ext-image-copy-capture-v1: scale not specified race-free

The ext_image_copy_capture_session_v1 interface has a ::buffer_size event to specify the size of the buffer to be attached; however, neither it nor ext_image_copy_capture_frame_v1 provide information about what the logical size of the capture source is at the time the frame is captured. (This information is useful when capturing multiple outputs at the same time or when recording a resizing window.) When capturing outputs, the logical size can be estimated by looking at zxdg_output_v1::logical_size; however, that event is not synchronized with the frame capture operations.

For example, say the following sequence of events happens on a system with a single wl_output.

• To start: Output scale is 1.5, logical size is 1000x1000 (according to zxdg_output_v1::logical_size), and physical size is 1500x1500 (according to wl_output::geometry.)
• Image copy capture session created for the output
• ext_image_copy_capture_session_v1::buffer_size(1500,1500) is sent
• In parallel:
  • Output scale changes to 3 and logical size is reduced to 500x500; physical size remains unchanged. A new zxdg_output_v1::logical_size is sent.
  • client requests ext_image_copy_capture_session_v1::create_frame, attaches and damages a 1500x1500 buffer, sends ext_image_copy_capture_frame_v1::capture

The compositor may have received ext_image_copy_capture_frame_v1::capture either before or after the output scale change, and because the same buffer size works in both cases, could have sent a frame from either before or after the output scale change. The screen copy operation may be done asynchronously, so that in either case, ready may arrive after the zxdg_output_v1::logical_size message. As a result, the client has no way to tell what the logical size of the content is for a specific frame.

Possible solutions:

• introduce sequence numbers for captured frames and zxdg_output_v1 (may not work for foreign toplevel capture sources)
• Add a logical_size event to ext_image_copy_capture_frame_v1 (Or, to be more precise when fractional scaling is used and the mapping from physical to logical pixels can be ambiguous, a fractional_scale event.)