If you know those things then you basically know ERB. And to your point the Ruby LSP works with ERB.
With actual XML and JSON (or csv etc) I would agree: for that you need a serializer, not a string templating env.
But for HTML, especially when used to build GUIs, string templating makes sense - to me -.
What I meant with "leaky abstraction" is that when I write a web-app, my goal is to write certain HTML, the actual strings, to a browser. I'll have my developers console open, inspect DOM/source and then modify the code of the app until the DOM/source matches what I want or need it to be. The closer my "code of the app" is to this actual DOM/source, the less I need to think through layers.
I prefer to just "move this UL out of the DIV and next to the H2¹" rather than manipulating the structure in a tree of objects. To each their own, I guess, but I rather prefer to stay "as close to the metal" as possible, especially if that metal isn't complex or difficult.
¹ I realized I wrote my HTML elements in capitals. I am that old; and I don't anymore, but old habits creap in, I guess :)
The thing about Phlex is it’s actually one less thing to learn. Instead of learning Ruby and ERB and how to use them both with HTML — and all the context switching that goes with using three languages in one file, you only need to know Ruby and HTML semantics (elements, methods, attributes, text, comments, doctype).
> Another syntax/LSP/docs to add to the IDE, or editor.
If you’re writing an app in Ruby, you’ll already have Ruby LSP set up in your editor. There is no other syntax. I would strongly advise against using Phlex if not already writing an application in Ruby, though I have heard of someone using Phlex to generate PHP.
> Another leaky abstraction that has to be debugged, monitored, performance-tested.
I don’t know what you mean by “leaky abstraction”. I consider Phlex to be an air-tight abstraction around the parts of HTML I’ve ever had to use in the last 15 years of programming. There are a couple of things you can’t do without using `raw` — mainly Ctags and other non-HTML5 doctypes.
If I saw a convincing argument for supporting those parts of HTML, I would.
The real leaky abstraction for HTML is string templating, where none of the HTML semantics are validated and XSS vulnerabilities abound.