try our own tile snapping #51
Conversation
|
Referring issue libvips/build-win64-mxe#77 |
|
It should be 1.0 / scale, oops. It's working on linux now with hardware desktop scaling, including fractional scaling! Edge tiles are not being snapped to image edges, so you get annoying strips of background appearing around the sides if the image is smaller than the window. I'm not sure how to fix that. |
|
I made a test release of this branch to make builds for other platforms simpler: https://github.com/jcupitt/vipsdisp/releases/tag/v4.1.0-tile-snap There's a win binary on that page. |
Ah of course, edge tiles ARE being snapped to image edges, it's the rendering of the background that isn't. I'll snap that as well. |
|
It's working on windows and linux! But there's a problem with test images like this (1024 x 1024 pixels of a 1 pixel checkerboard): 
At 200% desktop scaling and 100% vipsdisp magnification, 256x256 image pixel tiles are being drawn as 256x256 "virtual" pixel tiles. These get stretched by gtk scaling and become 512x512 display pixels, with (I think?) trilinear interpolation. You get a blurry mess! 
The 0 / 255 checkerboard becomes mostly 96 / 159, but with brighter bits at tile edges. I think we need to draw tiles smaller by the surface scale factor to get 1 image pixel == 1 display hardware pixel. I'll have a go. |
|
There's also a slight "jelly" effect if you load a large image, select 1:1, and pan sideways by holding a cursor key. Maybe rendering to the display hardware pixel size would fix this too. |
but the checkerboard PNG still fails to render correctly
|
I scaled tiles quickly, but I don't get 1:1 image pixels to display pixels. Maybe |
|
I moved the high-dpi experiment to another branch: https://github.com/jcupitt/vipsdisp/tree/add-high-dpi I'll merge this, it's much better than the simple +1 we had before. |
Experiment with our own tile snapping.
Seem to work on linux, needs testing elsewhere, ep. on win with non-100% desktop scaling.