wayland: do a roundtrip to wait for image description creation

[llyyr]

No description provided.

[kasper93]

Works, thanks!

[github-actions]

Download the artifacts for this pull request:

Windows

• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-aarch64-pc-windows-msvc
• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-aarch64-pc-windows-msvc-pdb
• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-i686-w64-mingw32
• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-x86_64-pc-windows-msvc
• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-x86_64-pc-windows-msvc-pdb
• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-x86_64-w64-mingw32

macOS

• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-macos-14-arm
• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-macos-15-arm
• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-macos-15-intel
• mpv-v0.41.0-dev-g74143fc01-pr17495-22603585021-macos-26-arm

[Dudemanguy]

Looks fine.

[mahkoh]

I don't believe this is correct. wl_display_roundtrip_queue performs one roundtrip on the queue. But if the image_description callbacks are not called within that roundtrip, the queue will never be dispatched again and the callbacks will never be invoked.

[llyyr]

I'm aware, but this was sufficient on every compositor I tested on. In fact the VK_hdr_layer layer also does a roundtrip like this.

Of course I'm perfectly fine with doing a dispatch until the callback is invoked, but there was some aversion among mpv maintainers with that approach. @Dudemanguy thoughts?

[mahkoh]

Am I trippin or didn't you add the "VO is ready" state to handle the image description not being ready while allowing the rest of mpv to run normally?

[llyyr]

didn't you add the "VO is ready" state

There were several problems with it. If the callback didn't get invoked within the next two draw_frame calls, mpv would end up in a permanent deadlock. Fixing it would be far more complex than it is worth. I didn't like the idea of littering the rest of the codebase just because Wayland being special. It's better for Waylandisms to be contained in one place instead of in mpv core.

[mahkoh]

Mesa loops until the image description is ready. If that was good enough for waylandvk until now, then it should also be good enough for the rest of mpv.

[llyyr]

Mesa loops until the image description is ready. If that was good enough for waylandvk until now, then it should also be good enough for the rest of mpv.

I completely agree, but I'm not mpv maintainer and I've already made that PR before and it was rejected. Something like llyyr@d330627 would be my preference

[kasper93]

Mesa loops until the image description is ready. If that was good enough for waylandvk until now, then it should also be good enough for the rest of mpv.

I completely agree, but I'm not mpv maintainer and I've already made that PR before and it was rejected. Something like llyyr@d330627 would be my preference

I don't want to add an api that spans three layers of abstraction, only to wait on some internal Wayland queue.

Why not just put while (wl->image_description_pending) if (wl_display_dispatch_queue(wl->display, wl->color_queue) < 0) break; instead of round trip, which is apparently not enough?

Alternatively, we can wait in other entry points, like flip_page or draw_frame.

[llyyr]

Why not just put while (wl->image_description_pending) if (wl_display_dispatch_queue(wl->display, wl->color_queue) < 0) break; instead of round trip, which is apparently not enough?

This is fine too, in fact that was the first version of the PR I made. I can change it to that if that's fine, it's exactly what Mesa does