Archives / 2015 / June
  • WebAssembly – A new Hope

    Recently, WebAssembly was announced, a new standard that defines a binary format and execution model for the Web. With Google, Mozilla, Microsoft and others voicing their support, my first thought was: Wow, so there is actually hope to get rid of JavaScript someday!

    In this blog post I’ll explain why I think WebAssembly is such a big step in the right direction, even if it’s something that won’t have an immediate impact.

    What’s the problem with JavaScript, anyway?

    JavaScript is a programming language that has quirks (see Seven JavaScript Quirks I Wish I’d Known About or JavaScript garden), lacks important features (e.g. proper arithmetic data types) and isn’t type safe (by design).

    All this doesn’t matter for JavaScript’s original purpose – writing a few lines of script code. In fact, the dynamic nature of JavaScript, i.e. being able to add object properties or change data types on the fly, enables some pretty elegant solutions.

    But writing large amounts of JavaScript in a team of average developers comes with a risk that is higher than in other languages. Bugs that the compiler for a typed language would catch immediately can remain unnoticed for a long time. Renaming identifiers (via find-and-replace in a text editor) or refactoring code can be downright dangerous. Anders Hejlsberg puts it this way: “[…] you can write large programs in JavaScript. You just can’t maintain them.”.

    So why are people using JavaScript?

    • JavaScript in the browser (used in conjunction with other web technologies) enables highly responsive, cross-platform, zero-deployment applications…
    • without any additional client installation (most would consider their favorite non-IE-browser as part of the “basic operating system setup” before any work can be done on a new computer) …
    • … and is good enough for developing almost any kind of application – at least given enough time, budget and/or determination.

    The problem: this “good enough” is an enormous hurdle for any alternative technology. At this point in time, it’s totally unrealistic that a completely new technology will magically appear and win the uphill battle to be available on all platforms, with the same reach as JavaScript-enabled web browsers.

    Fact is, JavaScript (plus other web tech) won. Some people are perfectly fine with this, others (me included) are a bit skeptical what this means for the future of software development –  to put it mildly.

    Is JavaScript the QWERTY of programming languages?

    I started programming in 1983 and since then I’ve witnessed incredible progress. From writing Z80 code that fit into a few kilobytes of RAM to simply using a component without caring, or even knowing, what language it was written in. And then, at some point, writing lots of code using a scripting language (using a document markup language to create application UIs, but that’s another story) became state-of-the-art. I thought we’d be much further in 2015.

    Ok, if replacing the JavaScript ecosystem with something completely new isn’t a realistic option, what can be done to improve the current situation for software developers?

    Option #1: Make JavaScript better

    JavaScript can be improved if all major players agree on the how, when and what. This can happen (see ECMAScript 2015), but sometimes it doesn’t (see ECMAScript 4). Improving a standard takes time, and it’s definitely not a playground for “let’s try this crazy idea and see how it works out when all browser have it”. Plus, some language characteristics simply cannot be changed without breaking a lot of existing code.

    Of course, this “you want to turn X into Y, but you cannot change X and don’t want to start from scratch” is nothing new in the world of programming languages. It’s a recurring pattern that if people have an idea for a programming language [feature] and are not in the position to write a full-blown compiler/interpreter and the matching runtime environment, they choose the route of more or less advanced a preprocessor.

    Which is exactly what happened with JavaScript, too.

    Option #2: Compile-to-JavaScript

    There are quite a few source-to-source compilers (see List of languages that compile to JS). These “transpilers” let you write programs in one language and translate the source code to another – JavaScript.

    In the early days, debugging was kind of painful, because you had to debug the resulting JavaScript. Now that browsers support source maps, the debugging story is straightforward: You step through the original source code, and the debugger keeps track of where program execution is in the JavaScript code behind the scenes.

    Note that source maps are not magic; they can tell the JavaScript debugger the original filename, line and column. If you want on-the-fly inspection of variables, the source language shouldn’t be too far away from JavaScript and its concepts (see the article “Beyond Source Maps” for more on source maps).

    TypeScript is a language that is close to JavaScript by design, extending it e.g. by offering optional typing. The TypeScript compiler translates the source code to idiomatic JavaScript code, i.e. code that a JavaScript developer following certain patterns would write by hand. TypeScript improves the developer experience, but cannot overcome inherent limits of JavaScript. The performance is determined by how fast the JavaScript engine executes JavaScript, obviously, and you still don’t get better (native) arithmetic data types.

    asm.js is “a strict subset of JavaScript that can be used as a low-level, efficient target language for compilers” (quote from the spec). So this is not a language intended to be written by human developers. Instead, a compiler like emscripten compiles a source language to “JavaScript-as-assembly” code. The code still runs in any browser, but for good performance you need an engine that either knows about asm.js or at least can achieve good performance from the JavaScript code that tries to avoid type conversions and garbage collections.

    There are a couple of things to keep in mind regarding asm.js:

    • Currently it is intended for statically-typed languages like C/C++ that use manual memory management. For other languages, you either have to translate the source to e.g. C first, or even translate the whole interpreter/runtime environment.
    • Browser vendors have to adapt their JavaScript engines for significant performance gains.
    • The resulting JavaScript code can be pretty large, so parsing the code on the client becomes a bottleneck (sending the code over the write can be an issue in certain scenarios, too).
    • Debugging is a problem, because the information stored in source maps is not sufficient.

    Finally, what’s WebAssembly about?

    WebAssembly can be thought of as an evolution of asm.js, to have “something” that can execute low-level code which can be optimized and precompiled. The central idea is transport the code in a binary format instead of (mis-)using JavaScript. This results in smaller file sizes and removes the parsing of source code on the client (good for mobile or IoT).

    This “something” will be part of JavaScript engines some day. Part of the high-level goals is to have a polyfill at least for the first step, the minimum viable product, which enables the MVP to run on existing browsers, but I’m not sure how important this will be in practice.

    It is important to note that at first, WebAssembly won’t be a general purpose compilation target for languages other than C/C++. The idea is to work incrementally, gain experience and collect feedback. The long-term vision does contain support for garbage-collected languages, though. Looking at the long list of To-dos it becomes clear that a lot of work is waiting to be done (including better source maps).

    Does WebAssembly have a chance?

    Judging from the collaboration of the big players alone, I’d say yes. Also important: Both the vision and the incremental approach to get there sound reasonable, clearly trying to avoid XHTML 2‘s “death by overambitious design”. WebAssembly is not about throwing away or replacing JavaScript engines. WebAssembly will be added to the engines, which helps interop and means that low-level infrastructure can be reused.

    Again, things will take time, and the first versions will not be suitable for all scenarios. In the long run (we’re talking years, not months), though, WebAssembly is very promising. If you have a general purpose engine/virtual machine in every browser, developers are free to use whatever language they want because the language matters only at compile time, not at runtime.

    And this is the point where we’ll be able to get rid of JavaScript.