[FEAT] Use browser cursors

[felipecrs]

Is this a new feature request?

• I have searched the existing issues

Wanted change

It would be a game changer (for me) if there was an option to render the cursor in client rather than in server.

Reason for change

I run Webtop in a VM in Sweden, accessing from a Windows 365 VDI hosted in Azure/Canada, accessing from my personal desktop in Brazil.

No wonder latency is bad, however, Webtop has provided me the best performance so far out of all options I tried. I'm very thankful for it.

However, because of such latency, what hurts the smoothness the most is that moving the cursor is not 1:1, giving it's rendered in the server.

If the cursor were rendered in client (like Xpra does), the smoothness feel would be a lot better.

Proposed code change

I don't even know in which project this feature would be implemented, I'm just open this issue to start the discussion.

What do you think?

[thelamer]

The cursor is rendered clientside

[felipecrs]

That's weird. What do you think may be going on then? See Xpra in left side and Webtop in right side. Used Powertoys to draw crosshair on the real cursor position.

PowerToys_k1xSmGOddg.mp4

Started with:

$ docker run -d \
  --name=webtop \
  -e PUID=$(id -u) \
  -e PGID=$(id -g) \
  -e PASSWORD \
  -e TZ=America/Toronto \
  -e DISABLE_IPV6=true \
  -p 3001:3001 \
  -v /mnt/disk0/data/webtop/config:/config \
  -v /var/run/docker.sock:/var/run/docker.sock \
  --shm-size=1gb \
  --restart unless-stopped \
  linuxserver/webtop:ubuntu-xfce

[thelamer]

Zero clue I cannot replicate that and I wrote the code for it, the only way you can render the cursor serverside in the current stack is in trackpad mode.

2025-08-25_15-32-04.mp4

[felipecrs]

Just to scratch that hypothesis, trackpad mode only shows up in touch devices, right?

Because I don't see any toggle for it:

[felipecrs]

Just to emphasize that in my test, it's not my first computer directly controlling Webtop. Webtop is being accessed through my second computer, which is being accessed through my first computer via RDP.

I run Webtop in a VM in Sweden, accessing from a Windows 365 VDI hosted in Azure/Canada, accessing from my personal desktop in Brazil.

So maybe this issue can only be reproduced when the client (the browser accessing Webtop) is running from within an RDP session.

I will try testing that with some other local computer here.

[felipecrs]

Yes, that seems to be the case. Maybe the server-side cursor rendering logic gets triggered when the client is running within an RDP session, for some reason.

I was able to reproduce the problem using a laptop to access my PC through RDP, which then accesses a Webtop instance running in Cloudflare close to my region.

Naturally the latency is much better, but I can still reproduce the issue, notice how the cursor is lagging behind the real cursor.

video5001401588405765520.mp4

[felipecrs]

And indeed, I cannot reproduce anything like that when accessing directly from my PC or laptop, only when accessing from a PC that is being accessed through RDP.

[thelamer]

I mean is this really a bug? I don't understand what the problem is I don't really care about multi layered network access like this and it works directly.
I would imagine there is some kind of thing in RDP where it cannot use the predicted cursor api or something. I am not going to patch the code for stuff like this unless you feel up to the task this seems like a "doc it hurts when I do this" situation.

[felipecrs]

I don't understand what the problem is I don't really care about multi layered network access like this and it works directly.

I understand your standpoint as a developer.

But from a user perspective, this means Webtop cannot be used whereas Xpra can.

I mean is this really a bug?

Bug or not, it's an undesired thing.

I am not going to patch the code for stuff like this unless you feel up to the task this seems like a "doc it hurts when I do this" situation.

That is ok. I didn't open the issue to obligate anyone to fix this. My intention is just to open the discuss about it. And yeah, documenting the issue is certainly better than nothing.

You mentioned you wrote the code for the client vs server cursor rendering. Can you point me to some of it? I may try to debug this.

[thelamer]

Server side cursor sending to the client:

https://github.com/selkies-project/selkies/blob/main/src/selkies/input_handler.py

Clientside cursor rendering is mixed between input and core:

https://github.com/selkies-project/selkies/blob/main/addons/gst-web-core/selkies-core.js
https://github.com/selkies-project/selkies/blob/main/addons/gst-web-core/lib/input.js

The option to enable the server side cursor rendering (literally embedded in the stream and used for trackpad mode) is in the core server:
https://github.com/selkies-project/selkies/blob/main/src/selkies/selkies.py
And to process and enable the cursor rendering that is in pixelflux, but I would not worry about that as that is a mobile only thing:
https://github.com/linuxserver/pixelflux/blob/master/pixelflux/screen_capture_module.cpp

[felipecrs]

That is awesome. Thank you so much.

[felipecrs]

I haven't looked into anything yet. Just wanted to say that I found a much easier way to reproduce this: simply try to access Webtop from within Windows Sandbox.

20250829-0327-06.7917801.mp4