.

But there are some disadvantages to this design:

• Although it represents the same behavior as the first design (the synchronous code), the code is now a whole lot more complicated.
• Here we’ve only shown one closure signature. But what if we needed a return value to be passed to the closure? In general, each different return type is going to need its own type definitions for ClosureFunc and Closure, which will add up to a lot of boilerplate.
• The memory efficiency is not great because it uses malloc and free.
  • On my Windows machine with Visual C++, I measure malloc to have an overhead cost of 40 bytes per allocation (compiling for x86).
  • In FreeRTOS, each allocation has an overhead of 8 bytes on a 32-bit platform. With this figure, the closure in the example takes 20 bytes of heap space.
  • The heap in C/C++ can get fragmented, which costs additional memory.

Using Microvium Instead

We can write this same example in JavaScript using nested functions, as follows:

function sendToServerAndLog(num, caller_callback) {
  sendDataToServer(num, continue_sendToServerAndLog);

  function continue_sendToServerAndLog() {
    console.log(`This num was successfully received by the server: ${num}`);
    caller_callback();
  }
}

The nested function continue_sendToServerAndLog has access to variables in the outer function (in this case the parameters num and caller_callback). Here I tried to keep the function names consistent with the C example, but in practice, it may be more convenient to do the same thing using arrow function syntax, as follows:

function sendToServerAndLog(num, caller_callback) {
  sendDataToServer(num, () => {
    console.log(`This num was successfully received by the server: ${num}`);
    caller_callback();
  });
}

Either way, the values num and caller_callback are automatically captured into a closure on the JavaScript heap, making them available to the nested function automatically.

If you’re using the Microvium JavaScript engine, this created closure has a very similar structure in memory to the final example we did in C — it’s a single structure with a function pointer and two other variables. You may see now why I called the struct in the earlier C example a “closure”. The C code is a more explicit way of representing the same runtime structure, with similar benefits from a decoupling and modularity perspective, although clearly the JavaScript is more syntactically simple.

This closure heap allocation in Microvium will have the following characteristics:

• If the num is an integer in the range¹ -8192 to 8191, the closure occupies 8 bytes of memory, including a 2-byte allocation header, compared to the 20 bytes consumed by the C example on a FreeRTOS heap.
• There is no fragmentation overhead, since the Microvium heap is compacting.
• Allocating of the closure generally happens in constant time. Since the Microvium heap is contiguous, creating new allocations is similar to just bumping a free pointer forward.

Conclusion

We’ve walked through an example that’s representative of a common situation in program development, especially when networking is involved: network requests take time, and we can either block the current thread or we need another way to remember what we were doing so we can get back to it when the request completes. When writing your code in a modular and decoupled way, it’s better not to assume anything about the caller of your long-running application, so it’s better not to block the thread or hard-code anything about which callback to run or what state to hold onto.

In this case, Microvium actually offers you a way to make your code more memory efficient than the equivalent C code, while also making it easier to follow, and preserving the nice decoupling characteristics. Depending on your situation, this might make Microvium a good choice for orchestrating this kind of high-level program flow, especially when long-running tasks are involved and when you need to keep track of state across those tasks.

---

1. I’d say this is another advantage of using Microvium: numbers automatically grow in size as-needed. Integers in the range -8192 to 8191 use 2 bytes of memory ↩

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:

A 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. ↩

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 try…catch 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. ↩

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. ↩