This is an attempt to add a way for EGL clients to tell software renderers like llvmpipe to render to an existing chunk of memory. The end goal is to allow llvmpipe to work with wlroots, which uses only EGLImageKHR and never uses EGLSurface.
Clients create an EGLImageKHR with the target EGL_HOST_POINTER_MESA, and a pointer to the chunk of memory in buffer. Width, height, stride and format are provided as attributes.
Is this an acceptable approach?
Prior art: VK_EXT_external_memory_host