diff --git a/README.md b/README.md index 369edf0..5ca1d97 100644 --- a/README.md +++ b/README.md @@ -236,7 +236,7 @@ Windows 10 1809 or newer. Grab the release, or build it yourself. ### Download -Take `featherwall-v0.1.0-win-x64.zip` from the +Take `featherwall-v0.2.0-win-x64.zip` from the [latest release](https://github.com/RitvikDayal/featherwall/releases/latest). It unzips to a single `featherwall.exe`, so there is no installer and nothing to uninstall. Run it and you are done. @@ -248,7 +248,7 @@ release to 7 MB instead of shipping a copy of .NET with it. Verify the download if you like: ```powershell -Get-FileHash .\featherwall-v0.1.0-win-x64.zip -Algorithm SHA256 +Get-FileHash .\featherwall-v0.2.0-win-x64.zip -Algorithm SHA256 # 300f59a28269fc2fc034ad86ee64d6fd320cd17f8c2e7a31c789bcc28844dfd7 ``` @@ -314,9 +314,10 @@ the accent colour rather than a broken box; pausing dims it rather than emptying **The record is the one thing in FeatherWall that runs on a frame clock**, and it is deliberately hard to leave running: it turns only while something is actually playing *and* the desktop is uncovered *and* the display is on. Any of those going false stops it dead — not throttled, stopped. -Turning off "Spin while playing" creates no timer at all, and you keep the record, the artwork and -the progress ring as a still image. Everything else in the app still repaints only when Windows -pushes it an event: a battery percentage moving, a track changing. +Turning off "Spin while playing" leaves its timer permanently disarmed, and you keep the record, +the artwork and the progress ring as a still image. The battery and now-playing sources repaint +only when Windows pushes an event — a battery percentage moving, a track changing — and the clock +ticks once a second, as it has since v0.1.0. Everything else: per-monitor wallpapers, Fill, Fit and Stretch modes, DPI-aware rendering, a full tray menu, JSON config for anything the UI does not expose, and auto-pause on fullscreen apps, diff --git a/docs/architecture.md b/docs/architecture.md index 1975bcf..c1cbae0 100644 --- a/docs/architecture.md +++ b/docs/architecture.md @@ -38,7 +38,9 @@ That gets a window into the right z-slot. It does **not** get pixels on screen. DPI does still matter, just one layer up. Sizes a *human* authored — the clock's font size and its margins — are in config as plain pixels, so on a second monitor at a different DPI the widget would be the same pixel count and therefore the wrong physical size. `MonitorTracker.DpiScale` resolves that per monitor from `GetDpiForMonitor(MDT_EFFECTIVE_DPI)`, **relative to the primary monitor rather than to 96**, so a single-DPI machine scales by exactly 1.0 and renders identically to v0.1.0. `featherwall --diag` prints each monitor's DPI and its computed widget scale. -One more practical discovery: DWM did not reliably compose a *second* per-widget window target inside the wallpaper layer, so the clock is not a window at all — it is an **overlay visual** in the same composition tree as the wallpaper content. Fewer HWNDs, trivial z-order, click-through by construction. The clock is rendered in a Mond-inspired style — a large light-weight time, a hairline separator, and a small dimmed date — as a GDI+ bitmap drawn onto the overlay surface at most once per second (boundary-aligned timer, re-armed each tick so it never drifts). +One more practical discovery: DWM did not reliably compose a *second* per-widget window target inside the wallpaper layer, so the widgets are not windows at all — each is an **overlay visual** in the same composition tree as the wallpaper content, held by key (`clock`, `info`, `halo`, `nowplaying`) so several can coexist above one content surface. Fewer HWNDs, trivial z-order, click-through by construction. + +The widgets are event-driven with one deliberate exception. The clock re-arms a boundary-aligned timer; the battery and now-playing *sources* repaint only when Windows pushes a power-setting notification or a media-session event. The **now-playing record** is the exception: it runs a 15 fps frame clock to turn, gated on playing **and** desktop-uncovered **and** display-on, and when `disc.rotate` is false its timer is constructed but never armed, so nothing ever fires. The clock is rendered in a Mond-inspired style — a large light-weight time, a hairline separator, and a small dimmed date — as a GDI+ bitmap drawn onto the overlay surface at most once per second (boundary-aligned timer, re-armed each tick so it never drifts). ## Settings panel diff --git a/docs/benchmark.md b/docs/benchmark.md index 4bace10..221fd9a 100644 --- a/docs/benchmark.md +++ b/docs/benchmark.md @@ -45,7 +45,9 @@ The check exists because it caught that. | **Video 4K60 playing** | **1** | **199.4** | **188.6** | **220.6** | **58.5** | **0.318 %** | **7.6 %** | **5.8 % (videodecode)** | playing, full window | | **Video 4K60, auto-paused** | **1** | **219.5** | **204.2** | **231.6** | **68.1** | **0.034 %** | **0.8 %** | **0 % (copy)** | paused, full window | | Settings panel open | | | | | | | | | | -| Still image + info widget | | | | | | | | | | +| **Still image + widgets** | **1** | **93.0** | **127.8** | **31.5** | **2.2** | **0.135 %** | **3.3 %** | **0.1 % (copy)** | playing, full window | +| **Video 4K60 + widgets** | **1** | **208.5** | **209.9** | **220.6** | **63.5** | **0.988 %** | **23.7 %** | **6.6 % (videodecode)** | playing, full window | +| **Video 4K60 paused + widgets** | **1** | **210.0** | **212.6** | **220.6** | **63.5** | **0.052 %** | **1.2 %** | **0 % (3d)** | paused, full window | *Playing rows measured 2026-08-22 on build 26200, 24 cores, Intel Arc, single 2560×1600 display at 150 %. Source `scripts/bench/results/featherwall-playing-verified.json`. Paused row measured @@ -84,29 +86,57 @@ refused rather than measured. An interactive run is *expected* to produce it, by that produced the auto-paused row — but expected is not verified, no run has yet produced it, and this file should not read as though one has. -**The widget rows have not been measured, and the design said they should be.** The battery halo -was built to a stated cost budget — no timer, no poll, one small bitmap per change — and the -now-playing record then added the app's first frame clock on top of it. The way to check both is -to run the still-image row twice, once plain and once with `-Widgets`, and see whether it moves. +## What the widgets cost — measured 2026-08-29 -**"No timer anywhere" is no longer true of FeatherWall as a whole, and this file used to say it -was.** The record turns at 15 fps while music plays and the desktop is visible, and stops -completely otherwise. That cost has never been measured, which is exactly why the row below -matters more now than it did when the claim was safe. Two attempts were lost to a full-screen browser that ignores `MinimizeAll` and +Two runs, back to back in one session, on the same machine and the same media. Comparing across +sessions would be meaningless: this machine's CPU figure moves two- to threefold run to run. + +| Still image | Without widgets | With widgets | Change | +|---|---|---|---| +| CPU, whole machine | 0.026 % | **0.135 %** | **x5.2** | +| CPU, one core | 0.6 % | **3.3 %** | **x5.5** | +| Working set | 110.7 MB | 127.8 MB | +17 MB | + +| 4K60 playing | Without widgets | With widgets | Change | +|---|---|---|---| +| CPU, whole machine | 0.721 % | **0.988 %** | x1.37 | +| CPU, one core | 17.3 % | **23.7 %** | x1.37 | +| Busiest GPU engine | 6.4 % decode | 6.6 % decode | unchanged | + +**The still-image row moved, and a lot in relative terms.** Five times is far outside the two- to +threefold spread this machine shows between runs, so it is a real signal rather than noise. + +That is the record's frame clock, and it is the trade the design accepted knowingly. The still +image is the state where FeatherWall previously did *nothing at all* after first paint, so any +number above zero is an infinite proportional increase and a small absolute one: **0.135 % of a +24-core machine is about a thirtieth of one core.** Set `disc.rotate` to false and no timer is +created — the record, its artwork and the progress ring stay as a still image. + +On a video wallpaper the same widgets cost proportionally far less, because the decoder already +dominates, and the GPU figure does not move at all. + +**Sources:** `scripts/bench/results/featherwall-baseline-verified.json` and +`featherwall-widgets-verified.json`. The record was confirmed turning during the sampling window +from FeatherWall's own log, so this measures the timer running rather than a static widget. + +**"No timer anywhere" was never quite true, and this file used to say it was.** The clock widget +has re-armed a 1 Hz timer since v0.1.0. What that claim actually meant, and still means, is that the +*battery and now-playing sources* repaint only on pushed events. The record is the new exception and +the only continuous one: 15 fps while music plays and the desktop is visible, nothing otherwise. Two attempts were lost to a full-screen browser that ignores `MinimizeAll` and takes the foreground back between samples, so FeatherWall correctly paused and the harness correctly refused every playing row. The harness now names that window and stops before running rather than discovering it four rows later. -To fill the row, from an interactive session with nothing in full screen: +To reproduce, from a session with nothing in full screen: ```powershell .\scripts\bench\Run-Bench.ps1 -Video -Image .\docs\media\hero.jpg -Seconds 30 .\scripts\bench\Run-Bench.ps1 -Video -Image .\docs\media\hero.jpg -Seconds 30 -Widgets ``` -Compare the two still-image rows. **CPU spans roughly 2x run to run on this machine**, as the note -above records, so a difference inside that spread is not evidence of anything and should be -reported as inconclusive rather than as a pass. +Compare the two still-image rows, and only ever from a single session. **CPU spans roughly two- to +threefold run to run on this machine**, so a difference inside that spread is not evidence of +anything. What *is* verified is which parts have a timer and which do not. Every repaint is logged. Over an hour of running, the battery and info widgets repainted only at startup, when a media session diff --git a/docs/index.html b/docs/index.html index bf22639..f2467a4 100644 --- a/docs/index.html +++ b/docs/index.html @@ -4,9 +4,9 @@ FeatherWall — live wallpapers for Windows - + - + @@ -479,7 +479,7 @@
FeatherWall - v0.1.0  /  MIT  /  Windows 10 & 11 + v0.2.0  /  MIT  /  Windows 10 & 11
Source @@ -535,8 +535,8 @@

FEATHER
WALL

Live wallpapers for Windows that don't eat your laptop.

- - Download for Windows ZIP 7.5 MB + + Download for Windows ZIP 7.2 MB Read the source
@@ -545,7 +545,7 @@

Live wallpapers for Windows that don't eat your laptop.

- Open source live wallpaper engine  /  v0.1.0  /  7.5 MB  /  MIT  /  100% C# + Open source live wallpaper engine  /  v0.2.0  /  7.2 MB  /  MIT  /  100% C# Scroll to lay it flat ↓
@@ -560,8 +560,8 @@

Live wallpapers for Windows that don't eat your laptop.

0Process, not five
-
0MB for a still wallpaper
-
0Percent CPU at rest
+
0MB for a still wallpaper
+
0Percent CPU at rest
0Accounts or trackers
@@ -604,12 +604,12 @@

It gets out of the way

02Measured

What it actually costs

-

25 one-second samples on a running instance, with the pause state checked before and after each window so a silently paused wallpaper could never be recorded as a cheap one. Pick a state and watch what moves.

+

Produced by scripts/bench/Run-Bench.ps1, which reads FeatherWall's own log and refuses to report a row whose pause state changed inside the sampling window — because measuring a silently paused wallpaper is the easiest way to publish a flattering lie. Pick a state and watch what moves.

- - + +
@@ -620,38 +620,45 @@

What it actually costs

Name
CPU
Memory
GPU
GPU engine
featherwall.exe
-
0.79%
-
212 MB
-
29.4%
-
GPU 0 – Video Processing
+
0.32%
+
199 MB
+
5.8%
+
GPU 0 – Video Decode
GPU engines · this process
-
Video processing29.4%
-
Video decode6.8%
-
3D7.5%
-
Graphics memory122 MB
+
Video processing0%
+
Video decode5.8%
+
3D0%
+
Graphics memory221 MB
Adapters
-
GPU 0 · Intel Arc, integrated29.4%
+
GPU 0 · Intel Arc, integrated5.8%
GPU 1 · NVIDIA RTX 5090 Laptop0%

All of this lands on the integrated GPU. nvidia-smi lists no FeatherWall process on the RTX 5090, which stayed at 0% throughout. The compositor lives on whichever adapter drives your display, so the wallpaper never competes with a game on the discrete card.

-

Two things the numbers do not say on their own. GPU percentages are per engine and do not sum, which is why the table shows the busiest one, the same as Task Manager. And graphics memory is a separate allocation from the working set: dedicated VRAM reads 0 MB because integrated graphics borrow system RAM, so a 4K60 wallpaper really costs about 480 MB of system memory rather than 248 MB. Paused is the largest memory row because buffers are retained so resuming is instant, while CPU and both video engines drop to nothing.

+

Two things the numbers do not say on their own. GPU percentages are per engine and do not sum, which is why the table shows the busiest one, the same as Task Manager. And graphics memory is a separate allocation from the working set, so the honest total for a wallpaper is the working set plus both graphics figures. Paused is the largest memory row because buffers are retained so resuming is instant, while CPU and the video engines drop to nothing.

+ +

Two corrections this page owes you. It used to say dedicated VRAM reads 0 MB “because integrated graphics borrow system RAM”. That is wrong: this machine's integrated Arc reports 2048 MB of adapter RAM and Windows attributes 232 MB of dedicated usage to FeatherWall on it. The old explanation made the footprint look smaller than it is. It also published 0.004% CPU for the paused state; the harness measures 0.034%. Both numbers here now come from the script rather than from reading Task Manager, and the correction goes against us on the paused row.

+ +

1080p60 has never been measured, so its tab shows dashes rather than a number. An empty cell means nobody has run it — not zero, and not unknown-but-probably-fine. CPU also moves two- to threefold between runs of the same clip on the same machine, so treat any single figure as one sample.

+ +

What the widgets cost. Measured across two runs in one session. On a still image the widgets take CPU from 0.026% to 0.135% of the machine — about a thirtieth of one core, but roughly five times the baseline, because a still wallpaper previously did nothing at all after first paint. That is the record's frame clock, and switching off “spin while playing” gives it back. On a 4K60 video the same widgets move CPU from 0.721% to 0.988% and the GPU not at all, because the decoder already dominates.

- 03The widget -

Any font you own

+ 03The widgets +

A clock, a charge, a record

-

Nine positions, any installed typeface, adjustable size, colour and opacity. The settings panel previews it with the same renderer that paints your desktop, so what you pick is what lands.

+

The clock takes any typeface you own, at any of nine positions, with adjustable size, colour and opacity — previewed in the settings panel by the same renderer that paints your desktop. Alongside it: a battery halo whose ring is the charge and whose colour steps with it, and a record that shows what you are listening to, with the real album artwork on its label, turning while the music plays.

+

The record is the only part that runs on a frame clock, and it stops the moment anything covers the desktop, the display sleeps, or the music does. Turning off “spin while playing” leaves its timer permanently disarmed and the artwork and progress ring as a still image. Everything else repaints only when Windows pushes an event — a battery percentage moving, a track changing.

Clock set in Bahnschrift Light over the Hubble Ultra Deep Field @@ -680,14 +687,14 @@

Get it, check it, or build it

Download Fastest

Unzip anywhere and run FeatherWall.exe. It lives in the tray.

- Get the ZIP 7.5 MB + Get the ZIP 7.2 MB

SHA-256
300f59a28269fc2fc034ad86ee64d6fd320cd17f8c2e7a31c789bcc28844dfd7

Verify Then

Check what you downloaded before you run it. PowerShell, no extra tools.

-
Get-FileHash .\featherwall-v0.1.0-win-x64.zip ` +
Get-FileHash .\featherwall-v0.2.0-win-x64.zip ` -Algorithm SHA256

Should match the hash on the left, character for character.

@@ -745,7 +752,7 @@

Who made this

06 / Get it

Put something alive behind your work.

- Download for Windows ZIP 7.5 MB + Download for Windows ZIP 7.2 MB Star it on GitHub
@@ -846,10 +853,15 @@

Put something alive behind your work.

/* the Task Manager view. Every figure here was measured, not modelled: 25 one-second samples per state with the pause state verified either side. */ var STATES = { - still: {cpu:'0.06%', mem:'156 MB', gpu:'0.1%', eng:'GPU 0 – 3D', proc:0, dec:0, d3:0.1, gmem:44, gmemLabel:'44 MB'}, - p1080: {cpu:'0.79%', mem:'212 MB', gpu:'29.4%', eng:'GPU 0 – Video Processing', proc:29.4, dec:6.8, d3:7.5, gmem:122, gmemLabel:'122 MB'}, - p4k: {cpu:'0.84%', mem:'248 MB', gpu:'41.8%', eng:'GPU 0 – Video Processing', proc:41.8, dec:17.9, d3:7.6, gmem:234, gmemLabel:'234 MB'}, - paused: {cpu:'0.004%', mem:'309 MB', gpu:'0.1%', eng:'GPU 0 – 3D', proc:0, dec:0, d3:0.1, gmem:269, gmemLabel:'269 MB'} + /* Every figure here comes from scripts/bench/Run-Bench.ps1, which refuses to report a row + whose pause state changed inside the sampling window. The earlier set was read off Task + Manager by hand and the harness contradicted it — paused CPU in particular was published as + 0.004% and actually measures 0.034%. 1080p60 has never been run and says so rather than + carrying a number that looks measured. */ + still: {cpu:'0.008%', mem:'110 MB', gpu:'0%', eng:'GPU 0 – idle after first paint', proc:0, dec:0, d3:0, gmem:49, gmemLabel:'49 MB'}, + p1080: {cpu:'—', mem:'—', gpu:'—', eng:'not measured', proc:0, dec:0, d3:0, gmem:0, gmemLabel:'—'}, + p4k: {cpu:'0.32%', mem:'199 MB', gpu:'5.8%', eng:'GPU 0 – Video Decode', proc:0, dec:5.8, d3:0, gmem:221, gmemLabel:'221 MB'}, + paused: {cpu:'0.034%', mem:'220 MB', gpu:'0%', eng:'GPU 0 – idle', proc:0, dec:0, d3:0, gmem:232, gmemLabel:'232 MB'} }; var GMEM_MAX = 300; /* bar scale, so graphics memory is comparable across states */ @@ -868,7 +880,9 @@

Put something alive behind your work.

document.getElementById('b-dec').style.width = d.dec + '%'; document.getElementById('b-3d').style.width = d.d3 + '%'; document.getElementById('b-mem').style.width = Math.min(100, d.gmem / GMEM_MAX * 100) + '%'; - document.getElementById('b-igpu').style.width = parseFloat(d.gpu) + '%'; + /* The unmeasured row shows an em dash, and parseFloat of that is NaN — which would set an + invalid width rather than an empty bar. */ + document.getElementById('b-igpu').style.width = (parseFloat(d.gpu) || 0) + '%'; } document.querySelectorAll('.tm-tab').forEach(function(tab){ tab.addEventListener('click', function(){ @@ -877,7 +891,9 @@

Put something alive behind your work.

showState(tab.dataset.state); }); }); - showState('p1080'); + /* 4K60 is the heaviest state that has actually been measured. Defaulting to 1080p60 now shows + dashes, which reads as broken rather than as honest. */ + showState('p4k'); /* copy buttons on the install commands */ document.querySelectorAll('.copy').forEach(function(btn){ diff --git a/docs/release-notes-v0.2.0.md b/docs/release-notes-v0.2.0.md new file mode 100644 index 0000000..c81c77d --- /dev/null +++ b/docs/release-notes-v0.2.0.md @@ -0,0 +1,117 @@ +# FeatherWall v0.2.0 + +Widgets that read the system, and performance numbers you can reproduce with a command instead of +taking on trust. + +## Widgets + +**Info widget.** A second, independently anchored block fed by the operating system. **Off by +default** — one checkbox on the Info page turns it on. A v0.1.0 config loads unchanged and the +desktop renders exactly as it did until you switch it on. + +**Battery halo.** A ring with the charge inside it: the arc is the level, the colour steps with it +— red low, orange middling, green when healthy — and a bolt sits beside the number while charging, +replaced by a tick when full. Three palette presets, five colour pickers, adjustable thresholds and +size. It sits beside the text, or detaches entirely and takes its own anchor anywhere on screen. + +**Now-playing record.** What you are listening to, drawn as a vinyl record with the **real album +artwork** on its label — read from the same Windows media session everything else here uses, so +there is still no network call. It turns at 33⅓ rpm while something plays, with the track's +progress on the rim. A track with no artwork gets a flat disc in the accent colour rather than a +broken box; pausing dims it rather than emptying it. It reads any app with media controls, a +browser tab included. + +**Clock.** Independent date styling — its own font, size as a percentage of the time, colour and +opacity, or inherit each from the time. Per-edge margins for both. + +## The record animates, which is new + +The record is the first **continuous** frame clock in FeatherWall. To be precise about what changed: +the clock widget has re-armed a 1 Hz timer since v0.1.0 to tick over the minute, so the app has +never been entirely timer-free. What is new is something redrawing many times a second. + +It is deliberately hard to leave running. It runs only while something is actually playing **and** +the desktop is uncovered **and** the display is on. Any of those going false stops it dead — not +throttled, stopped. Turning off *Spin while playing* leaves the timer permanently disarmed, so +nothing fires and the record, its artwork and the progress ring stay as a still image. + +The battery and now-playing *sources* repaint on events only: a battery percentage moving, a track +changing. + +## Rendering and reliability + +- **Per-monitor DPI.** Widgets are sized against the primary display's scaling, so they keep the + same physical size on a 100 % external monitor as on a 150 % laptop panel. +- **Device-loss recovery.** A driver update or TDR rebuilds the whole layer instead of leaving a + dead surface, with a bounded guard against a rebuild loop. +- **Pushed power and display signals** replace polling *for display state*. The wallpaper stops + rendering into a screen that is off because Windows says so, not because a timer noticed. Pause + decisions that depend on the foreground window — fullscreen apps, coverage — are still polled + twice a second, because Windows offers no push for those. +- **CodeQL** runs on the C# code on every push. + +## Measurement, and two corrections + +`scripts/bench/Run-Bench.ps1` drives FeatherWall through each wallpaper state and measures it. It +**refuses to report a row** whose pause state changed inside the sampling window — measuring a +silently paused wallpaper is the easiest way to publish a flattering number. That check is not +theoretical: the harness's own first run reported a "video playing" figure while the log showed the +wallpaper pausing and resuming three times inside the window. + +Building it corrected two published claims, and both corrections go against us: + +- The README and the website said dedicated VRAM reads 0 MB "because integrated graphics have no + memory of their own". **Wrong.** This machine's integrated Arc reports 2048 MB of adapter RAM and + Windows attributes **232 MB of dedicated usage** to FeatherWall on it. The old explanation made + the footprint look smaller than it is. +- The paused-state CPU figure was published as **0.004 %**. It measures **0.034 %**. + +The website carried both of those until this release and now carries the corrections. + +### What the widgets cost, measured + +Two runs back to back in one session, because this machine's CPU figure moves two- to threefold +between runs and comparing across sessions would mean nothing. + +| Still image | Without widgets | With widgets | +|---|---|---| +| CPU, whole machine | 0.026 % | **0.135 %** | +| CPU, one core | 0.6 % | **3.3 %** | +| Working set | 110.7 MB | 127.8 MB | + +**It moved about fivefold, and that is the record's frame clock.** Five times is well outside the +run-to-run spread, so it is a real signal. In absolute terms it is small — 0.135 % of a 24-core +machine is roughly a thirtieth of one core — but the still image is the state where FeatherWall +used to do *nothing at all* after first paint, so it is worth saying out loud rather than burying +in a ratio. `disc.rotate: false` leaves the timer disarmed and gives the number back. + +On a 4K60 video wallpaper the same widgets cost proportionally far less (0.721 % → 0.988 %) because +the decoder already dominates, and the GPU figure does not move. + +The record was confirmed turning during the sampling window from FeatherWall's own log, so this +measures the timer running rather than a static widget. + +### What is still not measured + +The 1080p60 row, the settings-panel row, and every competitor comparison row. The website's 1080p60 +tab shows dashes for the same reason. An empty cell means nobody has run it — not zero. + +## Upgrading + +A v0.1.0 config loads unchanged, and the desktop renders exactly as before until you switch the +info widget on. + +When you do, the halo and the record are already enabled inside it, so they appear with it rather +than needing two more switches. Both have their own toggles — *Show halo* on the Battery page and +*Show record* on the Music page — and removing `battery` or `nowPlaying` from `info.sources` drops +the corresponding widget too. + +## Known limitations + +- Multi-monitor widget placement is **unverified** — issues #3 and #4 need hardware this project + has not been able to test on. Help wanted. +- No row of `docs/recovery-matrix.md` has been executed. +- On the Battery settings page, the anchor grid stays visually undimmed when it does not apply. It + is inert, but it does not look inert. +- Releases are **unsigned**, so SmartScreen will warn. Verify the SHA-256 against the published + `.sha256`, or build from source. diff --git a/packaging/winget/manifests/RitvikDayal.FeatherWall.installer.yaml b/packaging/winget/manifests/RitvikDayal.FeatherWall.installer.yaml index 2e0bc6f..3f51a91 100644 --- a/packaging/winget/manifests/RitvikDayal.FeatherWall.installer.yaml +++ b/packaging/winget/manifests/RitvikDayal.FeatherWall.installer.yaml @@ -1,6 +1,6 @@ # yaml-language-server: $schema=https://aka.ms/winget-manifest.installer.1.6.0.schema.json PackageIdentifier: RitvikDayal.FeatherWall -PackageVersion: 0.1.0 +PackageVersion: 0.2.0 MinimumOSVersion: 10.0.17763.0 InstallerType: zip NestedInstallerType: portable @@ -8,13 +8,21 @@ NestedInstallerFiles: - RelativeFilePath: featherwall.exe PortableCommandAlias: featherwall UpgradeBehavior: install -ReleaseDate: 2026-08-15 +ReleaseDate: 2026-08-29 Dependencies: PackageDependencies: - PackageIdentifier: Microsoft.DotNet.DesktopRuntime.10 Installers: - Architecture: x64 - InstallerUrl: https://github.com/RitvikDayal/featherwall/releases/download/v0.1.0/featherwall-v0.1.0-win-x64.zip - InstallerSha256: 300F59A28269FC2FC034AD86EE64D6FD320CD17F8C2E7A31C789BCC28844DFD7 + InstallerUrl: https://github.com/RitvikDayal/featherwall/releases/download/v0.2.0/featherwall-v0.2.0-win-x64.zip + # PLACEHOLDER — must be replaced before this manifest is submitted to winget-pkgs. + # + # The release workflow builds featherwall-v0.2.0-win-x64.zip on tag push and writes its hash to + # featherwall-v0.2.0-win-x64.zip.sha256 beside it. Take the value from that published sidecar + # rather than from a local build: a local zip can differ byte for byte from the one GitHub + # serves, and the whole point of the hash is that it matches what a user actually downloads. + # + # Left as zeros deliberately. A stale v0.1.0 hash would look plausible and be wrong. + InstallerSha256: 0000000000000000000000000000000000000000000000000000000000000000 ManifestType: installer ManifestVersion: 1.6.0 diff --git a/packaging/winget/manifests/RitvikDayal.FeatherWall.locale.en-US.yaml b/packaging/winget/manifests/RitvikDayal.FeatherWall.locale.en-US.yaml index 32a098b..7d2e895 100644 --- a/packaging/winget/manifests/RitvikDayal.FeatherWall.locale.en-US.yaml +++ b/packaging/winget/manifests/RitvikDayal.FeatherWall.locale.en-US.yaml @@ -1,6 +1,6 @@ # yaml-language-server: $schema=https://aka.ms/winget-manifest.defaultLocale.1.6.0.schema.json PackageIdentifier: RitvikDayal.FeatherWall -PackageVersion: 0.1.0 +PackageVersion: 0.2.0 PackageLocale: en-US Publisher: Ritvik Dayal PublisherUrl: https://github.com/RitvikDayal @@ -31,7 +31,7 @@ Tags: - customization - windows - dotnet -ReleaseNotesUrl: https://github.com/RitvikDayal/featherwall/releases/tag/v0.1.0 +ReleaseNotesUrl: https://github.com/RitvikDayal/featherwall/releases/tag/v0.2.0 Documentations: - DocumentLabel: Architecture DocumentUrl: https://github.com/RitvikDayal/featherwall/blob/main/docs/architecture.md diff --git a/packaging/winget/manifests/RitvikDayal.FeatherWall.yaml b/packaging/winget/manifests/RitvikDayal.FeatherWall.yaml index 289c99c..5806aed 100644 --- a/packaging/winget/manifests/RitvikDayal.FeatherWall.yaml +++ b/packaging/winget/manifests/RitvikDayal.FeatherWall.yaml @@ -1,6 +1,6 @@ # yaml-language-server: $schema=https://aka.ms/winget-manifest.version.1.6.0.schema.json PackageIdentifier: RitvikDayal.FeatherWall -PackageVersion: 0.1.0 +PackageVersion: 0.2.0 DefaultLocale: en-US ManifestType: version ManifestVersion: 1.6.0 diff --git a/scripts/bench/Run-Bench.ps1 b/scripts/bench/Run-Bench.ps1 index 385d905..4748673 100644 --- a/scripts/bench/Run-Bench.ps1 +++ b/scripts/bench/Run-Bench.ps1 @@ -77,9 +77,18 @@ function Set-Wallpaper { color = '#F0FFFFFF'; marginX = 48; marginY = 48; twentyFourHour = $true; monitor = '*' } } if ($Widgets) { - $config.info = @{ enabled = $true; monitor = '*'; anchor = 'BottomLeft'; marginX = 48; marginY = 48 - fontSize = 34; fontFamily = 'Segoe UI'; color = '#F0FFFFFF'; shadow = $true - maxCharacters = 48; sources = @('nowPlaying', 'battery') } + # Written out in full rather than leaning on the C# defaults, so a run is reproducible even + # if those defaults move. This is the heaviest honest configuration: halo on, record on, + # and the record spinning if anything is playing. + $config.info = @{ + enabled = $true; monitor = '*'; anchor = 'BottomLeft'; marginX = 48; marginY = 48 + fontSize = 34; fontFamily = 'Segoe UI'; color = '#F0FFFFFF'; shadow = $true + maxCharacters = 48; sources = @('nowPlaying', 'battery') + halo = @{ enabled = $true; size = 44; detached = $false; placement = 'Left' + colorByLevel = $true; lowThreshold = 20; midThreshold = 50 } + disc = @{ enabled = $true; size = 112; rotate = $true; showProgress = $true + anchor = 'BottomRight'; marginX = 48; marginY = 48 } + } } $config | ConvertTo-Json -Depth 6 | Set-Content -Path $configPath -Encoding UTF8 } @@ -91,6 +100,7 @@ Add-Type -Namespace Bench -Name Fg -MemberDefinition @' [DllImport("user32.dll")] public static extern bool IsIconic(IntPtr h); [DllImport("user32.dll")] public static extern bool GetWindowRect(IntPtr h, out RECT r); [DllImport("user32.dll", CharSet=CharSet.Unicode)] public static extern int GetWindowTextW(IntPtr h, System.Text.StringBuilder s, int n); +[DllImport("user32.dll", CharSet=CharSet.Unicode)] public static extern int GetClassNameW(IntPtr h, System.Text.StringBuilder s, int n); [DllImport("user32.dll")] public static extern uint GetWindowThreadProcessId(IntPtr h, out uint pid); [StructLayout(LayoutKind.Sequential)] public struct RECT { public int L, T, R, B; } '@ @@ -109,12 +119,25 @@ function Find-CoveringWindow { #> param([int] $Samples = 6, [int] $IntervalMs = 700) + # The shell's own windows and FeatherWall's are not "covering" anything. Progman in particular + # IS the desktop: with everything minimised it becomes the foreground window and it is + # full-screen by definition, so without this the check fires on a perfectly clear desktop and + # refuses every run. Same list PauseDecision.cs uses, for the same reason. + $shellClasses = @( + 'Progman', 'WorkerW', 'Shell_TrayWnd', 'SHELLDLL_DefView', 'SysListView32', + 'FeatherWallSurface', 'FeatherWallClock', 'FeatherWallMessage' + ) + $screen = [System.Windows.Forms.Screen]::PrimaryScreen.Bounds for ($i = 0; $i -lt $Samples; $i++) { if ($i -gt 0) { Start-Sleep -Milliseconds $IntervalMs } $h = [Bench.Fg]::GetForegroundWindow() if ($h -eq [IntPtr]::Zero -or [Bench.Fg]::IsIconic($h)) { continue } + $cls = New-Object System.Text.StringBuilder 256 + [void][Bench.Fg]::GetClassNameW($h, $cls, 256) + if ($shellClasses -contains $cls.ToString()) { continue } + $r = New-Object Bench.Fg+RECT if (-not [Bench.Fg]::GetWindowRect($h, [ref]$r)) { continue } diff --git a/src/FeatherWall/FeatherWall.csproj b/src/FeatherWall/FeatherWall.csproj index c56396e..e6fe113 100644 --- a/src/FeatherWall/FeatherWall.csproj +++ b/src/FeatherWall/FeatherWall.csproj @@ -13,7 +13,7 @@ app.manifest true en - 0.1.0 + 0.2.0 A featherweight live-wallpaper engine for Windows.