Try covering dynamic-sized image with an svg surface. Try outlining a hovered element so that outline/shadow covers other items around. Try to baseline align text in non-shallow div hierarchy (which is required to workaround other issues).

When there is a moderate visual requirement I’d better mess around with something like GTK+/Cairo, NSView or QPainter rather than pulling hairs at CSS trying to slap few incompatible js-y solutions into one big ball of bs.

I'd appreciate a visual example of what you are talking about. Seems like a far fetched, obscure-as-hell example to use as an argument. A strawman as they would say. CSS isn't perfect but it's pretty damn powerful for something I can teach a 9 year old.

We can certainly get a lot of work done within CSS's limits. For the auto-generated billion div soup, god speed and good luck.

You can also hit the ugly points very quickly.

E.g., suppose I want to draw diagrams of text connected by Bezier curves. Looks like SVG is the obvious choice for me. But suppose I add editable text to the mix-- well, HTML5 certainly has plenty of options for me there. And the Javascript to glue a solution together however I want.

Yet, none of those options come with any CSS that allows me to align the baseline of the text in my choice of HTML element with an SVG text element of my diagram. I have to scour Stackoverflow for some holy-shit hack of nested div bullshit just to tell the browser to position HTML stuff the way it does for an SVG text element.

These rought seams joining SVG, HTML, and CSS remain very frustrating until you start reading the specs and list archives to figure out that they were very different tech developed by very different teams with very different initial purposes and use cases in mind

(That's not to say there aren't Javascript exeskeletons that inject the SVG directly into your DOM for you.)

Covering a image with something (whether SVG or not) is very common, though most of the time the image’s size will be known, or at least known in one axis. We’re talking things like putting captions on top of images.

Needing hover effects to render on top of elements around is even more common—if you want a row of buttons that share borders, by far the easiest way involves this very situation. Example where `position:relative` does the magic:

  data:text/html,<style>button{border:2px solid red}button:is(:hover,:focus,:active){border-color:lime;position:relative}button+button{margin-left:-2px}</style><button>ⅰ</button><button>ⅱ</button><button>ⅲ</button>

It doesn't make them any less of a hack.

I know many things that are hacks and that I will acknowledge are hacks, but these two aren’t hacks at all. They’re straightforward and natural CSS, working in exactly the way it was designed (and in areas where that design is completely sensible).

The absolute "obvious ways":

- a wrapper div that has to be position: relative. And a display block/inline-block because otherwise it won't work. An svg and an image inside that div. The svg has to be position: absolute.

- a full-blown grid. Two elements must be forced to occupy the same cell in the grid

Ah yes. Absolutely logical and coherent and not hacks at all.

> Getting hover effects to render on top, well, I wouldn’t consider that a hack either, though it’s a little more subjective

"Not a hack": rendering is done in order, but you still have to "shunt the hovered element onto a new layer or increasing its z-index above".

And this still doesn't precisely solve the problem of "outlining a hovered element so that outline/shadow covers other items around". Because most likely you'll have layers on top of layers of "logical coherent" wrapper divs to make everything work together, and you have to spend some time figuring out how to make that particular problem actually work.

And god forbid you want to make it pretty: https://tobiasahlin.com/blog/how-to-animate-box-shadow/

Not everything is straightforward, but the things I was talking of are. (Incidentally on the matter of performance on animating box shadows, that’s actually not fundamental; Firefox has proper GPU rendering of it all, via the WebRender project, so that advice is irrelevant and actually (negligibly) slightly harmful in Firefox; it’s just that Chromium and WebKit haven’t caught up.)

We can certainly get a lot of work done within CSS's limits

All I really want is a linear model over current layout model. E.g.

  ...your regular css...

  .wrapper:relations {
    div.text/left == svg.line.underliner-segment/left + 5px
    div.text/baseline == svg.line.underliner-segment/bottom + 2px

    #item-a/center-x == #item-b/center-x

    #item-d/right >= #item-c/left + 20px

    */right <= $this/right - 10px
    */bottom <= $this/bottom - 10px
  }

Thanks for this but

> I don't want play a detective with css properties ever, when there is a proper mathematical model for coordinate relations.

Designers and UI devs would probably rather just write custom shaders on canvas if they have that level mathematical prowess. For most, offloading that cognitive load to an abstraction like CSS is worth it. Let the browser do the math. Playing detective then seems like a woefully arbitrary issue to have.

Those who are fine with current CSS could then continue to craft wands and potions and everyone'd be happy.

I believe it's called the dark arts.

Sounds challenging to implement that in straight CSS :) I'd love to see a demo

This isn’t difficult. Two reasonable approaches immediately occur to me on this.

① The traditional one: <div style=position:relative;display:inline-block><img><svg style=position:absolute;inset:0 /></div>

② Using Grid’s ability to put multiple things in the one cell: <div style=display:grid><img style=grid-row:1;grid-column:1><svg style=grid-row:1;grid-column:1 /></div>

> Try outlining a hovered element so that outline/shadow covers other items around.

On hover, ensure it’s on a layer of its own, and if its surrounds are also on layers bump its z-index above theirs. e.g. `:hover { position: relative; z-index: 1 }`. Definitely hacky, but not difficult, unless you want the decoration to render beyond the bounds of a parent element’s layer, such as if it’s in a scrolling area with 0.5em of padding and you want a 1em box-shadow around it—that simply can’t be done, layers are inviolate; but as consolation, other platforms almost certainly have the same limitation.

It indeed is [1].

Expected: a yellow line crossing an entire 236x236 image, in all three cases.

Explanation: svg default size is lunatic (300x150) and it doesn't care of your layout modes, so it clips all drawing beyond that, unless you specify either width/height attributes (not css properties!), or a viewBox, from which it infers missing width/height.

I lost a day to this once, when parent-filling svg clipped 3 of my 5 shapes for no fucking reason. Shapes coincidentally fell either fully into 300x150 box or fully outside of it -- no easy hint for you.

You can pre-calculate that in js, but if you specify .wrapper's dimensions in non-px, or use some aspect fit, you're screwed as well.

[1] https://codepen.io/mhmdqrn/pen/yLMvGJW

For the sort of thing that you’re doing here, there are two most likely possibilities for what you want:

① Give the <svg> element a preserveAspectRatio attribute, most likely preserveAspectRatio="none" if you’re just trying to draw a diagonal line across the image. This’ll allow it to stretch the SVG, effectively throwing your stroke-width off a bit so that it could look wrong if you use it on multiple images of different aspect ratios.

② Accomplish a dynamically-sized viewBox with percentages: <svg width="100%" height="100%"><line x2="100%" y2="100%"/></svg>

Fortunately all of this stuff is nailed down pretty well now. IE always had trouble with these more unusual aspects of SVG, even to IE11. I fondly remember a design I implemented a decade or so ago that exercised just these sorts of fiddly areas, causing such issues on IE9 that I gave up and made it use the simpler, not-as-pretty non-SVG fallback that was necessary at that time because of IE6–8.

If you pick any other technology, e.g. an airport gateway system or a shop checkout line, and apply this "can be surprising if you don’t know what you’re doing with it", you'll get a big congestion instead of a functioning airport or a shop. Users are not to blame when the system is based on some cryptic nonsense. Good systems just work, bad systems require custom effort to fly and never scale. These congestions induced by web tech only pull us back and I can't understand why you're so positive with it (if not condescending; sorry if I mistook that).

You could embed the dynamic-sized image within an svg, perhaps even a fully dynamic inline svg.

> Try outlining a hovered element so that outline/shadow covers other items around. Try to baseline align text in non-shallow div hierarchy

I mean at some point you need to use JS. Still, IMHO it's quite amazing how far you can go with a purely declarative language. That also means a myriad of side effects including smooth reflow being taken care of automatically as long as best-practices are followed. Is there any way to write these examples completely declaratively in GTK+/Cairo, NSView or QPainter?

It gets a bad rap because it's so easy to use for non-designers that a lot of sites that use it don't bother customizing the UI elements like buttons or expand/collapse accordions etc. Or the layout of the page, for that matter. The famous "Every Bootstrap website ever"[0] site outlines this nicely:

[0] https://www.dagusa.com/