Wrong viewport on web+mobile

[asny]

Open this example in a browser on mobile and see that the green line is not going from top left corner to bottom right corner of the browser viewport.

The problem is that Window::innerWidth() used here includes the browser header and similar. Same for height. It's only a problem on mobile though. It seems like visualViewport might be correct, but that's not exposed by the web_sys crate.

[huand]

I guess this is also why object are not centered and aspect ratio is wrong? (see that the circle is not really a circle on the Shapes example on mobile)

[asny]

Yeah, I think so. Any help with this is much appreciated, I basically just need a way to get the correct viewport (without the browser header) from the web API, the rest is simple.

[huand]

Hi again, sorry i dont come with a solution, just some observations i just did:

• it seems that the .innerWidth() / .innerHeight() works as expected. For example in the line you mentioned, if you scale it by 0.9 you can see that it takes 90% of the height (and if you go 95% the remaining space shrink by half etc until 100%)
• When reducing the canvas size by a factor then we have the correct aspect ratio, see image
• When increasing the factor, then we have this issue. Could it be a limitation from the canvas, like hitting maximum size? see the image with a big factor

let k = 10.;
html_canvas.set_width((k * browser_inner_width) as u32);
html_canvas.set_height((k * browser_inner_height) as u32);

Also I'm a bit surprised by the canvas size.

log::warn!(
    "browser inner width/height: {} --- {}",
    browser_window.inner_width().unwrap().as_f64().unwrap(),
    browser_window.inner_height().unwrap().as_f64().unwrap()
);
log::warn!(
    "canvas width/height: {} xxx {}",
    html_canvas.width(),
    html_canvas.height()
);

gives the log (in pc browser)

browser inner width/height: 1423 --- 971
canvas width/height: 2846 xxx 1942

is it normal that they are not in the same "unit"? it seems that the inner values are logical pixel but canvas physical (factor 2 between the 2) but the function HtmlCanvasElement::Width() should actually gives logical pixel if i understand it correctly. So this would mean that we have a canvas actually too big by pixel_ratio factor. Would that be possible?

[huand]

So going to the original live demo on mobile, and halving my screen size by 2 with the split screen phone feature, actually the example works (ratio + accurate touching)

You can see in the upper half of the image the the correct view.

I dont know where to go from this ? any idea?

[huand]

so i went on with this : browser inner size == canvas size.
the change required then are for the viewport function + GUI mouse position and then it seems functional for me but i'm not sure that this is the correct fix?

[huand]

after Xmas I can open a PR for this so some more knowledgeable people could review this, as usability/fidelity on mobile web could be a nice to have for some projects.

[asny]

Thank you so much for looking into this @huand 🙏

I've spend a lot of time on these kind of problems, especially when I handled the canvas myself. A year ago or something, I realized that winit started supporting web so I removed my own code and started using that. I since realized that winit web support is not very good 😕 so it is possible that this is a bug in winit.

However, I think it's something else. To make the canvas cover the entire document I set the CSS margins to zero in index.html. That might be the issue since that will make the canvas be underneath the header as well? And the width and height that we retrieve to set the viewport is without that header and then the canvas width height is different from the screen viewport and we get the problem?

About canvas physical size, if the dpr is 2, the canvas size needs to be twice as big as the canvas logical size in the document, otherwise we'll get a pixelated result. It is very confusing, took me a long time to understand that.

The split screen experiment might be better than without split screen but it is still wrong. The green line doesn't hit the corner exactly.

I'll be very happy to review a PR, though I'm not sure I'm more knowledgeable than you 😄 just so you know, I'm perfectly ok with a hack if that's what it takes, just as long as it doesn't add a lot of code.

[huand]

For info, I saw in winit changelog a lot of changes lately. Maybe there is some stuff related to this issue? It might be worth it to try to upgrade this dependency?