Benchmark: snapDOM vs html2canvas | Hacker News Reader

Benchmark: snapDOM vs html2canvas

zumerlab.github.io

47 pointsjmm7716 comments

For the 1st example, snapdom is not only faster, but actually correct, while html2canvas has some artifacts (on Linux/Firefox)

Thank you for mentionig this. There are still some caveats I hope fix soon. And I working on a plugin system to allow user modify all aspects of the capture

Second this. Congratulations for snapDOM by the way. It performs great!

Very cool! How does this work from a technical standpoint? Do you have to implement a full HTML/CSS renderer in JS, or is there an official browser API to capture the elements as image data?

I'm using several functions to clone the DOM and insert it into an SVG <foreignObject>. This approach is now well supported across modern browsers, although <foreignObject> is still a rather primitive way to handle HTML and styles within SVG, which can sometimes lead to results that aren't fully precise in certain cases. To reduce the size, I created internal CSS classes to avoid repeating CSS properties, and I'm also using several caches to prevent re-cloning styles that have already been processed. There's still a long journey ahead, but I hope a plugin system will allow for more fine-grained control over all capture stages

Congrats!

If a render is 13ms (under 16ms), it means it could, in theory, also record page video capture at 60fps or 30fps?

Yes, I'm afraid it depends a lot on what elements you are capturing. But maybe with a plugin designed to perform a capture burst and collect them in an array, it would be possible to generate an animation as output.

SnapDOM can capture HTML elements as SVG - what are use cases for this?

Yes, the default capture output is a svg dataurl. But there many other export options. This lib was made as an internal tool for another project called Zumly that is zoomable engine and the svg format was the faster output I found. But then I realized that could be an independent tool like html2canvas

Zumly seems to be really interesting idea - are there any real world examples?

I really hope there will be, but it's still experimental. If it works well, I think there could be use cases such as dashboards, home automation UIs, charts, or even industrial line control.

Zumly is actually a spinoff of ZircleUI, the first project I ever worked on. Zircle eventually split into Zumly (which is still a work in progress) and Orbit, a CSS tool that makes it easy to create any kind of radial design.

It seems like it would be a lot more efficient to use SVG for screenshot testing design system components rather than raster images.

The SVG in question contains HTML elements as <foreignObject> nodes and thus still depends on the browser to render them. A bitmap has zero dependencies, and can be processed and compared fully outside a browser.

Oh wow, I didn’t realise that. That means SVG output is way less useful than I expected.

Good job! This is very promising

Thank you!

Reply on news.ycombinator.com