Prefer use of gdk_memory_texture_new()

[kleisauke]

As the gdk-pixbuf dependency is expected to be dropped in GTK5, see:
https://gitlab.gnome.org/GNOME/gtk/-/merge_requests/8061

[jcupitt]

Ah, nice! I'll copy this over to nip4 as well.

[jcupitt]

libvips/nip4@d2c967a

[kleisauke]

Great!

BTW, I just came across this GTK merge request: https://gitlab.gnome.org/GNOME/gtk/-/merge_requests/8327. The issues mentioned there are very similar to the ones we've encountered with visible tile boundaries (e.g. when opening this zone plate image).

[kleisauke]

... i.e. we could probably remove the +1 here:
https://github.com/jcupitt/vipsdisp/blob/13560a8f2fe0020b64e7ef2faa08e9c355ada466/src/tilecache.c#L1026-L1030

with that merge request (untested).

[jcupitt]

Oh, interesting!

vipsdisp has much better seam hiding now, but you can still see them under some circumstances.

It was going to int for the tile position at various points, but now it's now double all the way through, plus a 1 pixel tile stretch during final render. The one pixel stretch hides the joins, but obviously can make a visible step with some images. A 0.5 pixel stretch almost works, but not quite.

It's these lines, if you're curious:

https://github.com/jcupitt/vipsdisp/blob/master/src/tilecache.c#L1026-L1030

[kleisauke]

It seems to work! Using commit kleisauke@9ecdc97 and merge request https://gitlab.gnome.org/GNOME/gtk/-/merge_requests/8494 (with commit https://gitlab.gnome.org/GNOME/gtk/-/commit/97af44e6704a6c22593851efb761e92e392b9c39 cherry-picked to avoid assertion failures), I see:

Old	New
(click to enlarge)	(click to enlarge)

[jcupitt]

That's great! Looks like they are hoping to land that change for gtk 4.20, so fingers crossed!

I guess we could post a +1 on that draft, but it sounds like they are doing it anyway.

[kleisauke]

Indeed, hopefully that will land in GTK 4.20. I just opened draft PR #47 to track this.