With regards to 1), do not write/read structs directly to/from files. Instead write a proper serializer/deserializer. Without it, you may encounter another breakage soon when a different compiler/compiler options insert different struct padding bytes, which will then once again make your data non-portable, and a maliciously crafted save file with no length/size field validation on the deserializer level can lead to a variety of memory bugs.
struct layout is well specified, it should be possible to avoid any padding issues by just aligning and by padding (with dummy members) correctly. The problem in practice is mostly integer representation (big-endian vs little-endian).
Specified by whom? Not the C standard for sure. It is indeed soecified by individual ABIs, and ABIs don't tend to do anything too weird, but that's another question.
looks like I was wrong, but here is the de-facto standard I was relying on over the years ;-). Not that I've memcpied many structs to file directly btw. http://www.catb.org/esr/structure-packing/
FTA: I was serializing asset structs directly to disk (pak file) that had raw pointers in them
I’m surprised that that works in WASM. Wouldn’t a tiny change in your memory usage (say if you toggle your “log startup progress” flag) load data at a different address?
I love how WASM is the thing that finally blurred the line between Web and Native programming, formely two realms isolated from each other for a long time. This both develops better awareness of how the code is executed by the hardware, which JavaScript devs often lack, and also brings skilled folks from the Native platforms who seem to be not so against WASM as they were against JavaScript (and all other parts of the Web, really). Maybe this will bear fruit in that people will make more Native user interfaces again.
I wanted to love it. As someone who hasn't done any web stuff since I was a child, I thought it'd amazing for it to be "just another platform".
I'm a bit disappointed though:
* There's still no way to do DOM manipulation. So then it's tempting to just grab a canvas and draw everything yourself, which of course wreaks on things like accessibility. I'm no fan of the web, but at least it comes with a somewhat agreed-upon way to display graphical stuff – it's a bit of a shame if we're all gonna just treat it like a surface for pixels.
* WASI still leaves something to be desired. Why can't I have raw sockets and file access and stuff, in a POSIX-like way? I understand that sandboxing is important, so this can all be on a per-request-basis, but still. This "just another platform" is still too far from just that.
* The amount of JS glue needed to actually load WASM stuff in the browser is annoying. The idea of needing a bunch of magic "bundlers" is sad.
> WASI still leaves something to be desired. Why can't I have raw sockets and file access and stuff, in a POSIX-like way?
FWIW, that's exactly what they shipped first, with WASI preview 1 (wasip1). You can still use this today, and all runtimes with any level of WASI support will be able to run it.
If enough people adopt identical or similar js glue then they can use that for a new standard. If people dont care about a standard interface then why both creaing a new standard? Look what happened with jquery selectors and ajax. People loved it and it became the new standard built into browsers.
You can call JS in which you can manipulate the DOM.
Of course architecturally (also regarding your file access) it's better to use the wasm for logic as much as possible where the web (HTML/JS) provides the UI and IO, data flows into wasm for work and results flow back to the web.
This also has the benefit that you can keep your original C/C++ source code much more platform agnostic which helps reusability and testing.
> You can call JS in which you can manipulate the DOM.
Well sure. But for me, the promise of WASM was to make the browser "just another platform". Now it's "this special platform where you have to access some of the most important functionality through FFI interop with a very high-level, very opinionated language".
> Of course architecturally (also regarding your file access) it's better to use the wasm for logic as much as possible where the web (HTML/JS) provides the UI and IO, data flows into wasm for work and results flow back to the web.
OK, but like, I wanted the browser to be "just another platform". I don't want to use JS, and I consider HTML orthogonal to my logic. I realize that's not where we're at, but that's what I dreamt of. Hence my disappointment. Which is OK, I don't matter :)
> This also has the benefit that you can keep your original C/C++ source code much more platform agnostic which helps reusability and testing.
Hmm well I guess I don't quite get what counts as "just another platform." Surely every platform is going to have the native APIs that you need to abstract over. Why is WASM different?
Is it just a matter of WASM being too new to have full featured wrappers and APIs for your language of choice?
There's no way to draw on a canvas in WASM either. You just decided to write JS wrapper functions for that. But you didn't write wrapper functions for DOM manipulation.
1: Letting your code break on pointer size changes is a quite bad sign imho (it's a sign that many other things are probably done with aliasing,etc and has a high risk of breaking due to undefined behaviour once gcc/clang gets around to utilizing it for an optimization).
2: iirc WASM was initially designed to be shimmable via Asm.JS to force laggards(Apple, Google) to implement it, Asm.JS in turn relied on specific rules in JS to get reliable 32bit arithmetic (but impossible for 64bit).
Wasm64 is implemented and works in Chrome and Firefox.. Apple is lagging again with Safari.
1: True, although it also limits the addressable memory and the typical 4GB limit seems less these days. I’m thinking of large apps like Figma running in the browser.
2: Will existing 32-bit WASM binaries break on WASM64 engines or does the binary have a flag for compatibility?
1: Something like Figma could probably offload some of the memory pressure to GPU textures. (But they'd probably run into safety browser limits before that).
2: Most runtimes are 64bit already, A runtime detecting a wasm32 binary will just continue to generate code with the current JIT compiler whilst WASM64 will require another JIT (and perhaps memory system since WASM32 runtimes are often based on "hacks" where 4gb of address space is reserved but not given real memory so that the JIT compiler gets an easier job without security implications).
I believe 32-bit was chosen partially due to implementation efficiency reasons. It makes sense because you can allocate a 4GB mapping, so there is no need for a second software virtual memory layer. Also perhaps they internally require tagged pointers, which are much cheaper, especially if aligned, if the pointer is only 32 bits
32 is better for a lot of things like simd. the strength of it is wasm can do both types now and js can't unfortunately. a number in js is strictly 64.
It's not 100% better, but it cuts out a few tokens which helps readability and moves the significant asterix further left where I think it's easier to spot.
Maybe I find this harder to parse because I'm not used to sizeof without brackets (though I know it's valid). But I think the bigger deal is that your version has a bug if the star is missing whereas there's has a bug if the star is present; it's easier to spot something extra than it is to spot something missing.
It's slower. Wasm32 can just reserve 8 GiB (32-bit pointer + 32-bit offset) of the virtual address space from the OS for each memory, so checking for out-of-bounds memory accesses imposes no performance penalty. Wasm64 can't do that, so each memory access is a bit slower.
Sometimes I wonder whether it's possible to run the wasm code in a separate sandboxed process to eliminate a lot of checks. I mean optionally, because normally JS calls wasm code synchronously in the same address space. The bridge will add more latency when there is a transition between JS and wasm. It's obviously complicated because some data structures can also be shared, such as SharedArrayBuffer.
Oh that's interesting, never noticed it in my experience but I have never written anything in wasm where it would matter. Makes perfect sense now that I think about it though. Thanks!
You don't need 4GB and it wastes memory to make pointers twice as big? Even Linux supports running 64-bit code in a 32-bit address space ("x32 ABI") for this reason.
I've been porting Micropolis (SimCity Classic) to WASM / WebGPU / Svelte 5. Emscripten + Embind compile the C++ engine and glue it to TypeScript/Svelte/Runes/Reactivity; TypeScript owns UI, rendering, and callback handlers.
I agree with the article's main lessons: wasm32 pointer size, don't serialize structs with pointers, debug native 32-bit when you can, WebGL/WebGPU is stricter than desktop GL, Emscripten export flags still bite. I hit some of the same categories; the parts that were actually tricky for Micropolis are below.
Svelte 5 runes ($state, $derived, etc.) work in plain .ts modules, not just .svelte templates. That matters because the WASM bridge is a reactive module the HUD, command bus, and Vitest all import -- not a component-only trick. The file has to be MicropolisReactive.svelte.ts so runes compile under the same Vite/SvelteKit pipeline as the app; plain .ts breaks in Node with "$state is not defined".
Embind API surface -- what to expose and what to leave out:
// This file uses emscripten's embind to bind C++ classes,
// C structures, functions, enums, and contents into JavaScript,
// so you can even subclass C++ classes in JavaScript,
// for implementing plugins and user interfaces.
//
// Wrapping the entire Micropolis class from the Micropolis (open-source
// version of SimCity) code into Emscripten for JavaScript access is a
// large and complex task, mainly due to the size and complexity of the
// class. The class encompasses almost every aspect of the simulation,
// including map generation, simulation logic, user interface
// interactions, and more.
The comments in that file go on to describe the strategy for wrapping: Core Simulation Logic, Memory and Performance Considerations, Direct Memory Access, User Interface and Rendering, Callbacks and Interactivity, and Optimizations.
The engine callback virtual interface bridged C++ to JS via JSCallback:
In the old NeWS/Hyperlook, TCL/Tk/X11, SWIG/Python/PyGTK, and SWIG/Python/TurboGears/AMF/Flash versions, this callback interface used to be a stringly typed general purpose event callback interface, which I tightened up into a strict C++ interface and corresponding typescript interface, so embind could help me integrate it safely and cleanly with TypeScript and Svelte Runes.
TypeScript handlers that update rune-backed state (sendMessage, didTool, budget hooks, etc.):
The pattern: C++ fires callbacks with enough context for the UI; TS updates $state; components read micropolisReactive (peek / poke / memory / getSnapshot) instead of calling Embind or touching HEAP* directly. That is where the rubber hits the road for interactivity.
Heap access is its own footgun. Emscripten may expose Module.wasmMemory, HEAPU16, or neither until init; some getters throw if you read too early. Centralized helper:
Map rendering: WebGPU tile renderer with canvas fallback (legacy WebGL frozen, now reimplementing in WebGPU). The renderer reads 16 bit flags + tile indices from direct simulator memory views into WASM linear memory (mapData / mopData), not per-frame Embind copies.
City saves are a defined binary format (.cty), not fwrite of engine structs. Live map data is views into WASM linear memory (mapData / mopData), not embedded native pointers -- same idea as the article's side-table fix, but that is how this codebase is already structured.
Why I find this stack interesting: original SimCity engine lineage, narrow Embind surface on purpose, reactive TS facade so automation and UI share one sim without reviving the old Python/SWIG/pyGTK path. Sprites (trains, choppers, generic orange monsters wrecking chaos and havoc -- definitely not Godzilla [TM], but possibly Trump adjacent) simulate in C++; compositing them in the WebGPU path is still work in progress.
The WebGPU renderer is being built as a general stack with pluggable layers, including Sims content rendering (characters, animations, terrain, objects, walls, floors, ui effects, etc).
You really need a serializer for this sort of thing because it can also include forwards compatibility of your data structures.
I’m surprised that that works in WASM. Wouldn’t a tiny change in your memory usage (say if you toggle your “log startup progress” flag) load data at a different address?
UCSD Pascal:
https://archive.org/details/UCSD_Pascal_1.1_1
Wizardry:
https://archive.org/details/WizardryProvingGrounds
https://en.wikipedia.org/wiki/SWEET16
I'm a bit disappointed though:
* There's still no way to do DOM manipulation. So then it's tempting to just grab a canvas and draw everything yourself, which of course wreaks on things like accessibility. I'm no fan of the web, but at least it comes with a somewhat agreed-upon way to display graphical stuff – it's a bit of a shame if we're all gonna just treat it like a surface for pixels.
* WASI still leaves something to be desired. Why can't I have raw sockets and file access and stuff, in a POSIX-like way? I understand that sandboxing is important, so this can all be on a per-request-basis, but still. This "just another platform" is still too far from just that.
* The amount of JS glue needed to actually load WASM stuff in the browser is annoying. The idea of needing a bunch of magic "bundlers" is sad.
FWIW, that's exactly what they shipped first, with WASI preview 1 (wasip1). You can still use this today, and all runtimes with any level of WASI support will be able to run it.
Of course architecturally (also regarding your file access) it's better to use the wasm for logic as much as possible where the web (HTML/JS) provides the UI and IO, data flows into wasm for work and results flow back to the web.
This also has the benefit that you can keep your original C/C++ source code much more platform agnostic which helps reusability and testing.
Well sure. But for me, the promise of WASM was to make the browser "just another platform". Now it's "this special platform where you have to access some of the most important functionality through FFI interop with a very high-level, very opinionated language".
> Of course architecturally (also regarding your file access) it's better to use the wasm for logic as much as possible where the web (HTML/JS) provides the UI and IO, data flows into wasm for work and results flow back to the web.
OK, but like, I wanted the browser to be "just another platform". I don't want to use JS, and I consider HTML orthogonal to my logic. I realize that's not where we're at, but that's what I dreamt of. Hence my disappointment. Which is OK, I don't matter :)
> This also has the benefit that you can keep your original C/C++ source code much more platform agnostic which helps reusability and testing.
It feels the opposite to me.
Is it just a matter of WASM being too new to have full featured wrappers and APIs for your language of choice?
At any rate: this doubly makes my point.
> Web is 32-bit. Your 64-bit structs will break. This was the root cause of most of my bugs. WASM is 32-bit address space, pointers are 4 bytes not 8.
2: iirc WASM was initially designed to be shimmable via Asm.JS to force laggards(Apple, Google) to implement it, Asm.JS in turn relied on specific rules in JS to get reliable 32bit arithmetic (but impossible for 64bit).
Wasm64 is implemented and works in Chrome and Firefox.. Apple is lagging again with Safari.
1: True, although it also limits the addressable memory and the typical 4GB limit seems less these days. I’m thinking of large apps like Figma running in the browser.
2: Will existing 32-bit WASM binaries break on WASM64 engines or does the binary have a flag for compatibility?
2: Most runtimes are 64bit already, A runtime detecting a wasm32 binary will just continue to generate code with the current JIT compiler whilst WASM64 will require another JIT (and perhaps memory system since WASM32 runtimes are often based on "hacks" where 4gb of address space is reserved but not given real memory so that the JIT compiler gets an easier job without security implications).
The real mistake is requiring pointer to be 64 bit when most programs don’t use it.
Since this is one of the bugs, I always recommemd writing
Like this instead: It's not 100% better, but it cuts out a few tokens which helps readability and moves the significant asterix further left where I think it's easier to spot.But ACSHUALLY, how you write allocation is like this
The kernel people seem to finally have figured out this one in 2026.I like the word "everybug" :-D
I don't think that ever had much, if any, adoption and it looks like it will be removed in the next few releases.
[0] https://soft.vub.ac.be/Publications/2022/vub-tr-soft-22-02.p...
[1] https://www.usenix.org/system/files/sec20-lehmann.pdf
I agree with the article's main lessons: wasm32 pointer size, don't serialize structs with pointers, debug native 32-bit when you can, WebGL/WebGPU is stricter than desktop GL, Emscripten export flags still bite. I hit some of the same categories; the parts that were actually tricky for Micropolis are below.
Svelte 5 runes ($state, $derived, etc.) work in plain .ts modules, not just .svelte templates. That matters because the WASM bridge is a reactive module the HUD, command bus, and Vitest all import -- not a component-only trick. The file has to be MicropolisReactive.svelte.ts so runes compile under the same Vite/SvelteKit pipeline as the app; plain .ts breaks in Node with "$state is not defined".
Embind API surface -- what to expose and what to leave out:
https://github.com/SimHacker/MicropolisCore/blob/main/packag...
The comments in that file go on to describe the strategy for wrapping: Core Simulation Logic, Memory and Performance Considerations, Direct Memory Access, User Interface and Rendering, Callbacks and Interactivity, and Optimizations.The engine callback virtual interface bridged C++ to JS via JSCallback:
https://github.com/SimHacker/MicropolisCore/blob/main/packag...
In the old NeWS/Hyperlook, TCL/Tk/X11, SWIG/Python/PyGTK, and SWIG/Python/TurboGears/AMF/Flash versions, this callback interface used to be a stringly typed general purpose event callback interface, which I tightened up into a strict C++ interface and corresponding typescript interface, so embind could help me integrate it safely and cleanly with TypeScript and Svelte Runes.
TypeScript handlers that update rune-backed state (sendMessage, didTool, budget hooks, etc.):
https://github.com/SimHacker/MicropolisCore/blob/main/apps/m...
Simulator attach/detach, singleton engine load, wiring JSCallback into Micropolis:
https://github.com/SimHacker/MicropolisCore/blob/main/apps/m...
The pattern: C++ fires callbacks with enough context for the UI; TS updates $state; components read micropolisReactive (peek / poke / memory / getSnapshot) instead of calling Embind or touching HEAP* directly. That is where the rubber hits the road for interactivity.
Heap access is its own footgun. Emscripten may expose Module.wasmMemory, HEAPU16, or neither until init; some getters throw if you read too early. Centralized helper:
https://github.com/SimHacker/MicropolisCore/blob/main/apps/m...
Bridge design, Vitest against real WASM, teardown order with Embind lifetimes:
https://github.com/SimHacker/MicropolisCore/blob/main/docume...
Map rendering: WebGPU tile renderer with canvas fallback (legacy WebGL frozen, now reimplementing in WebGPU). The renderer reads 16 bit flags + tile indices from direct simulator memory views into WASM linear memory (mapData / mopData), not per-frame Embind copies.
https://github.com/SimHacker/MicropolisCore/blob/main/packag...
https://github.com/SimHacker/MicropolisCore/blob/main/docume...
City saves are a defined binary format (.cty), not fwrite of engine structs. Live map data is views into WASM linear memory (mapData / mopData), not embedded native pointers -- same idea as the article's side-table fix, but that is how this codebase is already structured.
Why I find this stack interesting: original SimCity engine lineage, narrow Embind surface on purpose, reactive TS facade so automation and UI share one sim without reviving the old Python/SWIG/pyGTK path. Sprites (trains, choppers, generic orange monsters wrecking chaos and havoc -- definitely not Godzilla [TM], but possibly Trump adjacent) simulate in C++; compositing them in the WebGPU path is still work in progress.
The WebGPU renderer is being built as a general stack with pluggable layers, including Sims content rendering (characters, animations, terrain, objects, walls, floors, ui effects, etc).
Character animation demo:
https://vitamoo.space
VitaMoo code:
https://github.com/SimHacker/MicropolisCore/tree/main/packag...
Unified WebGPU Renderer:
https://github.com/SimHacker/MicropolisCore/blob/main/docume...
Render Core Package:
https://github.com/SimHacker/MicropolisCore/blob/main/docume...
Renderer Plugin Roadmap:
https://github.com/SimHacker/MicropolisCore/blob/main/docume...
Live Micropolis tile renderer and simulator demo (no other ui yet, work in progress):
https://micropolisweb.com
Demo of the simulator, cellular automata, and tile engine to Jerry Martin's music:
https://www.youtube.com/watch?v=319i7slXcbI
Repo:
https://github.com/SimHacker/MicropolisCore