RectCut for dead simple UI layouts

https://halt.software/dead-simple-layouts/

by hsn915 • 5 years ago

89 43 5 years ago

halt.software

dstick• 5 years ago

Is this UI so simple that it’s meant to break on mobile?

rexelhoff• 5 years ago

ReactCut

Varriount• 5 years ago

Idiosyncratically, the website looks better on mobile if you view the page in desktop mode. That is, the content actually fits to the screen size, and the code examples aren't cut off mid-line.

chrismorgan5• 5 years ago

That’s half of the point of giving a desktop mode switch, that it makes the browser ignore the “no really, I support mobile” <meta name="viewport" content="width=device-width"> tag.

tobr2• 5 years ago

It’s a good illustration of the need for a simpler layout model than the three-dozen-piled-layers-of-cruft model CSS is based on.

dahart3• 5 years ago

OTOH, CSS can handle a vastly wider array of styles and use cases. I agree it’s crufty, but there are many historical / legacy reasons for it that are still in use today.

If used in practice for anything substantial, RectCut will start to acquire cruft too. First thing to get added would be resizeable / dynamic layouts. Next would be something akin to the CSS grid layout, which is already the successor to flexbox (closest CSS layout mode to RectCut).

chrismorgan5• 5 years ago

It’s a hopelessly bad illustration of that.

The web does a singularly excellent job of supporting varying viewport sizes out of the box, but this page actively sabotaged that by claiming to support mobile:

  <meta name="viewport" content="width=device-width, initial-scale=1.0" />

While applying CSS rules that ruin it:

  .article-card, article h1, article ul, article p, article ol {
    width: 600px;
  }

  pre {
    overflow-x: hidden;
  }

These rules are obviously bad. width should be max-width, and, well, what would you expect to happen if you tell it to hide the overflow?

(It would have been OK even with these rules if the page hadn’t claimed to support mobile sizes, as the mobile browser would have rendered the page on a larger viewport and scaled it down, more or less.)

fouric2• 5 years ago

The code examples are in C and make no mention of any webtech or mobile. I'm pretty sure that this is meant for desktop interfaces.

altschuler• 5 years ago

This approach is also used/encouraged by JUCE: https://docs.juce.com/master/tutorial_rectangle_advanced.htm... (see "Layout by subdividing rectangles")

slver3• 5 years ago

I’m unsure what’s the difference between this and hierarchical layouts. And say flexbox

wruza7• 5 years ago

Mostly the ease of definition (a good part), but also there is no calculate-demands / distribute-size loop, only the distribution part. If a medium gets too small, it clips content.

slver3• 5 years ago

Is... that good?

cwales95• 5 years ago

Looks nice, but unresponsive on small screens.

_448• 5 years ago

This is interesting! I had implemented a similar system a while back for a mobile OS company. It was used for optimisation of the rendering of the UI scenegraph. The rectangles were cut(taking into account alpha blending) before feeding it to the renderer to improve performance. This was all post-layout though. Didn't realise that it can be used for layout as well!

wruza7• 5 years ago

It is tAlign layout from Delphi. Alignments were cutting a parent surface in order of children appearance. It allowed to solve 95% of desktop ui without complicating things.

I’m aware of four main layout models: fixed (x,y), alignment model, boxing model (gtk, qt, html), and constraints model (apple ui). Can anyone list more?

helltone2• 5 years ago

How does layout in say LaTeX compare? Is text/font layout different fundamentally?

iainmerrick• 5 years ago

I would call that “flow layout”.

HTML includes a version of it, TeX and DTP apps use various versions of it.

Edit to add: I don’t know the details of TeX except that it uses a hierarchical “box model”. I suspect it’s fairly similar to HTML before flex-box. HTML flex feels like a refined version of Java AWT layout managers (a more successful evolution than Android’s painful variant).

vbsteven• 5 years ago

The article and parent talk mostly about layout in terms of rectangles and subdividing those into smaller rectangles. The basics like Box and Flex layout are pretty simple.

Once you start adding constraints like min/max size and relative sizes it becomes more complex but still manageable. You’re still only reasoning about rectangles and their constraints.

Adding rendered text in the mix increases complexity immensely. Determining the size of rendered text is a very complex problem space in itself involving fonts, dpi, text flow, code points, LTR/RTL, ligatures and more. It’s not just rectangles anymore.

dahart3• 5 years ago

Yes text layout is fundamentally different in the sense that text layout is about placing lines of text inside a rectangle, where RectCut is about placing the rectangles and doesn’t help much with text. It would be possible to use RectCut for text flow, but you probably wouldn’t want to, it’s more designed for a small number of on-screen panels in an application.

Core LaTeX is concerned with the flow of text in printed articles, and Knuth published the text layout algorithm “Breaking Paragraphs into Lines” http://www.eprg.org/G53DOC/pdfs/knuth-plass-breaking.pdf

Originally, HTML did the same thing, it was used to emulate the written page and layout text into a box. But not everyone wanted their site to look like a printed article, so CSS and new layout modes sprouted over the years. Today there are seven of them: https://developer.mozilla.org/en-US/docs/Web/CSS/Layout_mode

RectCut is fundamentally most similar in spirit to CSS flexbox, and fundamentally similar to a bunch of other box layout managers in other languages and UI toolkits, as you can see from the comments.

cmrdsprklpny• 5 years ago

JUCE also has this kind of layout.

wruza7• 5 years ago

It seems that it positions itself as RAD-like, so not surprising. Boxing and constraints models have one huge downside – your visual interface builder always sucks.

kroltan• 5 years ago

An extension of the fixed layout would be the anchor+margin layout commonly used for game UIs, where each rect is defined by two opposing corners, each of those's axes defined as a % position within its parent, plus a fixed amount of pixels margin.

In this regime, centering a 100px*50px element would be to make TopLeft=(50%-50px, 50%-25px) BottomRight=(50%+50px, 50%+25px)

wruza7• 5 years ago

Another approach I remember uses a pre-sized template and attaches sides to other sides (unlike to parent sides only). E.g. you have a form of width 1000, then you drop button A at (10,10)-(495,50), button B at (505,10)-(990,50), and then connect A.right to follow B.left instead of Form.left (which is default). B.left is then told to be proportional to Form.right, and B.right follows Form.right. When parent resizes, margins and the spacing between them (10) are naturally maintained from the fact that they were 10px apart at design time, i.e. the layout stores their initial coords for using in calculations. A minor issue is that proportional at 505 is not equidistant to proportional at 500 (a true middle of the spacing) at all times, but you just ignore that.

All the rules are of the form:

  A.side {follows|proportional} B.side

By default

  *.side follows Form.left (or top)

And few shorthands, e.g. “A expands hotizontally” means A.right follows Parent.right, etc.

It is actually a dumbed down version of a constraints model and is a subset of it, though it looks and feels more like boxing or alignment models. It was used in pre-constraints apple toolkits and few other products. The main difference is that these follow/proportional constraints are tree-like (cycles forbidden), when true constraint-based layout is a linear problem that requires a special solver optimized for ui-related [in]equality patterns to update in realtime.

fouric2• 5 years ago

I'm curious about the four that you listed. Fixed is relatively obvious, but I don't have a handle on the conceptual models of the other three. Where can I learn more about them without reading through hundreds of pages of webtech and Apple documentation?

wruza7• 5 years ago

Not sure if there is a single resource, but check these links. As you can see, some frameworks mixed layouts and extended size allocation ideas to better fit a given job. (Web has never done this, we're literally stuck with a tree of MSWords)

Constraints-based:

  https://blog.gtk.org/2019/07/02/constraint-layouts/
  https://ebassi.github.io/emeus/
  https://developer.apple.com/library/archive/documentation/UserExperience/Conceptual/AutolayoutPG/AnatomyofaConstraint.html
  https://developer.apple.com/library/archive/documentation/UserExperience/Conceptual/AutolayoutPG/VisualFormatLanguage.html

Boxing:

  https://developer.mozilla.org/en-US/docs/Learn/CSS/Building_blocks/The_box_model#examples_of_different_display_types

Alignment/anchor-based:

  $subj
  https://wiki.lazarus.freepascal.org/Autosize_/_Layout

maskros• 5 years ago

This is also how the Tcl/Tk pack layout works.

cmacleod4• 5 years ago

... and has done for at least 25 years :-)

Sharlin• 5 years ago

As they say, everything old is new again.

blacktriangle• 5 years ago

That was my first thought as well.

I found Tk was really great for simple layouts, but I worked on a Tk project that had some very complicated control panels. Ultimately it became unwieldy and difficult to refactor the UI as moving code around would have breaking consequences.

HelloNurse• 5 years ago

Trees of "dumb" rectangles don't seem very suitable for responsive or adaptive layouts in which the hierarchical structure is built before deciding rectangle sizes. For a retained mode GUI traditional container widgets and layout managers negotiating sizes with child widgets seem a more general technique; for an immediate mode GUI there is no obvious advantage over explicit and more general size computations at the same abstraction level.

nine_k• 5 years ago

From screenshots, this looks like a game console or an embedded controller screen.

For that purpose, simplicity and low footprint may trump flexibility.

no_time• 5 years ago

rect->minx = min(rect->max.x, rect->minx + a);

Did you mean maxx instead of max.x?

jakelazaroff• 5 years ago

This may be good if you know all your rect sizes in advance. But it seems like it will quickly fall apart if you have rectangles that need to change their size based on content, especially if that content is multiple levels deep.

Say you have a floating toolbar the width of two buttons, which themselves are the width of their text contents. You're back to square one — you have to measure the text of each button, then measure both buttons and finally you have the size of the toolbar.

hsn915• 5 years ago

You will generally know the rough content of the text and allocate a size that looks good enough.

A lot of people do this anyway even with css: specify the button width in pixels. Because a toolbar with buttons that have varying widths depending on their content is not always a very good looking toolbar.

RectCut for dead simple UI layouts

Related Stories