Skip to content

omamobile-device-husky 0.1.17: screen look stopgap (#537) - #880

Open
Somnius wants to merge 2 commits into
derekross:mainfrom
Somnius:husky-screen-look-1118
Open

Somnius wants to merge 2 commits into
derekross:mainfrom
Somnius:husky-screen-look-1118

Conversation

@Somnius

@Somnius Somnius commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

omamobile-device-husky 0.1.17: screen look stopgap for the 8 Pro (#537)

This ships the screen shader I've been running on my 8 Pro since 11-14, for the picture looking washed out after boot. It's off by default and one file turns it on. It's pow(rgb, 1.55) then times 0.88, as a Hyprland screen shader.

It depends on #204 (#204). That PR adds the optional /usr/share/omamobile/hypr/device.lua hook to phone.lua, loaded with pcall after the shared rules. This PR only adds husky's own device.lua and the shader, no shell changes, so without #204 the file is installed but nothing loads it. Inside device.lua the hl.config call is in a pcall too and prints an error if it fails.

The numbers come from a first pass with the names showing, then two blind gamma rounds and one blind gain round by eye on the panel. Gamma 1.5 to 1.6 won the blind gamma rounds, no change looked washed out, more saturation looked worse, and the gain round picked 0.85 to 0.92, so 0.88. Details and the correction are here:
#537 (comment)
#537 (comment)
#537 (comment)

The cause is still unknown. Stock's tables don't explain a curve like this, and @gitColinX saw his unit look deeper for a while on 11-15 with nothing changed, so panels or the panel state after the bootloader may differ. So this is a stopgap, the real fix belongs in the display path.

To turn it on: touch ~/.config/omamobile/screen-look-on and hyprctl reload. Remove the file and reload to turn it off. Without the file device.lua doesn't call hl.config at all, so a screen_shader line in the user's own config still applies, and nothing changes for anyone who doesn't create it. My own phone has the shader in looknfeel.lua, which is how I've been running it since 11-14.

Screenshots, recordings and the VNC mirror don't show the shader: they go through Hyprland's screencopy, and v0.57.0's ScreenshareFrame.cpp sets blockScreenShader for its frames. So the look can only be judged on the panel or with a camera.

Husky goes to 0.1.17 (#836 is in as 0.1.15, #860 has 0.1.16). The README also says now that an updated phone gets #836's Wi-Fi power save from its next boot. glslang and lua are now checkdepends, and the test fails if they're missing instead of skipping.

Tested: husky's test/all and tools/pr-check pass. The test checks the GLSL ES header, compiles the shader with glslangValidator, runs device.lua with a stub hl with and without the marker (no marker means no hl.config call at all) and with a failing hl.config, and checks no other device package ships screen-look.frag. The shader itself, same file, has been live on my phone since 11-14 and the GPU stayed at its lowest clock with it on.

Not tested: I haven't built and installed the package on the phone, so install from the package and the load through #204's hook aren't tried in a real session. Turning it on and off with the marker and hyprctl reload isn't tried on the phone either. And it's only tried on my unit, which is why it's opt-in.

@derekross

Copy link
Copy Markdown
Owner

Thank you, Lef. The blind rounds are the right way to pick numbers by eye, the pcall around hl.config and the opt-out are careful, and the test does real work: I removed the shader = "" line and the pcall in turn and it caught both. I also checked the two things you took from Hyprland's source against v0.57.0. An empty screen_shader does turn it off (applyScreenShader() returns on an empty path). And your open question has an answer: captures don't include the shader. ScreenshareFrame.cpp sets blockScreenShader for every capture, so screenshots, recordings and the VNC mirror show the picture without the look. That fits what you saw on #537, and it means the look can only be judged on the panel or through a camera. Please put that in the README.

It can't go in yet, for these reasons:

  1. On by default for every 8 Pro. I'd like it off by default for now, with the marker file turning it on. My reasons: both you and @gitColinX have seen the panel go "deeper" by itself, and on a unit in that state the shader would take mid-grey to 0.30 on top of an already darker picture; Colin can't say his usual look is washed out and hasn't tried the shader; and the cost is certain (white at 0.88 of its brightness at any backlight setting, and the lowest 16 grey levels folded into 3). Opt-in keeps it one touch away for anyone who wants it and can be flipped to default-on with one line once a second phone agrees or the cause is known. @gitColinX, would you try it by hand on yours (the .frag from this PR and one screen_shader line) and say plainly whether it looks better, worse or too dark?
  2. The red check is this PR's own line: tests/grep-q-pipes flags head -1 hypr/screen-look.frag | grep -qx … at test/all:272 (the suite runs under pipefail). [[ $(head -1 hypr/screen-look.frag) == '#version 300 es' ]] does the same without the pipe. Because tests/all stopped there, the package suites never ran on CI.
  3. After that, CI would fail on the tools: its container has neither glslang nor lua (only lua5.4, pulled in by something else), and your test fails when they are missing. I'd rather keep it failing than skipping, so I'll add the two packages to the workflow myself. Please don't touch the workflow in this PR.
  4. omamobile-shell 0.2.135: an optional device.lua after the shared phone rules #204 comes first. Until it is merged nothing loads device.lua.
  5. A run from the package. With this package and omamobile-shell 0.2.135: an optional device.lua after the shared phone rules #204's shell installed, and your own looknfeel.lua line taken out:
    • pacman -Q omamobile-device-husky omamobile-shell;
    • hyprctl getoption decoration:screen_shader after login, after creating the marker and hyprctl reload, and after removing it and reloading;
    • anything in the session log with screen look: or Screen shader;
    • one screen off and on, and the lock screen;
    • tools/phone-smoke.

Smaller:

On the numbers: #836 merges first as 0.1.15, and this keeps 0.1.17. It conflicts with #836 in test/all (both add a block at the end: keep both) and version, so please rebase once #836 is in.

A Hyprland screen shader, pow(rgb, 1.55) then x0.88, from blind rounds by
eye on the 8 Pro's panel. The cause of the flat mid-tones is unknown, so
this is a stopgap. It is set from /usr/share/omamobile/hypr/device.lua,
which the phone.lua hook from derekross#204 loads; this depends on derekross#204. It is off
by default: ~/.config/omamobile/screen-look-on turns it on, and without
it device.lua doesn't call hl.config. glslang and lua become checkdepends
for the test.
@Somnius
Somnius force-pushed the husky-screen-look-1118 branch from 8db9aa6 to 12ad95d Compare November 18, 2026 16:40
@Somnius

Somnius commented Nov 18, 2026

Copy link
Copy Markdown
Contributor Author

@derekross thanks for checking the Hyprland side, and for answering the capture question. Done in 12ad95d, rebased on main with #836 in (test/all keeps both blocks, version stays 0.1.17):

  • Off by default now. ~/.config/omamobile/screen-look-on turns it on, and without that file device.lua doesn't call hl.config at all. The test checks that no marker gives no call and the marker gives the shader path.
  • [[ $(head -1 hypr/screen-look.frag) == '#version 300 es' ]] in place of the pipe.
  • The loop now only checks that no other device package ships screen-look.frag, so it isn't a rule for anyone's README or their own device.lua.
  • README: captures don't show the shader (screencopy, blockScreenShader in v0.57.0's ScreenshareFrame.cpp), and the omamobile-device-husky 0.1.15: Wi-Fi power save on by default #836 sentence that an updated phone gets the Wi-Fi power save from its next boot. "Not yet booted from the package" is gone: my 8 Pro has rebooted on 0.1.15 since, and iw reads power save on.
  • I didn't touch the workflow.

I rewrote the description for opt-in too. Still waiting on #204, and then I'll do the package run with your list (versions, getoption after login, with the marker and reload, after removing it, the session log, screen off and on, the lock screen, phone-smoke).

@derekross derekross left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you, Lef. Every point is done, and I checked each in the code: the look is off unless ~/.config/omamobile/screen-look-on is there, and without it device.lua doesn't call hl.config at all; the header check no longer pipes into grep -q; the loop only looks for another package shipping your shader file; the README has the capture note and the Wi-Fi sentence. I ran the suite here and broke device.lua twice (calling hl.config without the marker, and without the pcall), and your new cases caught both.

The check is still red, and it isn't yours: CI has no glslang and no lua, so your test stops at "glslangValidator (glslang) is installed for the shader check", as I said it would. Every other suite in that run passed. I'll add the two packages to the workflow and re-run.

What's left, in order:

  1. I add glslang and lua to CI.
  2. #204 merges.
  3. Your run from the package, with the list from my last comment (versions, getoption after login, with the marker and a reload, after removing it, the session log, screen off and on, the lock screen, tools/phone-smoke).

On the number: #899 and #860 also bump this package, so I'll set the version when I merge, whichever comes first. Nothing for you to do about it.

derekross added a commit that referenced this pull request Nov 18, 2026
omamobile-device-husky's coming test (#880) checks a shader with glslangValidator and stops without it, and the Lua cases of package tests were skipped on CI for want of an interpreter. Both are in Arch's extra repository.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@gitColinX

Copy link
Copy Markdown
Contributor

@derekross, @Somnius, tried by hand on my unit: the .frag from this PR (head 1a747c173) and one screen_shader line at the end of looknfeel.lua. Hyprland 0.57.0, Omarchy 4.0.6, published kernel. It applied live with no config error.

It looks better to me. I only tried 1.55 with 0.88, no other values, and I have no stock phone beside it. I'm leaving it on for now and will say if it ever looks too dark.

@Somnius

Somnius commented Nov 19, 2026

Copy link
Copy Markdown
Contributor Author

@gitColinX thanks for trying it by hand. That's a second unit where it looks better, even if neither of us has a stock phone beside it to compare. Good to hear it applied live on 4.0.6 too. If it ever looks too dark on yours, please say so here.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants