Offset rendering on Linux since glutin upgrade

[FauxFaux]

Since 5de27b0 ("update glutin and gl_generator"), on Ubuntu Artful (16.10) with nvidia-384, things render off-centre. Here's the cargo run --example image:

Here's what it should look like. Same machine, but glium at commit 63577da, before the updates. Note the (r) to the right of the L, and the even margin:

It's nothing to do with the window size, it happens at all sizes; the discrepancies in the above screenshots are my laziness.

Not super visually significant in this example, but presumably related to PistonDevelopers/conrod#1120 which is awful.

I am still unable to find a pure glutin example that shows the same behaviour.

[tomaka]

Probably a HiDPI issue, as always.

[FauxFaux]

I have large monitors (2 * 2560x1440), but they're not HiDPI by the normal standards, nor do I have any scaling config as far as I can remember. Issue happens in Gnome and in i3, which should use different scaling settings.

% xdpyinfo | grep -B 2 resolution
screen #0:
  dimensions:    5120x1440 pixels (1204x342 millimeters)
  resolution:    108x107 dots per inch

[spearman]

Reverting the change to the X11 window get_inner_size method made in (rust-windowing/winit@4890229) seems to fix it:

spearman/winit@d3cc220

Not sure if that's the real problem or if the change was intentional and the problem is elsewhere.

[tomaka]

Hmm, maybe glium applies the HiDPI stuff on top of glutin.

[daboross]

I've been trying to replicate this on my computer running Ubuntu 16.10, but I haven't found any setting which makes DPI reported by xdpyinfo anything other than 96x96. Was there some setting you had to change to get it to 108x107 that I could also change?

[FauxFaux]

Ah, right, I didn't even try changing the DPI manually. If I run xrandr --dpi 200 it gets much worse, as expected:

And xrandr --dpi 96 fixes it to look like the reverted commit.

[daboross]

Ah cool! I can also reproduce this now with xrandr --dpi 200 on Xorg.

[daboross]

Alright, thank you! I think I've found the culprit in glium: it accounts for get_inner_size() returning dpi-scaled points, while the commit in winit mentioned earlier changes it to return absolute pixels. (code at https://github.com/glium/glium/blob/master/src/backend/glutin/mod.rs#L265).

This was much easier to find with the commit you found, @spearman and the isolation of the issue to the winit upgrade and dpi scaling you found, @FauxFaux.

This seems to really be just winit making a breaking change in what get_inner_size meant with the 0.9 upgrade, and the associated upgrade in glium not accounting for that change.