I work in this space so I am curious how your library is different from its similar competitors. I read the docs, but I didn't find anything about the actual model for how it works, like whether the whole UI rerenders on any observable change or not. Maybe consider this paragraph a request for documentation.
PS: A minor suggestion: the word "blazing" is so overused in the JS world at this point, when I see it I always at first suspect the app is a parody. If it's fast, then the word "fast" suffices.
I've started learning Rust after many years of NodeJS. In the past 3 weeks I've seen more "blazing" in Rust library marketing than in all my years of NodeJS combined.
Kind off-topic but related: sometimes I wonder how web development would be if Lisp had somehow won as standard, since we can use to do everything the stack HTML+CSS+JS+Backend-language does, in addition to standardized syntax and standard everything.
We were so close. Twice, JavaScript had a chance to be Scheme; both times marketing reasons prevailed (first, to capitalize on the marketing spending of Java; later, to compete against Microsoft's JScript).
If Scheme was the language of the web, then HTML and CSS being represented as s-expressions would be the most obvious next steps, and we'd be like 15 years ahead in terms of progress.
One is people pretending to like something or wrongly saying people like something. The other is people liking something for the wrong reasons. These are completely two different concepts.
I kinda don't get the whole "put html in template literals". Wouldn't there be no syntax highlighting? And the framework is either putting the string into the DOM with some regex manipulations, which would lead to some very confusing error messages, or it's parsing the string into a tree, which means it needs to implement what's essentially XML parsing, which would come with an overhead. And for what? Avoiding JSX? If you don't like JSX, just desugar it to nested function calls. Or just use a templating language.
Of course you can write a syntax highlighter for HTML in template literals. The lit-plugin VS Code extension provides syntax highlighting plus type checking for bindings.
lit-html works by passing the template strings to the built-in HTML parser, so it doesn't need to bring its own.
The benefit isn't just avoiding JSX and build steps, but more importantly avoiding VDOM and expensive diffs. Template literals separate the static parts if a template from the dynamic and remove the need to diff the parts that never change. It's more efficient.
The preprocessing is far from an HTML parser. It just helps pick what marker to join template fragments with before passing to the built-in parser.
You seem to think this is enough of a problem to comment on every time I mention lit-html, but you never back it up with any reasoning. Why should it be bad to do some string processing? What's "ungodly" about joining the template fragments?
The fact is, lit-html is very fast, and allows embedding parameterized, updatable markup in standard JavaScript. You don't have to use it.
My personal opinion is that tagged template literals are the worst addition to Javascript [1]
As HN user stevebmark said [2]:
--- start quote ---
…we’ve learned not to write code in strings. You’d think this would be a fundamental law of software engineering
…
Writing code, not strings, means you can do everything to it that you can do to code. You can type check code. You can lint it. You can optimize it, compile it, validate it, syntax highlight it, format it with tools like Prettier, tree shake it…
> You’d think this would be a fundamental law of software engineering
Agreed, but not for the reasons you quoted.
It should be a law of software engineering because when you're writing code, you aren't writing strings, you're writing a serialized and prettified form of a tree. So when you have two pieces of code you want to join together, you should join trees, not strings.
Breaking this law is precisely why things like SQL Injection even exist in the first place.
I think your blog post misses an important aspect of tagged template literals, which is that the first argument is an immutable object that is unique to the source location of the template. That doesn’t necessarily invalidate the rest of your argument, but it is a very powerful feature that is not really equivalent to anything else in JavaScript.
What do you mean by 'source location'? Why does this matter? I'm intrigued (and also suffering from conjunctivitis so excuse me if I'm being a bit dense and totally missed your point).
If you put a tagged template literal on line 42 in foo.js, every time the tagging function is called from there, the array of strings passed as the first argument is the same object. It’s unique for the source location of the template literal, so even if you paste the exact same code into bar.js, it will have a different object. So, the function can recognize that it is being called from the same location in the source. This has some interesting implications for meta programming and debugging.
I’m not sure that it matters so much to the discussion, but it is a completely unique feature, and really has to be mentioned when discussing what tagged template literals offer.
@localvoid thanks for the investigation, Sinuous can get around this with the template module I believe.
Also Sinuous doesn't use the same library as Solid or S.js for tracking dependencies. I'll do some investigating if it has the same issue that you posted.
`o` is an alias for `observable`.
`observable` can be used as well.
`h` is the pragma, comes from hyperscript.
it's used by a lot of libraries as the createElement function name so I didn't think this needed a more descriptive name.
I took it for a super simple test run and behavior seems very… weird and incorrect: https://codesandbox.io/s/7o92r. Perhaps I'm just using it wrong, but it's not very clear why this example's state fails to update as you would expect. And I found a few examples of rendering that don't behave as expected.
The issue below is still to be fixed.
It currently returns an array when it should be a DocumentFragment. It's related to the popular `htm` package that Sinuous uses.
I've toyed around with the template-strings-to-markup approach before, but couldn't quite figure out how using `map` to generate dynamic lists avoided doing a full O(n) loop upon each update without some kind of diffing engine (other than the browser's own markup diffing).
Is anyone else bothered by the two different meanings of the word "observable" (data streams vs getter/setter/proxy), when you are never sure which one was meant until you read the code?
Introduction of D3 observable notebooks didn't help :-(
PS: A minor suggestion: the word "blazing" is so overused in the JS world at this point, when I see it I always at first suspect the app is a parody. If it's fast, then the word "fast" suffices.