feat(wayland): implement ext-image-capture-source-v1 and ext-image-copy-capture-v1 protocols #1902
Conversation
Implements ext-image-capture-source-v1 and ext-image-copy-capture-v1 protocols for screen capture functionality. Design uses callback-based extensibility with UserDataMap, allowing compositors to attach custom data to capture sources. This enables COSMIC-style workspace capture through custom globals without infecting the core API with generics. Key features: - Full session/frame lifecycle with RAII cleanup - Buffer validation (size, format, modifier) before frame callbacks - Cursor session support for pointer capture - Anvil integration demonstrating handler implementation Based on cosmic-comp's screencopy implementation by Victoria Brekenfeld and Ian Douglas Scott at System76. Closes Smithay#1900
glamberson
changed the title
The get_dmabuf import is only used in validate_dmabuf() which is gated by #[cfg(feature = "backend_drm")]. Without this gate, the import is unused when building with use_system_lib but without backend_drm, causing a compile error with -D warnings.
Split image_capture_source into separate State + Handler per source type: - ImageCaptureSourceState + ImageCaptureSourceHandler: Core source handling - OutputCaptureSourceState + OutputCaptureSourceHandler: Output capture - ToplevelCaptureSourceState + ToplevelCaptureSourceHandler: Toplevel capture This allows compositors to choose which source types to support independently. Compositors with custom foreign-toplevel implementations can skip the toplevel module entirely and handle the protocol directly. Each module has its own delegate macro for cleaner integration.
Drakulix
left a comment
Drakulix
left a comment
There was a problem hiding this comment.
This looks mostly good for, I think the latest design iteration covers everything we need for custom implementations.
For merging this still needs a more in-depth review to make sure this covers everything cosmic-comp and potentially niri need.
Which clients did you test the anvil implementation with?
So is this PR in a "ready" state to implement the protocol in other compositor? I want to implement this protocol in Niri and maybe I'll give it a go with this PR. |
Well, yes, presumably, although I"m testing still and can't probably get through all of it until morning. |
|
So far this looks good. I'm working on updating |
|
@ids1024 Great. I wanted to run this through testing but I just haven't had a minute to get to it. So I'm going to mark it as ready with the proviso it should undergo more testing. |
The BufferConstraints.dma field only exists when smithay is built with backend_drm. Gate the field assignment in anvil behind the features that enable backend_drm (udev or winit) to fix --no-default-features build.
|
Seems to be working (though that's not surprising given this is based on the cosmic-comp version). I was able to adopt the So it seems to meet our requirements, and I think should suit the requirements of other compositors. |
I've actually been contemplating this for a bit since I'm using screencopy-v1 and I think that's an important feature. While I haven't tried using it there seem to be a lot of desktop abstractions built in by default, so I appreciate that there seems to be an option to bypass that without having to just write the protocol implementation myself. |
Yes this is going to save me a lot of time. Thanks @ids1024 and @Drakulix for allowing me to adapt your work! |
External source implementations (e.g., workspace capture) need to call add_instance() to associate protocol resources with custom sources.
Drakulix
left a comment
Drakulix
left a comment
There was a problem hiding this comment.
This PR looks good to me and is imo okay to merge now.
Slight but however:
- I'd love to see a proper anvil implementation, that doesn't just fail the actual capture. Anvil has a standable
render_output-function and the implementation doesn't have to be perfect (anvil itself is far from it). It could simply re-render the output, which would work for all backends.
But our merge-requirement has always been an implementation proving the code works in any compositor. And since @ids1024 already adopted cosmic-comp, this has been done.
- I'd like to hear some input on this from @YaLTeR, given niri seems to have some interest in implementing the protocol eventually. Just to hear if the overall design seems to be a good fit for niri now as well.
The x11 feature enables smithay/backend_vulkan which pulls in backend_drm, so the dma field exists and must be initialized.
Looks alright at a glance. I was adding some weird code to my wlr-screencopy impl over the past days but it seems that none of that code is necessary for these ext protocols because of the much improved protocol design. So I expect I'll be able to use this Smithay impl. Can't say definitively without trying though. |
7 participants
Implements ext-image-capture-source-v1 and ext-image-copy-capture-v1 protocols for screen capture functionality.
Design uses callback-based extensibility with UserDataMap, allowing compositors to attach custom data to capture sources. This enables COSMIC-style workspace capture through custom globals without infecting the core API with generics.
Key changes:
This implementation is adapted from cosmic-comp's screencopy module, originally authored by Victoria Brekenfeld (@Drakulix) and Ian Douglas Scott (@ids1024) at System76. Re-licensed to MIT with their approval per discussion in #1900.
Closes #1900