> *It's NOT a proposal to incorporate JSX into the ECMAScript spec itself.* It's intended to be used by various preprocessors (transpilers) to transform these tokens into standard ECMAScript.
because react decided to return JSX like element syntax instead of being simply a function returning a dom element doesn't mean that this should be standardized.
The argument to standardize it isn't really "because React does it", it's "because a ton of developers find it really useful". That seems kind of undeniable to me, even as someone that dislikes a lot of React.
I don't think that it is sound to incorporate a DSL that imitates html only approximately into a language. That should remain an external library at best.
And that's irrespectively of my view on jsx, the latter which I find non ergonomic and confusing.
That many people are being force fed it to the point of familiarity does not mean that it is the best DX.
we went from (html + js> to html in js and js in html. Quite very confusing to me. Knotty (naughty) complexity pun intended.
It's more akin to stockolm syndrom to not improve on the statu quo at this point.
To provide a quick feedback, I believe the 'without jsx' version could have been improved.
One advantage it already has is that it doesn't need the repetition of /closing> elements.
But it doesn't go far enough in terms of function composition to pass arguments. The signature (that resembles react.createElement I guess) is too cumbersome.
That creates a succession of braces and comma which is unfortunate.
Just to specify some object attributes. That could be easily a function Attr().
We have higher order functions easily in javascript. We should make use of it.
That eliminates html looking jsx since we just use functions. That's less confusing than pseudo html syntax.
True. Then can you suggest a standardized ECMAScript JSX proposal that doesn't turn this into a "children" key on an object? Would you just transform it into an array? ["foo", { bar: qux }, "baz"] ?
if it returned dom elements it should not be standardized either. JS is supposed to move forward with clean abstractions, not blindly inherit things from the web that happen to be convenient to some people.
My proposal transforms <foo bar={qux}>bla</foo> into { [Symbol.for('jsx')]: 'foo', bar: qux, children: "bla" }
It's self-contained, generic, doesn't rely on imports or globals, and avoids tag key collisions. It's the only way I can imagine it ever being standardized.
I think it is still useful as a function call syntax. There's a variety of data structures that people convert JSX to already, including direct-to-DOM. Choosing a blessed one seems harder than a function calling convention, and a function calling convention resembles other language things like tagged templates.
I agree that the current "auto-imports" to find that function are nonsense and far too React specific. But the current "global" approach isn't actually "require a global" it is "requires a variable in scope" so it works with "Bring-Your-Own-Import" just fine. We just need a better standard for what that BYOI function is called by default. `React.createElement` is obviously silly. I've been happy in my own projects standardizing on `jsx` as the function name. (It resembles the auto-imports, too, even if I still don't understand why React thought it needed an extra underscore.)
I think the biggest tweak that would be nice if we are also wishing for ponies would be a way to set that function per block of JSX like the way that you can tag a template. That would make it far easier than the current per-file or per-project configurations.
jsx<foo bar={qux}><zoot>bla</zoot></foo>
That doesn't look terrible. Not great either. But I'm sure the big problem with it is that it makes the `jsx < foo` less than versus `jsx<foo />` tag parsing a lot harder.
ETA: New idea, what if it was a fake dot tag .< operator?
if (condition) {
// new scope
const jsx = // ...
return <foo/>
}
In practice, "variable in scope" is essentially the same as "global in scope" and React used to require you to import 'react' at the top of every JSX file for this reason. Your solution is tantamount to just changing that to import jsx from 'react' but still requiring it explicitly.
The main problem with this is that it's non-trivial to make sure variables are in scope before/while loading a file, and they should be loaded on an as-needed basis instead of always, which is what inevitably ends up happening when you need React classic, you import React at the top of every HTML file. These problems seem to be what led to auto-imports.
I admit that my solution still does require you do something like import interpretJsx from 'something', and even further it requires you to manually call interpretJsx(<foo/>) which can be tedious and verbose.
The main benefit is that at least it becomes a standard. And besides, we probably won't need to call those functions everywhere, just at the top of a tree, like the same place you call React.render(root)
The main downside is lack of monomorphism, especially if you're passing an entire tree to interpretJsx(). I admit this is not solved by my proposal. But I still wanted to put the proposal out before smarter eyes than mine anyway.
A problem with going straight to objects is you potentially create a lot of GC churn if the `interpretJsx()` converts it to a different data structure. Also because it is a tree structure, you are probably talking a recursive depth-first-search or other tree-walker algorithm that can easily blow up the stack while it works. If you are doing something Virtual DOM-like and evaluating lots of trees, that memory/GC and CPU churn compound quickly.
This is part of why JSX has always been function calls rather than a data structure to interpret: the recursion to walk the tree is flattened at "compile time".
I don't see why it is a problem you'd need to import your `jsx` function in every file that uses JSX syntax, but perhaps because that's how I've always preferred to use JSX, even with React. Explicit imports are better than implicit ones. If you use lit-html you have to import its `html` function everywhere to get html`` template literals to work. It's not a lot of overhead and it works well. You can add the auto-import smarts to your editor, to your snippet files and template files. Typescript already has a ton of auto-import suggestions as you write a file.
Yeah I guess you're right. This proposal has serious performance issues.
Still, I don't like the React-classic style, the React-autoimport style, or the hybrid that you're advocating for. None work be very ergonomic and require tooling to at least give you errors that you forgot an import, which defeats the purpose of standardizing it at that point.
Maybe it’s ugly, that seems like a matter of personal taste. But it’s syntactically unambiguous which is valuable especially for a language which mixes expressions and statements in often very subtle ways. There are also other ambiguities to consider:
const trailingIntetpolation = <><div/>{foo}</>;
const leading = <>{foo}<div/></>;
const bare = <>{foo}</>;
Template elements don't have great Typescript type checking today. I've been very happy with both, writing things in TSX with deep type checking and then statically rendering them to Template tags.
I did this in the past, but I found that templates simplified the process. The same functions I use to populate/update elements can be reused on existing elements and newly cloned template elements. It can be done in JS by returning strings of unpopulated elements, but then my display is further mixed with logic.
I like to create the HTML and CSS as I'd like it with test data, then just wrap that with <template> tags. Easy to preview without triggering function calls or pasting it into code.
Probably not important, but as I recall I think there was some minor overhead in translating from a JS String to an Element.
Can you recommend examples and tutorials of happy <template> usage, showing advantages over building values for innerHTML etc. as text from strings and template string literals?
This MDN page is where I first discovered the <template> element, and I wasn't very impressed: verbosely operating on one textContent at a time, using ordinal indices into very untyped querySelectorAll results, apparently gratuitous complications with document fragments and shadow DOM.
Verbosity never concerned me. Readability, maintenance, page weight and separation of concerns does.
Typically I do not find myself using querySelectorAll within the context of a template element. When I do I am usually applying changes universally to all of the matched children within the element. The CSS class of the element reflects these things semantically. I find this highly readable. It also resolves most concerns around the type of the element.
Of course in Vanilla JS, types are a mess. Even Microsoft's finest duct tape script doesn't fully resolve these issues. Overall I find Vanilla JS and browser APIs to be a lesser evil than the build systems, package managers and indirection of the React/Next.js approaches. After all, these compiled environments and their libraries will boil down to Vanilla JS when deployed in the browser.
I made this proposal because JSX is more generally useful than generating HTML. You can use it for configuration, or views for other GUIs like I do in https://90s.dev/os/ or to describe basically any kind of tree.
Now that I am paid to occasionally write templates in languages other than JSX, I do miss it so much. All the ERBs and ng-for s and handlebars of this world don't remotely come close.
https://facebook.github.io/jsx/ is a mirage, apparently.