Simon Ser

When one of the `SRC_*` properties changes, we don't necessarily need to perform a re-allocation. We should probably still re-alloc when enabling/disabling scaling, because some drivers can't always do scaling.

Skip setting properties that haven't changed

1

~~Atomic test-only commits set property values for the next commit when they succeed.~~ We don't need to keep setting the same property values over and over again.

Better support for KMS enums and bitfields

2

Enum and bitfield properties may have different definitions across planes. This would be silly and annoying, but could happen. For instance, plane 1 may have `"rotation": bitmask {"rotate-90" = 0x1,...

Each plane can advertise the set of enum/bitfield values it supports. When applying layer properties to a plane, mark the plane as incompatible if the layer value isn't advertised by...

Migrate to gitlab.freedesktop.org

1

gitlab.freedesktop.org would be better suited than GitHub for this project. We need support for our CI (builds.sr.ht) before migrating.

This would be useful for debugging.

Fast access to basic properties

1

Have a list of basic properties and provide fast access to them. These are: - CRTC_ID - FB_ID - CRTC_X, CRTC_Y, CRTC_W, CRTC_H - zpos

Our current plane allocation algorithm is per-output. This means all planes will go to the first output. Instead, we want to split planes across outputs depending on each layer's priority.

https://github.com/emersion/libliftoff/issues/21 was about exposing a per-output API to users. Now that our API allows for multiple outputs, we still need to manage and allocate planes at the device layer. Basically,...

Benchmarks

8

This would allow us to figure out whether an optimization is worthwhile and to prevent big performance regressions.