Add explicit display backend configuration - #345
Conversation
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
| } | ||
|
|
||
| func (s *ApiService) doPressKey(ctx context.Context, body oapi.PressKeyRequest) error { | ||
| if s.usesWayland() { |
There was a problem hiding this comment.
Mouse position still uses xdotool
Medium Severity
GetMousePosition always runs xdotool getmouselocation and is not routed through the Wayland CDP input path. With DISPLAY_BACKEND=wayland there is no X display, so the endpoint returns 500 instead of the pointer position.
Reviewed by Cursor Bugbot for commit 793b3fb. Configure here.
| } | ||
| final := chromiumflags.MergeFlagsWithRuntimeTokens(baseFlags, runtimeTokens) | ||
| final = withDefaultPrivateNetworkBypass(final) | ||
| final = append(final, displayFlags...) |
There was a problem hiding this comment.
Ozone flags overwrite enable-features
Medium Severity
Wayland display flags are appended after merged CHROMIUM_FLAGS, including a separate --enable-features=UseOzonePlatform. Chromium last-token-wins for --enable-features, so that token replaces any earlier feature list from base or runtime flags rather than merging with it.
Additional Locations (1)
Triggered by learned rule: Chromium flag merging must preserve last-token-wins ordering
Reviewed by Cursor Bugbot for commit 793b3fb. Configure here.
| events = append(events, cdpclient.MouseEvent{Type: "mouseMoved", X: float64(point[0]), Y: float64(point[1])}) | ||
| } | ||
| end := body.Path[len(body.Path)-1] | ||
| events = append(events, cdpclient.MouseEvent{Type: "mouseReleased", X: float64(end[0]), Y: float64(end[1]), Button: button, ClickCount: 1}) |
There was a problem hiding this comment.
Wayland drag omits pressed buttons
Medium Severity
Wayland drag sends mouseMoved events with coordinates only. CDP Input.dispatchMouseEvent needs a non-zero buttons bitmask (and usually button) on moves between press and release, otherwise pages see hover rather than a drag.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit 793b3fb. Configure here.
| }); err != nil { | ||
| return &executionError{msg: fmt.Sprintf("failed to press Wayland key: %v", err)} | ||
| } | ||
| return nil |
There was a problem hiding this comment.
Wayland key path ignores hold duration
Medium Severity
doPressKeyWayland immediately emits paired keyDown/keyUp and never reads duration or hold_keys. Unlike click types, which are rejected as unsupported, those fields are silently dropped so held keys and modifiers do not behave as on X11.
Reviewed by Cursor Bugbot for commit 793b3fb. Configure here.
| } | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
Recording dies on first capture error
High Severity
streamFrames returns on the first CaptureFrame or write error, which closes ffmpeg stdin and ends the Wayland recording. Transient CDP failures (devtools not ready, navigation, Chromium restart, oversized screenshot frames) therefore stop capture for good instead of retrying, with no error recorded on the recorder.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit 5a7066d. Configure here.
| captureClient = nil | ||
| } | ||
| return frame, err | ||
| } |
There was a problem hiding this comment.
Hung capture blocks all recordings
Medium Severity
Wayland CaptureFrame holds a process-wide mutex for the full CDP screenshot and uses a non-cancelable context with no timeout. If DevTools stalls, streamFrames never observes exited, the mutex stays locked, and later recordings (and further frames) block indefinitely.
Additional Locations (2)
Reviewed by Cursor Bugbot for commit 5a7066d. Configure here.
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.
There are 7 total unresolved issues (including 6 from previous reviews).
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit c415be4. Configure here.
| @@ -0,0 +1,8 @@ | |||
| [program:weston] | |||
| command=/bin/bash -c 'exec weston --backend=headless-backend.so --use-pixman --debug --shell=desktop-shell.so --socket="${WAYLAND_DISPLAY:-wayland-0}" --width="${WIDTH:-1920}" --height="${HEIGHT:-1080}" --idle-time=0' | |||
There was a problem hiding this comment.
Desktop shell restores window chrome
Medium Severity
Switching Weston from kiosk-shell.so to desktop-shell.so drops the always-fullscreen kiosk behavior. Desktop shell defaults to a top panel and window decorations, so Chromium no longer fills the configured WIDTH×HEIGHT output. Compositor capture then includes that chrome, and CDP viewport size can fall short of the intended display (for example 1920×1024 versus 1920×1080).
Reviewed by Cursor Bugbot for commit c415be4. Configure here.
1 participant
summary
validation
-Wall -Wextra -Werrordependency
The default Neko image remains unchanged until the companion Wayland backend change is released. Wayland WebRTC stays disabled in deployed images until that dependency is available and end-to-end live-view validation passes.
Note
High Risk
Touches display startup, Chromium launch, input/screenshot APIs, and recording capture. Wayland paths change core computer-control and media pipelines; a misconfigured backend can break headful sessions.
Overview
Adds an explicit
DISPLAY_BACKEND(x11default,wayland) so headful Chromium can run on Weston instead of X11, without silently falling back.On Wayland, the wrapper starts Weston, Chromium uses Ozone/Wayland, and computer APIs (mouse, keys, type, scroll, drag, screenshot, resize) go through CDP rather than xdotool/X11. Recordings and CDP-monitor screenshots also capture via CDP PNG frames instead of
x11grab.The image builds a Weston screenshooter helper (
weston-capture) and Neko is started with Wayland capture flags when that backend is selected. The Neko base image is parameterized for local validation; X11 remains the production default.Reviewed by Cursor Bugbot for commit a38b174. Bugbot is set up for automated code reviews on this repo. Configure here.