-
Notifications
You must be signed in to change notification settings - Fork 374
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Template Instantiation Developer Productivity Proposal #1013
Comments
Please note that this is now quite different from how it started. |
Regarding the introduction, we shouldn't try to make that ergonomic. It's an antipattern and a significant accessibility and UX footgun, especially with ARIA attributes and form controls. We should instead be nudging developers away from that antipattern (and export default function Multiple() {
...
return (
<form onSubmit={handleSubmit}>
<label>
<div>Name:</div>
<input type="text" name="name" value={formData.name} onChange={handleChange}/>
</label>
<label>
<div>Email:</div>
<input type="email" name="email" value={formData.email} onChange={handleChange}/>
</label>
<label>
<div>Message:</div>
<textarea id="message" name="message" value={formData.message} onChange={handleChange}/>
</label>
<button type="submit">Submit</button>
</form>
);
}
And besides, you can get the names of the inner form controls without extra IDs via TL;DR: I'm not fond of the idea of adding built-in support for automatic ID generation for anything. It's a recipe for disaster in multiple ways. |
I learned today that Vue 3.5 is adding support for auto id generation, and apparently React supports it already. You make good points @dead-claudia , I agree it is much nicer to wrap inputs inside labels, and I hadn't considered the benefits to accessibility for doing that. I also hope the platform does provide alternative mechanisms for linking elements together other than only id's. Perhaps by the time we start to see the platform actually going full speed ahead with declarative templating, such an alternative will be in place, and we can reexamine the utility of it. But for now, it seems many libraries are finding it beneficial. Thanks for your thought-provoking input! |
Template Instantiation Developer Productivity Proposal
Author: Bruce B. Anderson
Last Updated: 2024-3-10
The following outlines some productivity, and higher reliability enhancement proposals, that would make template instantiation more effective.
Support for Microdata, Name and ID attributes, toLocaleString/toLocaleDateString
Idea for this updated section of the proposal is partly inspired by (misreading?) this article.
A common "chore" developers need to perform when binding to forms, is to "repeat oneself" when binding the value and name and id of a form field. For example:
The property names "name", "email", "message" are repeated no less than three time per field, causing unnecessary carpal syndrome, and room for errors.
So eliminating this unreliability caused by needing to keep all three in sync would be a great benefit. Especially if the same idea could be extended to at least one additional attribute: itemprop. I.e. it would be great if template instantiation provided...
Microdata support
A good percentage of websites use microdata. It is still lagging behind some competitors which aren't of the WHATWG standard, however. Let's try to fix that!
Historical backdrop
Given the age of the second link above, it is natural to ask the question, why did it take so long for anyone to raise the possibility of integrating template binding with microdata, at least for a while? Or is there some fatal flaw in even trying? I was ready to attribute this to a massive market failure on the part of the web development community, including myself, but I don't think the explanation is that simple, thankfully.
What I've learned is that for years, the microdata initiative was in a kind of simmering battle with another proposal, RDFa, as far as which one would be embraced as the one true standard.
The microdata proposal suffered a significant setback in the early 2010's, and only in the late 2010's did it experience a comeback, and it seems safe now to conclude that microdata has won out, permanently, in terms of built-in integration with the browser. Some sites haven't been properly updated to reflect that fact, which can partly explain why this comeback seems to have slipped under the development community's radar.
Nudging developers
I think nudging developers to make use of this standard by making it super easy and reliable, when working with template instantiation, would have a beneficial impact for the web and society in general. As we will see, we want to ensure that all bindings are in sync, by avoiding the need to repeat ourselves. Since what the search engine sees is so difficult to determine, I think it's important to build in that reliability, hence the extra complexity I'm requesting to ensure that we get that reliability.
Benefits
At a more mundane level, it could have significant performance benefits. It could allow applications to hydrate without the need for passing down the data separately, and significantly reduce the amount of custom boilerplate in the hydrating code.
With the help of the semantic tags and microdata attributes, we can extract "water from rock", passing the data used by the server to generate the HTML output within attributes of the HTML output, consistent with what the client would generate via the template and applied to the same data. The hydration could happen real time as the html streams in.
Caveats
The biggest cost associated with supporting microdata, is whether updates to the HTML should include updates to data tags' value attributes. Not updating them would have no effect on hydrating, or what the user sees, but might, I suspect, have an impact on limiting search result accuracy and indexing.
The specific syntax of this proposal is just my view of the best way of representing integration with microdata, and is not meant to imply any "final decision", as if I'm in a position to do so. I'm not yet an expert on the microdata standard, so it is possible that some of what I suggest below contradicts some fine point specified somewhere in the standard. But I do hope others find this helpful, especially if it triggers a competing proposal that tries to "get it right".
Because there's a tiny performance cost to adding microdata to the output, it should perhaps be something that can be opt-in (or opt-out). But if having microdata contained in the output proves to be so beneficial to the ability of specifying parts and working with streaming declarative shadow DOM, that it makes sense to always integrate with microdata, in my view the performance penalty is worth it, and would do much more good than harm (the harm seems negligible).
Highlights
This proposal consists of several, somewhat loosely coupled sub-proposals.
Simple Object Example
Let's say our (custom element) host object looks like:
There are two fundamental scenarios to consider:
I don't think template instantiation needs to care which scenario the developer is following. Despite my initial suspicion that it would be helpful, on reflection, I think we should stay clear of making template instantiation emit any itemtype information beyond what the developer specifies explicitly in the attribute.
Suggested symbol shortcuts
As mentioned above, what we want to do is allow developers to easily emit the name of the property they are binding to to various attributes of the element. Each attribute binding would be specified by a single character in the binding instruction, to keep this added support light and small. Suggested symbols are below:
Let's see some examples of this in action:
Binding to simple primitive values, and simple, non repeating objects with non semantic tags
Let's apply the following template:
... to the host object defined above. What template instantiation would emit is:
All numbers, dates, booleans are, unless specified, emitted to the user via .toLocaleString/toLocaleDateString.
If expressions involve more than one property, I think here we should leave it up to the developer to provide the needed tags (including meta) to provide the desired microdata reflection.
For the aria-checked property, if the value of isVegetarian is true/false, set the value to isVegetarian.toString(). Otherwise, set it to "mixed".
This proposal is also advocating support for "quaternion" conditional expressions (condition ? === true : === false : other).
Now let's talk about the dreaded interpolation scenario.
This would generate:
Should there exist, in the future, a semantic tag for read only numbers and booleans, the template instantiation would use it, but we would need to "enable" it for backwards compatibility. For now, use data tags for numbers and booleans, span's for string values, and time for dates.
What this example demonstrates is we apparently don't need the use of ranges, when performing interpolation, if we want to support microdata. If there is a significant performance benefit to using ranges, with meta tags, or link tags for binaries, that could be used as an alternative (that was my original thought on this question). If the performance difference is tiny, I think the simplicity argument should prevail.
id management
Referential support with user specified reference id.
It often arises that the id for one element should be dynamic, and another element needs to reference this dynamic value.
Examples are the label's for attribute, numerous aria- attributes, and microdata's itemref attributes. In many cases (itemref, output's for attribute) we allow for a space delimited list of id's.
To help with this, I propose:
... would generate:
Maybe the Y symbol should be followed by a *, but the point is, what follows the Y symbol, if anything, could be a css query to match for everything below the adorned element. To reference the previous element, use ^, followed by a css query if applicable. In either case, stop at the first match.
We could also perform a general css search inside the idref function, that would need to be done carefully within the #each block, so that if multiple elements are found matching the css query within that block, then the attribute is a space delimited list of all the id's of matching elements. In this case, instead of using ^ or v, the tentative recommendation is to start with query with a * followed by a space. If we want to specify a query based on one of the special symbols listed in the table at the top of this document, this could be done as well. Examples of such rules are spelled out in more detail here (WIP), where we are trying to make that userland library conform with this proposal, basically a POC of sorts.
Referential support with auto-generated id's.
Conditions with microdata
Suppose we want a conditional to output more than one root element. It's unclear to me what the final outcome will be as far as whether built-in template instantiation will use some "marker" in the live DOM tree that can be used to remember the contents of the template from which the conditional content derived. Such a marker could be useful, if the condition changes to be false, and the conditional content is removed, and then the condition changes back to true. Perhaps built-in template instantiation could be smart enough to do this without any markers in the live DOM tree. We assume below that template instantiation would rely on such a marker, and suggest that that marker be an empty template element (with the already parsed "content" stored in some memory location, so that no parsing of the inner HTML would be required).
If no such markers are needed, then it seems to me there will be some inevitable loss of information as far as microdata is concerned, and as far as "reverse engineering" the output in order to infer the template is concerned. But if the performance improvement from not needing a marker is significant enough, that should probably trump concerns about loss of information.
... would generate:
So we use itemref to maintain a hierarchical tree logical structure, even though the DOM structure is flat.
Also, the template instantiation would automatically add the equivalent of id={{generate-id()}} to all elements inside the condition, unless the developer specifies an id.
Loops
The scenarios get a bit complicated here. What follows assumes that the developer wants to be able to "reverse engineer" the output, and be able to guarantee they can distinguish between content that came from a loop, vs. content that came from a single sub property. In what follows, I assume the tougher case. If no such requirements exist, the developer can drop specifying the itemtype.
Example 1
Suppose we have a list of todo items:
would generate:
I think now when hydrating, even when there's a single child list item, that we can tell we are working with an array of items, rather than a sub property. All of the types pointing to schema.org and to mywebsite.com are optional. I don't think template instantiation would need them for anything.
Creating artificial hierarchies with itemref and loops
Similar to the conditional:
would generate:
Conveying searchable content via attributes.
One serious limitation of the microdata standard, is that much useful information is conveyed from the server in attributes, such as alt, title, value and data-*, and there is no way to specify itemprops for those attributes.
Suppose WHATWG adopted another microdata attribute, say itempropmap, that would allow us to provide more useful information:
Template instantiation could help generate these mappings with reliability:
The text was updated successfully, but these errors were encountered: