fix(linux/kms): resolve connector names in output_name config#4762
fix(linux/kms): resolve connector names in output_name config#4762itsDNNS wants to merge 4 commits intoLizardByte:masterfrom
Conversation
When a user sets output_name to a DRM connector name like "DP-2" or "HDMI-A-1", util::from_view() silently converts the non-numeric string to a garbage integer (e.g. "DP-2" becomes 23172), causing display init to fail with "Couldn't find monitor [23172]". This is a UX trap because Sunshine's own KMS logs show connector names (e.g. "Monitor 1 is DP-2: ..."), leading users to copy those names into their config. Add resolve_display_name() that accepts both numeric indices (backward compatible) and DRM connector names. Connector names are parsed using the existing kms::from_view() and matched against card_descriptors already populated by kms_display_names().
New config option `pre_probe_cmd` (same JSON syntax as `global_prep_cmd`) that runs before encoder probing. This allows enabling a virtual display on demand rather than keeping it active permanently. The "do" commands run before display configuration and encoder probing in launch() and resume(), with client parameters exposed as environment variables (SUNSHINE_CLIENT_WIDTH, HEIGHT, FPS, HDR). The "undo" commands run on probe failure, cancel, and session disconnect.
When pre_probe_cmd enables a virtual display (e.g. via kscreen-doctor), the connector is not yet present in card_descriptors which were populated at startup. resolve_display_name() now retries once after refreshing the display list via kms_display_names(), allowing dynamically enabled connectors to be resolved.
When output_name is a DRM connector name like "DP-2", it won't match the numeric display_names list (["0", "1", ...]). Without this fix, the code falls through to index 0, streaming the wrong monitor. Add a fallback that appends the connector name to display_names when no numeric match is found, allowing display_t::init() to resolve it via resolve_display_name().
|
|
Thank you for the PR. I'm assuming since you didn't use our PR template this is AI generated? Can you please use our PR template? https://github.com/LizardByte/.github/blob/master/.github/pull_request_template.md?plain=1 |
ReenigneArcher
added
the
ai
|
Thanks for the feedback! Updated the PR description with the official template. To answer your question: yes, the implementation was done with AI assistance (Claude Code / Opus 4.6). I designed the approach and did all testing on my setup (virtual monitor via custom EDID on KMS/Wayland), Claude helped with the actual code changes. Marked as "Heavy" in the AI Usage section for full transparency. |
|
I'll say we've needed this feature forever, but the pr's have just been disappearing for years. I was just thinking of doing it yesterday too. Needs to work with wlgrab and actual virtual displays though, i.e. HEADLESS-3 from I'm currently porting the wlr-screencopy code to prefer the new ext-image-copy-capture proto that will replace it, since Hyprland merged icc yesterday. I could also add this feature to wlgrab on the side... |
3 participants
Description
This PR adds support for DRM connector names in
output_nameand introducespre_probe_cmdfor virtual display lifecycle management on Linux/KMS.Commit 1 - fix(linux/kms): resolve connector names in output_name config
When
output_nameis set to a DRM connector name likeDP-2,util::from_view()silently converts it to a garbage integer. Addedresolve_display_name()that accepts both numeric indices (backward compatible) and connector names.Commit 2 - feat: add pre_probe_cmd config option
New config option (same JSON syntax as
global_prep_cmd) that runs commands before encoder probing. Enables virtual display on-demand lifecycle: enable on stream connect, disable on disconnect. Client parameters exposed as env vars (SUNSHINE_CLIENT_WIDTH,HEIGHT,FPS,HDR).Commit 3 - fix(linux/kms): refresh card_descriptors when connector not found
When
pre_probe_cmdenables a virtual display, the connector is not yet incard_descriptors(populated at startup).resolve_display_name()now retries once after refreshing the display list.Commit 4 - fix(video): add connector name fallback in refresh_displays
When
output_nameis a connector name, it won't match the numericdisplay_nameslist, causing the wrong monitor to be streamed. Added fallback that appends the connector name directly.Use case: Virtual monitor driven by custom EDID firmware, kept disabled when not streaming. On stream connect,
pre_probe_cmdenables the display viakscreen-doctor, on disconnect the undo command disables it.Example
sunshine.conf:Screenshot
N/A (no UI changes)
Issues Fixed or Closed
Roadmap Issues
Type of Change
Checklist
AI Usage
AI tool used: Claude Code (Opus 4.6). The concept, architecture, and testing were done by the author; Claude Code assisted with implementation, debugging, and commit messages.