Support <> standalone fragment syntax
smrq opened this issue · comments
This syntax (or similar) has been brought up in a number of comments across various issues already, but as far as I can tell has no standalone proposal.
The syntax is:
<>
<Child />
<Child />
<Child />
</>This is equivalent to an array, except that because the structure of the array is statically known, keys could be automatically assigned to each child.
I think a lot of people may mistakenly think React 16.x can already do this as a result of the new feature to return multiple components from render. I suspect the lack of this syntax might become even more of a pain point with that release.
(Correct me if I'm wrong or should clarify anything further. I'm just making the issue because it seems there should have been one already.)
cc @SanderSpies, if I recall you have something like this in Reason. How did it work out?
@smrq let me reverse position (not the devil's advocate ;) here and say that React 16.x now allows us to do that (by explicitely setting key= in <Child />) ; BUT lots of us would like this to be implicit for the sake of new "users".
The key thing is one issue that follows a very specific heuristic. It may not be enough to warrant new syntax. It might be possible to find other ways.
The more convincing argument to me is the ability to nest text content and dropping the comma separator.
return <>
Hello <strong>World</strong>!
</>vs.
return [
"Hello ", <strong>World</strong>, "!"
]Another variant would be to drop the comma between JSX elements in arrays. This doesn't solve the text content issue though. Maybe that's just an argument for dropping JSXText #35?
return [
<Foo />
<Bar />
<Baz />
]vs.
return [
<Foo />,
<Bar />,
<Baz />,
]One thing that would be neat is that you sometimes would like to put a key on a whole fragment.
var dynamicList = items.map(item => {
if (item.type === Foo) {
return <>
<Foo />
<Bar />
</>;
}
return <Foo key={item.id} />;
});In the proposed <> syntax there is no syntactic space to define a key. If you put something in there such as "frag" then there is.
var dynamicList = items.map(item => {
if (item.type === Foo) {
return <Frag key={item.id}>
<Foo />
<Bar />
</Frag>;
}
return <Foo key={item.id} />;
});This can be implemented today in React as a component.
I'll also point out that non-React usage of this would also greatly benefit (like Mithril + @jsx m, where Mithril already supports fragments).
The issue with Frag is that it is in the same namespace as regular components.
How about <[]>…</[]>? => <[] key="foo">…</[]>
Or this might work: <{...{key: 'foo'}}>…</>
What about something like <fragment />, which would be a host component that React could define?
Another proposal was to use namespaces.
<react:fragment>
<Child />
<Child />
</react:fragment>But if we were to do this, should we go all-in? For example key and ref are also React-specific (and technically don’t become props) so we could also do
<div react:key="1" react:ref={fn} />But this is awkward to assign when defining element config as plain object.
What about something like
<fragment />, which would be a host component that React could define?
This blocks HTML from ever defining it. Kind of like MooTools killed [].contains().
document.createDocumentFragment() already exists and HTML is highly unlikely to ever define <fragment>, but fair point 😬
I like the namespacing idea.
A few more ideas.
<@fragment>
<Child />
<Child />
</@fragment><$>
<Child />
<Child />
</$>Or, use <> and when you do need to add a key, do-it-yourself:
const Frag = ({children}) => childrenReact could provide this as a named export of course.
I really like how obvious the <> syntax is. OTOH. It prevents<> being defined as a JS operator at some point.
I like this.
<*>
<Child />
<Child />
</*>It kind of implies "multiple elements" and you can put a key on it if you want.
var dynamicList = items.map(item => {
if (item.type === Foo) {
return (
<* key={item.id}>
<Foo />
<Bar />
</*>
);
}
return <Foo key={item.id} />;
});Haha I guess that won't work because of comments 😛
you could always end it with </> :-p
The reason I dislike typing "fragment" is because it's one of its kind. It's not like we're going to add more primitives like this. It's pretty fundamental by itself, rather than a part of some bigger API of core primitives. So it's annoying to squat on a whole namespace or create a special export just for a single thing. If you see react:fragment you might expect there to also be react:if, react:for etc.
There's two counter arguments. One is making it explicit makes it very easy to Google. Googling characters is not very pleasant (although maybe they fixed it).
The other counter argument is we may in fact add more primitives. Such as yields and coroutines. But those seem to be solved better on syntax level than by adding special cases of elements. Especially since they're not even producing elements.
I mean you could support <key="..."></> if you wanted to. Although it may be somewhat confusing. (< key="..."></> is also possible but looks rather weird in context.)
<[arbitrary char symbol]> (e.g. <!>) seems like the obvious go-to but it tends to look just like a jumble of symbols (especially the closing tag, </!>).
<[arbitrary char symbol]fragment> (e.g. <!frag>) is nice in that it's descriptive and not really special, but it also becomes rather wordy. For fragments which is meant to signify something "less" than a regular element, it certainly looks like a lot more.
Personally. Despite its strangeness, I kind of like <> and <key="">, it looks like a fragment (i.e. nothing, just a hollow container) and it's short so that you can reasonably have it on a single line in some cases.
return <key="foobar">
<div />
</>;return (
<key="foobar">
<div />
<key="barfoo"> // someone could do this, but I'm not sure why they would
<strong />
</>
<FooBar />
</>
);Or is it too strange to stomach?
Can someone explain why this wouldn't work?
return (
<div>Element 1</div>
<div>Element 2</div>
);
Which could just be sugar for
return [
<div key="key1">Element 1</div>,
<div key="key2">Element 2</div>
];
Afaik it doesn't conflict with anything + doesn't need new syntax.
@rikkit Even if that would work it wouldn't solve <frag>You have <Counter /> items</frag>.
Here's an example of ambiguity this would introduce.
return (
<div>Element 1</div>
) this is just a paren in text node
<div>Element 2</div>
);Note how text nodes (which are normally valid in JSX) become broken as a result. It's not clear if ) is text or intends to close the opening paren.
There are similar issues with newlines, braces, etc. All of those are OK in JSX but become ambiguous if we don't have a shared JSX ancestor.
I'd just like to point out that .net/razor templating language introduces <text> for defining bare text blocks (no wrapping element), so it's not unheard of to introduce special tags (within the "HTML namespace").
I think I'd prefer either <frag> or <> but I'm not sure about this <key="x">...</> idea. It looks too unbalanced to me; even as I wrote that I tried typing </key>. I think IDEs will erroneously try to autocomplete that closing tag too actually.
A user-space <Frag> really isn't so bad except I imagine a great deal of libraries would be re-implementing the same thing over and over again, so it's probably better to make it core.
@mnpenner This is the JSX-namespace though, and not the HTML-namespace. Also, by reserving the frag element you can no longer create <frag /> elements in HTML, which is legal.
EDIT: "JSX-namespace" as in it applies to all current and future JSX/React renderers, not just HTML.
I think IDEs will erroneously try to autocomplete that closing tag too actually.
Unless JSX ends up reserving an element name then the IDEs need to update the autocomplete and syntax highlighter anyway.
This is the JSX-namespace though, and not the HTML-namespace.
What do you mean? <a>, <span>, <div> and all the other lowercase tags represent HTML elements. <frag> would fall into that space.
Oh... you mean for people that are using JSX for non-ReactDOM? Yeah, okay. I see your point. Using <frag> would use up a token.
Unless JSX ends up reserving an element name then the IDEs need to update the autocomplete and syntax highlighter anyway.
<frag>, <Frag>, and <react:frag><jsx:frag> then, no?] will all autocomplete today. <> OTOH, does indeed confuse PhpStorm and probably most other editors. Not that it's a huge deal -- I'm sure they'll be updated.
Anyway, I'm not sure what the best solution is then.
BTW, I like the namespace best:
- React's fragments are just plain fragments with children. No attributes are supported.
- Mithril's fragments can potentially have attributes, and this is officially supported.
- Mercury's virtual-dom backend doesn't support fragments at all.
- nativejsx (JSX to straight DOM) doesn't even really need fragments.
Here's how I feel it could be handled, via modifications to babel-plugin-transform-react-jsx:
- Add
fragName: string- the name used to define fragments. This could either be a JSXIdentifier, JSXMemberExpression, or JSXNamespace. This defaults to whatever React ends up using. - Add
fragAttrs: boolean- whether to prepend the attributes to the call. This defaults tofalse, as it's a limited use case. - Add
fragPragma: string- the function to replace fragments with, like the existingpragmaoption or inline@jsx foo. This defaults to whatever React ends up using. - If a JSXElement has an opening element whose tag matches
fragName, it's transformed to a call to the function described byfragPragma. IffragAttrsistrue, the attributes (ornullif none are passed) are the first argument. The children are appended as the remaining arguments.
Is this syntax necessary?
Just leave it up to the community to establish a convention. See:
@gajus Its semantics differ from "true fragments" and has additional performance overhead.
@gajus Its semantics differ from "true fragments" and has additional performance overhead.
Sorry, rushed into discussion. Regardless, I disagree that a new syntax (such as <> or <*>, anything thats not already valid JSX) is necessary.
Would it not be better to reserve a component name for the use case? (e.g. Aux, Fragment or whatever else) and treat it specifically.
Otherwise, every new edge case requirement such as this will require a new syntax. Syntax such as <> makes the code uneasy to the eye (because it breaks the JSX flow), it will require updating bunch of linters/ formatters/ plugins that depend on JSX syntax, and it is hard to Google/ describe unambiguously for the purpose of reporting issues/ documenting behaviour/ communicating in person the subject.
This works fine for me for rendering static arrays (React 16):
const autokey = (children) => children.map((c, i) =>
React.cloneElement(c, {key: i})
)
export const Element = (props) => autokey([
<li>one</li>,
<li>two</li>,
])For this use, I see no need for a custom element, custom syntax, or even a custom component to provide for this functionality. (Assuming the use of arrays is fine, just manually enumerating keys is not.)
@felixrabe yes but using the index with items that actually have unique qualities that are related to their props/content defeats the perf benefits of using key.
Yes, that completely defeats the purpose of the warning. Which is to warn you that if you don’t specify keys representing stable identity rather than just position, you risk having very obscure and complex bugs that are hard to debug.
That's why I said "static arrays" :) totally agree.
Ah, missed that part. Still, cloning every element is not very good for performance.
Not symmetrical but what about something like:
<[>
<Foo />
<Bar />
<]>I guess it breaks the closing / semantics 😕
Syntax such as <> makes the code uneasy to the eye (because it breaks the JSX flow)
Just a unnamed tag. I think <> is setting a clear intent: group tags but do not wrap. In my opinion the intent is clearer than if it had any name, be it aux, fragment, or arbitrary symbols.
it will require updating bunch of linters/ formatters/ plugins that depend on JSX syntax
Fragments have been requested for a long time, it would be nice to have an official syntax for it. Plugins can and will update.
and it is hard to Google/ describe unambiguously for the purpose of reporting issues/ documenting behaviour/ communicating in person the subject.
It's just an operator like any other. I'd google for "jsx empty tag" if i had no idea it's called a fragment.
<> would, generally speaking, break things and confuse people. That syntax would have a nonzero cost to the community. The breakage may not be huge, and all confusion can be fixed with some googling (though it's harder in this case), but we can probably get ~all of the benefits at a far lower cost with something like <frag>.
Eg; I can't tell from looking that <> is "just an unnamed tag" -- I might guess that, but now I need to look up what dom node it produces and what its other behaviors are. That's 5-10 minutes wasted across a large number of developers that probably didn't have to be.
Another point worth mentioning (but not worth deciding based on) is that this reserves the <> glyph from EcmaScript and (more likely) compile-to-js languages. I don't expect much sympathy there though.
this reserves the <> glyph from EcmaScript
I don't think it does, for the same reason that <Component> doesn't reserve the combination < + some characters + >.
After JSX is compiled to JS those glyphs are gone, so it doesn't matter if the browser interprets <> different than React developers expect. By then (if that ever happens) compilers might need to be smarter when to interpret <> as React and when as JS though.
Am I missing something thinking like this?
Am I missing something thinking like this?
Yes 😄 it prevents developers from writing const baz = foo <> bar; as valid syntax in a .jsx file. The possibility of TC39 approving such a syntax is small, however, so the bigger concern (which isn't really worth any of our time either) is some compile-to-js language which wants to use <> as a glyph now can't if it wants to support JSX.
It may be small, but it's nonzero. A number of people want a spaceship operator (<=>), and <> is a valid alternative for that - to name an unproposed, wildly speculative example.
It seems this task can broke the web easy :)
return ([|
<div>OK</div>
<div>OK</div>
|]); return ([:
<div>OK</div>
<div>OK</div>
:]); return (<!?>
<div>OK</div>
<div>OK</div>
</!?>);return (<;>
<div>OK</div>
<div>OK</div>
</;>)return (<``>
<div>OK</div>
<div>OK</div>
</``>)return (<[]>
<div>OK</div>
<div>OK</div>
</[]>)@rattrayalex const baz = foo <> bar; is compatible with JSX for the same reason that const baz = foo <bar; is compatible with JSX. <> as an operator is different from <> as an expression/value. Although elements as statements which is being considered could possibly complicated it a bit, I'm not entirely sure. Visually it could be a little jarring in edge-cases and error messages could suffer in some cases, but as far as I can discern they not mutually exclusive.
Compile-to-js languages will not be JS and thus not using JSX. They may choose to implement the same syntax or similar if the host language allows it, but in some languages it will simply not be possible to implement a similar syntax and they will have to implement their own unique dialect.
Is there a reason we cannot use named JSX tag such as <Aux>, <Frag> or whatever and make the name configurable at the babel-plugin options level?
This way,
- A quasi- standard will be established by the default babel config value.
- It does not break anyone's existing React project because everyone can configure the name that does not clash with their existing component names.
- It allows to add more special components in the future without needing to think of yet another syntax.
- It does not break the existing ecosystem of the plugins and other supporting tools.
- Should future React/ JSX implementation somehow change and the current special treatment of such construct become redundant, it would not require updating the code.
@gajus I already proposed almost exactly that, just in a more expanded form that admits namespaces as well as simple component names.
Note that React isn't the only framework that uses JSX - Mithril offers it via option, and Inferno uses it by default. In addition, Mithril had fragment support long before React did (by about a year), and has developed a separate API for it (raw arrays or m.fragment(attrs, children), the latter of which accepts keys and attributes).
Is there a reason we cannot use named JSX tag such as , or whatever and make the name configurable at the babel-plugin options level?
There's already enough configs in a modern react project that I'd really want to avoid having another knob, especially such a project-specific syntax setting. Why add a special string to be set in a config file instead of the status quo of having the string just be a variable/import name like you can already do? That config isn't doing anything more than an implied import.
import Aux from 'react-aux';
const j = <Aux><div>1</div><div>2</div></Aux>;
Personally I'd prefer one of the concise syntaxes (<>...<>, <>...</>, <*>...</*>, etc), but failing that, it would be nice if React included Aux's functionality in a property. (import {Frag} from 'react';, etc.)
If JSX tag names let you use reference properties and React.Frag existed, then you could do this without any extra imports:
const j = <React.Frag>...</React.Frag>;
Why we can't just avoid any wrapper syntax and parse <a/><b/> as a fragment? As far as I can understand parsing process (I have some experience: https://github.com/veged/ometa-js https://github.com/veged/krasota.js) it's possible to treat <a/> as a single element and <a/><b/> as an array of elements.
@veged that wouldn't work with eg; jsxtext.
@isiahmeadows would <frag> work for mithril et al?
It seems like an option which is:
- concise
- unambiguous
- not likely to conflict with present or future features of browsers or frameworks
Are there downsides?
@rattrayalex As long as I can remap it to either an array or a function call with possibly-null attributes, it should work with Mithril.
Note: magic component names do have the downside of preventing names from being used, so that's why I also proposed the use of namespaces. I'd personally prefer <m:fragment {...attrs}>...</m:fragment> ↔ m.fragment(attrs, ...) for Mithril's JSX API, but that's more of a design preference than anything else.
Inferno doesn't have native component support, and I know next to nothing of other JSX-based libraries and frameworks.
Yes, a namespace is one way of not stepping on anyone's toes; I'd hope that a shortening like <frag> would accomplish the same goal without as much ugly, but <react:fragment> certainly is unambiguous, so I wouldn't shy away from it. (It might seem like a "scary lib internal" to users though)
@rattrayalex To clarify, the namespace usage would be optional with my proposal.
@rattrayalex For React it seems it will be synonymous with <React.Fragment> (or importing Fragment into the local scope).
@rattrayalex For React it seems it will be synonymous with <React.Fragment> (or importing Fragment into the local scope).
Which sounds reasonable. Then just have a babel rule that interprets the matching component in a special way.
Workaround that works with React v16.0.0:
import React from 'react';
import { render } from 'react-dom';
class VirtualDiv extends React.Component {
render() {
return this.props.children;
}
}
const App = () => (
<VirtualDiv>
<h2>Hello</h2>
<p>Second element</p>
</VirtualDiv>
);
render(<App />, document.getElementById('root'));Preview: https://codesandbox.io/s/0m7204mlw
Oops, there's Frag somewhere in the middle of this post that does the same thing I missed. Sorry for the duplicated post.
Excuse my ignorance on the subject, but instead of introducing new syntax, could we not make JSX introduce a transparent wrapper around static arrays that allows consuming libraries to know for sure that they are static? Then the warnings for keys can be discarded with no new syntax, no performance implications—everyone is happy, world peace is attained, etc.
Example:
function render() {
return [
<div>One</div>,
<div>Two</div>,
<div>I am a snowflake!</div>
]
}Would result in:
function render() {
return React.StaticArray(
React.createElement( "div", null, "One"),
React.createElement( "div", null, "Two"),
React.createElement( "div", null, "I am a snowflake!")
);
}Where React.StaticArray is a simple subclass of Array
export class StaticArray extends Array {
constructor(...args) {
super(...args);
}
__static__ = true
}@JonDum Fragments are not really about the data representation, it's to have a useful syntax for this problem: #84 (comment)
@JonDum In your example, how would the transpiler know that it should be a StaticArray?
Excuse my ignorance on the subject, but instead of introducing new syntax, could we not make JSX introduce a transparent wrapper around static arrays that allows consuming libraries to know for sure that they are static? Then the warnings for keys can be discarded with no new syntax, no performance implications—everyone is happy, world peace is attained, etc.
This is a static fragment that is passed to a function that expects some type of element and wraps it.
render() {
const wrapWithContainer = (content) => <div>{content}</div>;
return (
<div>
{wrapWithContainer([
<span>A</span>,
<span>B</span>
])}
</div>
);
}This is an array of nodes passed to a function that expects an array and does array operations on it.
render() {
const processNodes = (nodes) => nodes.map((node) => <div>{node}</div>);
return (
<div>
{processNodes([
<span>A</span>,
<span>B</span>
])}
</div>
);
}How do you tell the difference?
how would the transpiler know that it should be a StaticArray?
How do you tell the difference?
You can use the AST to check for the few scenarios where it could be processed, like ensure we're in the render function, we're not a child of CallExpression (i.e., processNodes([])) or we're in a JSXExpressionContainer (not shown in example)
Example:
module.exports = function(babel) {
const { types: t } = babel;
const StaticArray = (args) => {
return t.newExpression(t.Identifier("StaticArray"), args)
}
return {
visitor: {
ArrayExpression(path) {
let fp = path.getFunctionParent();
const isInRenderFunc = (fp.node.id.name == 'render');
const isReturnedDirectly = path.parent.type == 'ReturnStatement';
const isNotTransformed = path.parent.type !== "CallExpression";
if (isInRenderFunc && (isReturnedDirectly || isNotTransformed)) {
path.replaceWith(StaticArray(path.node.elements));
}
}
}
};
}
Quick n' dirty tests:
https://github.com/JonDum/jsx-static-array-plugin
I'm sure I'm missing cases, but the concept is there. Maybe at this point though it's cleaner to just auto-key the elements instead of wrapping in a special Array subclass.
I created a component to handle this that I called a shadow wrapper. I wrap my child components in it instead of DIVs. I'd be interested to know it's limitations. I'm fairly new to React.
@clemmy @sebmarkbage @smrq Should this issue be closed now that #93 and #94 are merged in?
@matt-landers That's a good question - the difference is in the state preservation behavior when using a class or functional component to wrap it. Although the shadow wrapper works for common cases, it doesn't preserve state in some edge cases such as rendering something like
ReactDOM.render('<Stateful key="a" />')
and then having it be inside a fragment in a subsequent render
ReactDOM.render('<><Stateful key="a" /></>')
Cases can be found here.
Aside from that, there are some optimizations that can be done by putting the React Fragment export inside the core library.
@clemmy Thanks for the response. Looking forward to using the new spec.