x11: Add support for custom cursors #1801
Conversation
Signed-off-by: Uli Schlachter <psychon@znc.in>
Signed-off-by: Uli Schlachter <psychon@znc.in>
|
Hm, I think the x11rb-image dependency is not actually needed here. That code doesn't actually do anything useful here. Future-me should double-check that and possibly remove the dependency. (Note to self) |
|
looks like channels are wrong. Also it is less clear |
| ) -> Result<Cursor, ReplyOrIdError> { | ||
| // BEGIN: Lots of code just to get the image into a RENDER Picture | ||
|
|
||
| // No idea how to sanely get the pixel values, so I'll go with 'insane': |
There was a problem hiding this comment.
ImageBuf::raw_pixels should work
There was a problem hiding this comment.
Yeah, but then I need to handle all the different possible ImageFormat values. The GTK backend does not do that (I guess it only handles two of the four possible values properly), so this seems complicated/untested.
And with raw_pixels, RgbaSeparate still needs to be converted to premultiplied "by hand". The simplest way for that could be the already existing code, I guess.
|
Updated. Here is a screenshot. On the left is the actual cursor and on the right is a screenshot of gimp showing the same image. I added some more colors to the image for testing:
Seems like I got the component order completely wrong. And treating unpremultiplied pixels as premultiplied looks worse than I expected, so I also implemented that todo. The big-endian code is completely untested, but might work. Or not... I have no idea which component order are required.
Also: Thanks a lot for the hint on how to get druid to work with my ancient Rust compiler. |
While piet provides pixels in rgba order, the X11 server wants an u32 in its native endianness with order argb. I even called a variable argb32_format, but I still did not handle this difference. This commit premultiplies the color component with the alpha, handles the different component order and also converts to the server's endianness. Signed-off-by: Uli Schlachter <psychon@znc.in>
The code only used x11rb's image feature for splitting a large PutImage request into smaller pieces. Since X11 servers usually have a maximum request size of 16 MiB these days, it is unlikely that people create cursors large enough to go above this limit. You'd need a 2k x 2k cursor for that! Signed-off-by: Uli Schlachter <psychon@znc.in>
|
It just occurred to me that one could let piet-cairo and cairo handle the "complicated" part of uploading stuff to the X11 server in the right format. I briefly followed down this road, but when I figured out I'd need to call |
|
I am in favor of "existing code is good enough" |
|
😔 conflict resolution failed |
|
Here is my attempt at merging with master. I also added the use for I didn't test anything besides "it builds". |
This adds custom cursor support to the X11 backend, fixing an open todo from #1755.
Half of the added code is "RENDER initialisation" (is the necessary RENDER version supported? What is the id of the standard ARGB32 format?). The other half of the code "upload this image to the X11 server". I guess the later could be simplified somehow, but I was lazy and did it "manually". This could also be made faster, but I don't think creating cursors is performance critical...
Oh and: Nothing ever deletes custom cursors, so they stay around forever, eating memory in the X11 server. I wonder if one can make the X11 server go out-of-memory with cursors...
This code is completely untested since I can't build druid and thus cannot run the cursor example due to a too old Rust compiler.