Screen share from a Linux sharer arrives at 720p although 1080p was chosen, with black bars #876

Open
opened 2026-11-18 06:47:36 +00:00 by charles · 4 comments
Owner

Reported by Charles, 2026-11-18, while watching a share from a Linux (Wayland) user in Flaxative:

  • "I still see black bars" on that stream (a Windows user's share watched earlier the same day was clean);
  • "Resolution of the stream seems to be off? He selects 1080 and it's on 720p on the stream according to Flaxative (my client)".

Not yet known: the sharer's Flaxative version, GPU, desktop, monitor resolution and aspect ratio; whether the bars are steady (letterbox/pillarbox) or flicker; where they are.

Leads to check in the code, not assumed.

  1. A software clamp. PR #873's report mentions a "720p30 software clamp" that applies when a codec has no hardware encoder. If the sharer's machine has no usable hardware encoder (an AMD/Intel GPU on a nightly before the VAAPI fix of #865, or no GPU encoder at all), choosing 1080p would silently become 720p30. Find that clamp: when it applies, whether the UI tells the sharer, what the viewer's stream-quality label shows.
  2. Aspect ratio. When the captured monitor or window is not 16:9 (16:10, ultrawide, a window such as 1920×1046) and the share is scaled to a fixed 1280×720 or 1920×1080, is the picture letterboxed or pillarboxed with black bars baked into the encoded frames, or scaled preserving the aspect ratio with matching dimensions? Steady black bars would come from here, and would be a different defect from #589 (rows of another frame at the top).
  3. Simulcast layer on the viewing side. A sender's log from 2026-11-16 shows the server asking only for the low layer (subscribed_codecs="h264:[Low=true, Medium=false, High=false]") while a Flaxative viewer watched. Check how the Flaxative viewer chooses the layer (adaptive stream by tile size, a fixed quality, the focus view vs the grid tile), whether it ever asks for the high layer when the stream is large on screen, and what the resolution label reads.
  4. Which of the above the nightlies since 2026.1118.369 already changed (NVENC above 720p30, VAAPI, encoder order).
**Reported by Charles, 2026-11-18, while watching a share from a Linux (Wayland) user in Flaxative:** - "I still see black bars" on that stream (a Windows user's share watched earlier the same day was clean); - "Resolution of the stream seems to be off? He selects 1080 and it's on 720p on the stream according to Flaxative (my client)". Not yet known: the sharer's Flaxative version, GPU, desktop, monitor resolution and aspect ratio; whether the bars are steady (letterbox/pillarbox) or flicker; where they are. **Leads to check in the code, not assumed.** 1. **A software clamp.** PR #873's report mentions a "720p30 software clamp" that applies when a codec has no hardware encoder. If the sharer's machine has no usable hardware encoder (an AMD/Intel GPU on a nightly before the VAAPI fix of #865, or no GPU encoder at all), choosing 1080p would silently become 720p30. Find that clamp: when it applies, whether the UI tells the sharer, what the viewer's stream-quality label shows. 2. **Aspect ratio.** When the captured monitor or window is not 16:9 (16:10, ultrawide, a window such as 1920×1046) and the share is scaled to a fixed 1280×720 or 1920×1080, is the picture letterboxed or pillarboxed with black bars baked into the encoded frames, or scaled preserving the aspect ratio with matching dimensions? Steady black bars would come from here, and would be a different defect from #589 (rows of another frame at the top). 3. **Simulcast layer on the viewing side.** A sender's log from 2026-11-16 shows the server asking only for the low layer (`subscribed_codecs="h264:[Low=true, Medium=false, High=false]"`) while a Flaxative viewer watched. Check how the Flaxative viewer chooses the layer (adaptive stream by tile size, a fixed quality, the focus view vs the grid tile), whether it ever asks for the high layer when the stream is large on screen, and what the resolution label reads. 4. Which of the above the nightlies since 2026.1118.369 already changed (NVENC above 720p30, VAAPI, encoder order).
Author
Owner

2026-11-18: the sharer's GPU is an NVIDIA RTX 4080 (Linux, Wayland; Flaxative version still unknown). So the software 720p30 clamp cannot come from a GPU without a hardware H.264 encoder; if it applies, it comes from the NVENC probe or encoder failing on that machine (driver/library, pre-#838 level bug, mid-share fallback). Being checked in the #876 investigation.

2026-11-18: the sharer's GPU is an **NVIDIA RTX 4080** (Linux, Wayland; Flaxative version still unknown). So the software 720p30 clamp cannot come from a GPU without a hardware H.264 encoder; if it applies, it comes from the NVENC probe or encoder failing on that machine (driver/library, pre-#838 level bug, mid-share fallback). Being checked in the #876 investigation.
Author
Owner

2026-11-18, from Charles (sharer: Linux Wayland, RTX 4080; viewers on Flaxative): "When he switches to source, it goes to 1440p, when he goes to 1440p it shows 1080p (on the receiving clients)", and 1080p shows as 720p. Every setting arrives one step lower, so this is not the software 720p30 clamp (that would hold every setting at 720p30). The sharer's monitor resolution is not yet known.

2026-11-18, from Charles (sharer: Linux Wayland, RTX 4080; viewers on Flaxative): "When he switches to source, it goes to 1440p, when he goes to 1440p it shows 1080p (on the receiving clients)", and 1080p shows as 720p. Every setting arrives one step lower, so this is not the software 720p30 clamp (that would hold every setting at 720p30). The sharer's monitor resolution is not yet known.
Author
Owner

Hypothesis and result (measured, PR #879). Hypothesis: nothing is downgraded. Each setting is a 16:9 box, the 3440×1440 picture is fitted into it with its aspect kept, and the viewer's pill names a stream by the closest standard height.

Measured with run.sh --chain --setting (real Wayland capture of a 3440×1440 source, NVENC and OpenH264, local LiveKit, a second session decoding), sent → received:

  • 1080p: 1920×802 → 1920×800, labelled "720p"
  • 1440p: 2560×1070 → 2560×1068, labelled "1080p"
  • Source: 3440×1440 → 3440×1440, labelled "1440p"

That is exactly the report. In every run every frame was whole, with no black rows or columns inside it, so any bars beside the picture come from the viewer's layout. Screen shares are single-layer, with no simulcast. The server's Low=true is the name of that single layer (rid q), and the viewer gets it at full size. The software 720p30 clamp was not involved: it needs the NVENC probe to fail, and on an RTX 4080 that is logged at WARN by libwebrtc.

PR #879: the viewer's pill names a non-16:9 stream by its size ("1920×800"). The sharer's own pill shows what is sent, including the 720p30 clamp, and a WARN is logged when the clamp applies. The fit rule is unchanged; the original's pixel-budget fit is offered there as an option. The top strip is tracked on #589.

**Hypothesis and result (measured, PR #879).** Hypothesis: nothing is downgraded. Each setting is a 16:9 box, the 3440×1440 picture is fitted into it with its aspect kept, and the viewer's pill names a stream by the closest standard height. Measured with `run.sh --chain --setting` (real Wayland capture of a 3440×1440 source, NVENC and OpenH264, local LiveKit, a second session decoding), sent → received: - 1080p: 1920×802 → 1920×800, labelled "720p" - 1440p: 2560×1070 → 2560×1068, labelled "1080p" - Source: 3440×1440 → 3440×1440, labelled "1440p" That is exactly the report. In every run every frame was whole, with no black rows or columns inside it, so any bars beside the picture come from the viewer's layout. Screen shares are single-layer, with no simulcast. The server's `Low=true` is the name of that single layer (rid `q`), and the viewer gets it at full size. The software 720p30 clamp was not involved: it needs the NVENC probe to fail, and on an RTX 4080 that is logged at WARN by libwebrtc. PR #879: the viewer's pill names a non-16:9 stream by its size ("1920×800"). The sharer's own pill shows what is sent, including the 720p30 clamp, and a WARN is logged when the clamp applies. The fit rule is unchanged; the original's pixel-budget fit is offered there as an option. The top strip is tracked on #589.
Author
Owner

The label changes of PR #879 are in nightly 2026.1118.383. The top strip (#589) for this sharer is not resolved; see #589 for the log that is needed.

The label changes of PR #879 are in nightly 2026.1118.383. The top strip (#589) for this sharer is not resolved; see #589 for the log that is needed.
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
charles/fluxer-native#876
No description provided.