Maud: A Rust macro for writing HTML
maud.lambda.xyz
maud.lambda.xyz
Scala is the only other language I know of which has some kind of XML/HTML syntax support (scala.xml). (But it looks like it got removed. [2])
XHP was launched in 2010 (https://www.facebook.com/notes/10158791323777200/)
The first version of FaxJs (the precursor to React) was launched in 2011, being directly inspired from XHP (https://github.com/jordwalke/FaxJs)
React was made in 2012 by the same person who made FaxJs, taking the best ideas from FaxJs and creating React.
So yeah, JSX actually comes from the idea of XHP :)
The concept was popular at Facebook, so it made its way into HPHP, HHVM, and later ReactJS.
I find JSX extremely hard to read and work with and reason about.
Small example for people to understand the difference:
// HTML & JSX
<div id="person">
<h1>Niko</h1>
</div>
// Hiccup
[:div#person
[:h1 "Niko"]]In JSX you might be struggling with expressing certain things, because JS has a statement expression syntax. So you have to circumvent that. In Clojure everything is an expression already.
In JSX you can manipulate components with HoCs. In Clojure it’s just data, no special ceremony needed.
On the client, if you’re using React, you might use hooks to keep travk of state, which are a special construct that don’t exist in JS. In Clojure, if you use something like reagent, you just use atoms, they have a special purpose implementation, but they look and feel like regular old atoms.
However I have to say, JSX and React are very productive even with those limitations, simply because the tooling around it is so good.
This is a very common confusion because most people approach JSX via React, and as such never really grasp that JSX is a separate, independent thing. If you're interested in learning it standalone, I'd highly recommend working with either the Typescript `jsxFactory` compiler option, or the `@babel/plugin-transform-react-jsx`
The "opposite" approach being using a template language like the one in Svelte or Angular… or like PHP (or Velocity, or Jinja) (for me JSX is not a template syntax at all).
I think template languages are also prone to rendering / logic mix, I don't know what is an elegant solution to this problem apart from getting used to this reality. Maybe something like QML, using data models, instead of having a for loop to render lists for instance.
Yes. Which makes any issues the gp mentioned entirely up to the implementation of that function call.
> seems like everything the parent said could apply to any use of JSX, except for the attribute naming issue
If by "could apply", you mean anyone could write a function that has those issues, then yes I guess. My point was that if you want to avoid those, they're in no way inherent to JSX. Is that what you meant?
I mean, I guess "blurred line between rendering and code" is less about the function implementation, and more about project / file organisation choices (including conventions of a frontend view framework community), but it's still certainly not inherent.
> I think template languages are also prone to rendering / logic mix
True. I think you do need a lot of discipline if you want to keep a clear and consistent separation, and it's also not always worthwhile (there's a codebase-approachability trade-off with abstraction & locality of context).
For arguments' sake though here is a quick-and-horribly-dirty example of JSX syntax separation (not recommended but demonstrative): https://codepen.io/lucideer/pen/jOZvwVz?editors=0010
In contrast with HTML, where there is no way one can mix the logic and the UI because it is syntactically impossible. You have to get the DOM element from a script and to fill it data (well, actually, you can with the script tag and on* attributes, but let's consider HTML without them, only allowing script tags in <head> for instance so our restricted version of HTML remains useful).
I think you need the rendering to be data, not code, to "forbid" this mix (and soon enough, you reinvent some kind of lisp I guess)
That's fair. If that's what you're looking for, then JSX is not the answer. I don't think we disagree though - I was merely pointing out separation is possible (even if not strictly enforced).
Personally, I am a fan of the component-driven model (criticised above for "blurring the lines"). As I mentioned, it's a trade-off, and I think any potential "blurriness" is far outweighed by the context and reduced abstraction wins from a code-readability & codebase approachability perspective. Heavily separated templating efforts often leads to codebases with deeply nested hierarchies for views -vs- controller logic that require having 4+ split pane editors open just to understand extremely local snippets of data flow. This is particularly egregious when e.g. trying to audit codebases for injection mitigation.
Think it may have being easier if JSX is just included in ES as a standard. So yes and no, JSX introduced by FB via react is a just a syntax sugar, however the original JSX is a meant to be a ES standard, with each node as a tree structure.
https://www.ecma-international.org/publications-and-standard...
E4X was supported by Firefox for a while, but I think support was removed.
For instance classes in JS are sugar syntax to define a function that defines fields in this and prototype members of this function.
Arrow functions are sugar syntax to define functions bound to a specific this.
async / await is sugar syntax for promises.
for..of is sugar syntax for iterating over an iterable.
The attribute renaming is a React feature. JSX doesn't rename attributes. See here for example https://codepen.io/lucideer/pen/jOZvwVz?editors=0010
> only expressions in brackets but full function calls in loops (map)
Not sure what you mean here. Function calls are expressions - they work bare in brackets & also as method args (like the arg passed to the map method on an array prototype, which is called while "loop"-ing that array). This is all Javascript functionality, not really specific to JSX in any way?
The problem with "only expressions in brackets" is that then you can write code in the template, but it's massively different to regular code. Worst of both worlds: code in templates and weird subset of code.
If this is undesirable as a feature, then yes, JSX is not an appropriate choice. JSX is the furthest thing from logic-less templating. But that's a deliberate design decision.
> it's massively different to regular code
It's not. It's just regular code - there are no differences.
> weird subset of code
It's not weird. It's any JS expression. All JSX brackets are values, and JS statements don't have return values, so embedding a JS statement within a JSX value would serve no purpose. What would it do?
All statements in JS can be expressed as expressions directly (ternary/array methods) or indirectly (nested within a function expression), so there's no loss of functionality - it's not even really a subset in that sense.
> All statements in JS can be expressed as expressions directly (ternary/array methods) or indirectly (nested within a function expression), so there's no loss of functionality - it's not even really a subset in that sense.
Do you not see the contradiction?
Vanilla javascript assignments can be any expression.
JSX assignments (the brackets) can be any expression.
There's no weirdness nor extra limitations to JSX assignments compared to normal vanilla JS. It works exactly the same.
If you think not being able to assign JS blocks is weird you're going to have to explain to me what this piece of JS code would do:
const val = if (x < y) { func(); } <div>{if (list?.length > 0) {
list.map(e => {
if (e.items.length) {
<div>Excellent!</div>
} else {
<div>Not good.</div>
}
});
} else {
<div>Nothin to see.</div>
}}</div>
<div>{(!!list && list.length) // awkward way of writing code
? list.map(e => {
if (e.items?.length) { // normal way of writing code
return <div>Excellent!</div>;
} else {
return <div>Not good.</div>;
}
})
: <div>Nothin to see.</div>
}</div>
https://developer.mozilla.org/en-US/docs/Mozilla/Firefox/Rel...
Had E4X been more widely released, I might have used it, as might many others. In retrospect, though, it's probably good that it didn't catch on, as it introduced a lot of extra syntax and operators, like a.@b, a.@*, a.*, a.b::c, a..b, a.(@b==x) and had semantics that affected large parts of the JS Runtime (how the delete operator works, how calling works, what += means, and so on).
Somehow JSX seems to be "worst of both worlds" to me.
Some of these HTML-templates-as-code libs (like Kotlinx.html) are generated from a formal HTML specification. This is really nice, and makes it more type safe (only allow tag nesting that is allowed, and only allow attributes that are allowed; ofcourse with escape hatches).
Visual Basic .NET has XML literal support [1].
[1]: https://docs.microsoft.com/en-us/dotnet/visual-basic/program...
\<h1>Hello, world!\</h1>
\<img alt="Hello, world!"/>
\<h1>Hello, {place}!\</h1>
\<img alt={message}/>
\<img alt="Hello, {place}!"/>
<img alt={"Hello, " + place}/>
<img alt={`Hello, ${place}`}/>Mara's whichever_compiles! macro for example: https://github.com/m-ou-se/whichever-compiles -- that macro is forking your compiler to try out all the branches and throwing away branches which caused a compile error.
Clearly your IDE should throw its hands up and say, I don't understand what this does, I give up.
In general doing something useful with Rust macros is a more tractable problem for an IDE than say the C pre-processor, because Rust's macros have a stronger syntax, but the proc macro is potentially much too powerful / dangerous to try to evaluate.
https://rust-analyzer.github.io/blog/2021/11/21/ides-and-mac...
I don't remember experiencing Rust Analyzer crashes for a long time, though (I do remember some in the past).
The declarative macros aren't too head-twisting for tools, they just expand as declared, this can sometimes have a few surprising effects but generally it is very manageable. Procedural macros have essentially unlimited power and thus are sometimes entirely impossible to analyse.
I like, use, and write macros, but this isn't fully true. For example, recent Rust versions allow writing identifiers directly in the format string:
println!("{some_identifier}")
[1]: https://github.com/rust-lang/rust-analyzer/issues/11260#
But once you start getting to a few pages it's so much less cognitive overhead to separate concerns out into proper html files/templating.
It seems quite daunting to have a large project where all your html is inside rust files and a lot harder to find people who can work on it proficiently.
Plus, Maud has a definitive speed advantage from what I've noticed.
Sadly, there isn't any great alternative for making performant HTML templates in Rust from what I've seen, a lot of the other HTML templating libs aren't that great comparatively.
Which others have you tried, and why do you think they're not great?
Horroshow, ructe, typed-html and yarte were okay but inheritance was underdocumented (or generally not as well documented), which is a big no for me. And a few only give minor-quality help at compile time.
Maud on the other hand has decent compiletime help and doesn't require a lot of documentation as normal rust code is doing the heavy lifting.
My main worry on things is that I'd rather like being able to express the inheritance in the type system through generics. Ie, having a top level "Page<T>" struct in which I can start putting the various subparts needed to render a page. It seems askama is doing it the other way round with a "parent" member needed in the child page.
I'd say it's actually a disadvantage. Front end engineers will work with HTML with embedded logic (as little as possible - values, iterators, conditionals). Generating HTML from Rust code is nice for simple things, but won't work with complicated HTML templates when those are required and will be harder to integrate with front-end applications. It's possible to compile a template with embedded logic (Java Server Pages did that in the late 1990's) to performant and type safe code. An approach like that would be great for Rust (someone must have done it by now)
And I dispute it not working with complicated HTML templates, that's handling it fine and integrates very well since I share code between frontend and backend anyway.
So even if it's not the "optimal approach" as you title it, it's the ergonomic approach for me because it doesn't get in my way.
https://djc.github.io/askama/creating_templates.html#the-tem...
Regardless, it's good to see some variety and experimentation like Maud in software design
Ideally, I'd imagine just making a "Global<T>" struct, where I insert the current page context like so "Global<LoginPage>" to get the final template. Though by the looks, it sounds more like LoginPage will simply have to have all the global context available via some attribute like "LoginPage.global".
But as someone who works on a sprawling set of template files with loads of sprinkled conditionals in a template language... the idea of using a "real language" for the templating, and in particular having type checking and proper abstraction tools (you can just write functions! No weird template tag tricks) is so much nicer to me.
HTML templates have always felt super brittle to me, basically at the level of C preprocessor programming. Even just with React/TS you can get away with so much with way more confidence given the typed backing.
Though I think that you _could_ get away with some nice tricks in Rust thanks to macros, where you ingest the template files and generate render functions with all the properties... or something like that.
All these opening and closing quotes are pointless. Why don't write a single string with the verbatim html inside?