MilkDrop: preset thumbnails in a grid (#575), with engine buffer and light sampler fixes

README.md

@@ -12,7 +12,7 @@
 [![Platform](https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux-111997.svg)](#build--distribution)
 [![Electron](https://img.shields.io/badge/Electron-43-47848F.svg)](https://www.electronjs.org/)
 [![Downloads](https://img.shields.io/github/downloads/CaYatur/SoundVisualizer/total?label=downloads)](https://github.com/CaYatur/SoundVisualizer/releases)
-[![Tests](https://img.shields.io/badge/tests-2081%20passing-2ea043.svg)](#tests)
+[![Tests](https://img.shields.io/badge/tests-2108%20passing-2ea043.svg)](#tests)
 [![cayadev.com](https://img.shields.io/badge/cayadev.com-e11d2a.svg)](https://cayadev.com)
 
 </div>
@@ -360,6 +360,15 @@ that asserts the bar profile has no step in it.
   scanned in full before you are asked. No preset pack ships with the app. Measured with real
   packs: a 9,795-preset folder was scanned in 1.1 s and imported in 6.2 s, and importing it again
   added nothing.
+- **Preset thumbnails.** The MilkDrop panel's list has a *Grid* layout: every preset shows a small
+  picture of itself, so a preset can be found by its look. Each thumbnail is drawn once in the
+  background — the first two seconds of the preset with the demo sound, from a black screen — kept
+  in the app's data, and drawn again when the preset or a texture it uses changes. Only the
+  presets you scroll to are drawn. A visualizer window drawing MilkDrop at the same time barely
+  noticed: none of its frames took longer than 27 ms. Found while building this: when another
+  MilkDrop window was drawing and the computer was busy, a MilkDrop picture that had just set up its
+  buffers (a new window, a resize, a thumbnail, a video export) could start from the other window's
+  picture. Buffers now start from zeros.
 - **Sprites from `milk_img.ini`.** MilkDrop 2 draws images of your own over the picture during a
   show: each one is defined in `milk_img.ini` with an image, code that runs once and code that
   runs every frame, and launched by number. Choose the file in the MilkDrop panel and launch from
@@ -1259,7 +1268,7 @@ npm test
 npm start -- --smoke
 ```
 
-**2081 unit tests, all passing.** They are written to check answers, not to exercise lines:
+**2108 unit tests, all passing.** They are written to check answers, not to exercise lines:
 
 - **Formulas** are checked against values derived by hand from their definitions — Viviani's curve
   staying on its sphere, the torus tube radius, Chladni's m↔n antisymmetry, every attractor

README.tr.md

@@ -12,7 +12,7 @@
 [![Platform](https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux-111997.svg)](#paketleme--dağıtım)
 [![Electron](https://img.shields.io/badge/Electron-43-47848F.svg)](https://www.electronjs.org/)
 [![İndirme](https://img.shields.io/github/downloads/CaYatur/SoundVisualizer/total?label=indirme)](https://github.com/CaYatur/SoundVisualizer/releases)
-[![Test](https://img.shields.io/badge/test-2081%20geçiyor-2ea043.svg)](#testler)
+[![Test](https://img.shields.io/badge/test-2108%20geçiyor-2ea043.svg)](#testler)
 [![cayadev.com](https://img.shields.io/badge/cayadev.com-e11d2a.svg)](https://cayadev.com)
 
 </div>
@@ -353,6 +353,15 @@ profilinde basamak olmadığını doğruluyor.
   baştan taranıyor. Uygulamayla hiçbir preset paketi gelmiyor. Gerçek paketlerle ölçüldü: 9.795
   presetlik bir klasör 1,1 sn'de tarandı, 6,2 sn'de içe aktarıldı; ikinci kez almak hiçbir şey
   eklemedi.
+- **Preset küçük resimleri.** MilkDrop panelindeki listenin *Izgara* düzeni var: her preset kendi
+  küçük resmiyle görünüyor, yani bir preset görünüşünden bulunabiliyor. Her küçük resim arka planda
+  bir kez çiziliyor — presetin ilk iki saniyesi, örnek sesle, siyah bir ekrandan — uygulamanın
+  verisinde saklanıyor ve preset ya da kullandığı bir doku değişince yeniden çiziliyor. Yalnız
+  kaydırıp baktığınız presetler çiziliyor. Aynı anda MilkDrop çizen bir görselleştirici penceresi
+  bunu neredeyse hiç fark etmedi: hiçbir karesi 27 ms'den uzun sürmedi. Bunu yaparken bulundu: başka
+  bir MilkDrop penceresi çizerken ve bilgisayar meşgulken, tamponlarını yeni kuran bir MilkDrop
+  görüntüsü (yeni bir pencere, yeniden boyutlandırma, küçük resim, video dışa aktarımı) öbür
+  pencerenin görüntüsüyle başlayabiliyordu. Tamponlar artık sıfırdan başlıyor.
 - **`milk_img.ini`'den sprite'lar.** MilkDrop 2 gösteri sırasında kendi resimlerinizi görüntünün
   üstüne çiziyor: her biri `milk_img.ini`de bir resim, bir kez çalışan ve her kare çalışan koduyla
   tanımlı ve numarasıyla başlatılıyor. Dosyayı MilkDrop panelinde seçin; panelin listesinden, bir
@@ -1256,7 +1265,7 @@ npm test
 npm start -- --smoke
 ```
 
-**2081 birim testi, hepsi geçiyor.** Satır çalıştırmak için değil, cevap denetlemek için yazıldılar:
+**2108 birim testi, hepsi geçiyor.** Satır çalıştırmak için değil, cevap denetlemek için yazıldılar:
 
 - **Formüller**, tanımlarından elle türetilmiş değerlerle sınanıyor — Viviani eğrisinin küre
   üzerinde kalması, simidin boru yarıçapı, Chladni'nin m↔n antisimetrisi, her çekicinin sınırlı

ROADMAP.md

@@ -129,9 +129,9 @@ npm test
 npm start -- --smoke
 ```
 
-- **2081 unit tests, all passing** on `main`. 703 of those shipped in v3.1.0;
+- **2108 unit tests, all passing** on `main`. 703 of those shipped in v3.1.0;
   105 came with v3.1.1; 163 came with v3.1.2; 157 came with v3.1.3 — 1128 at
-  that tag — 469 more with v3.1.4, most of them from the MilkDrop work, and 484
+  that tag — 469 more with v3.1.4, most of them from the MilkDrop work, and 511
   on `main` since.
   Formulas are checked against values derived
   by hand from their definitions — Viviani's curve staying on its sphere, the
@@ -1737,6 +1737,127 @@ rest after. No version number yet.
     main-process handlers, the engine's texture counter, and the panel
     drawn with a fake DOM, confirmation and rescan included. 57 of 57
     mutations are caught.
+- **Preset thumbnails (#575)** · done. The MilkDrop panel's list gets a
+  *Layout* choice, *List* or *Grid*. In the grid every preset shows a
+  thumbnail, drawn once in a hidden window and kept in the user data
+  (`milkdrop-thumbs/`).
+  - **Recipe, measured** on a seeded 200-preset sample of the corpus: a
+    black start; 60 frames of the demo sound at 1/30 s; drawn at 640x360
+    and scaled to a 256x144 WebP (~3.6 KB). A fixed seed image (a builtin
+    preset for 45 frames, then a hard cut) lowered the black thumbnails only
+    from 20 to 18 — fixing four and breaking two — made each job 45% longer
+    and painted its own colours into the thumbnails of presets that draw
+    nothing like it. 90 and 150 frames gave 20 and 22 black thumbnails (some
+    presets fade out), at a median of 350 and 538 ms against 259 ms. Drawing
+    at 640x360 instead of 320x180 barely moved the time (250 against 259 ms
+    median): the work is the per-frame code, not the pixels.
+  - **The same every time.** Each thumbnail is drawn by a new engine
+    instance, whose context `dispose` releases, from black buffers.
+    `Math.random` and the preset's seed come from the key. A texture the
+    preset asks for is waited for right after the frame that asks, and a
+    thumbnail whose texture does not arrive within 8 s is not written;
+    neither is one whose frames were not all drawn on a live context (with
+    no context the engine draws a placeholder, after a loss it keeps the
+    last frame — either would be kept for good under a content key). A
+    crashed drawing process is replaced instead of making every later cell
+    wait for the 30 s limit. Checked in the real page: seven presets, each
+    drawn after two different busy presets, gave the same bytes both times
+    — the five busiest of 40 (6.8–13 KB), a textured one and one using
+    `sampler_randNN`. Also checked under load (below): one preset drawn 48
+    times with a visualizer window drawing MilkDrop and the CPU busy gave
+    the same bytes every time.
+  - **Key** = a hash of the recipe (the bytes of every file the thumbnail
+    page loads, plus the sizes and the frame count) + the preset's source +
+    a signature of the textures it asks for (name, size, modification time;
+    the whole list for `sampler_randNN`). An engine change (#580, #567)
+    therefore retires every thumbnail by itself, an edited preset or a
+    changed texture gets a new one, and two presets with the same content
+    share one. Once per session, files no current preset produces are
+    deleted.
+  - **Panel.** The visible cells are asked for; cached thumbnails come back
+    at once, the rest as they are drawn, newest request first, one at a
+    time. A waiting cell is striped and a failed one has a dashed frame, so
+    neither looks like a thumbnail that is really black (about 9% are). The
+    layout is the viewer's own choice, kept in browser storage rather than
+    in the settings. Images come through an `sv-thumb:` protocol that
+    accepts only a key and reads only the thumbnail folder. The hidden
+    window closes after 20 s idle and when the panel closes — otherwise it
+    would keep the app from quitting.
+  - **Found on the way — a picture from another window in a new buffer.**
+    The engine allocated its render targets empty and cleared them with
+    `gl.clear`. With another MilkDrop context drawing on the same GPU and
+    the CPU busy, a freshly allocated feedback buffer could still hold that
+    context's picture, and the feedback loop grew it: a preset whose
+    thumbnail is black came out as the starburst the visualizer window was
+    showing at the time, in 13 of 24 draws with the flash limiter off and 3
+    of 16 with it on. Neither condition alone did it (18 of 18 clean with
+    the window drawing bars under load, and with the window drawing MilkDrop
+    and the CPU idle). The same GPU, context attributes and formats were
+    reported every time. Targets are now allocated from zeros — a shared
+    zero buffer that is never written (8 MB at 1080p) — and the clear
+    stays; the same conditions gave 0 of 48. This is not only about
+    thumbnails: any window whose buffers are allocated again (a second
+    window opening, a resize, a restored context) and the video export
+    could start from another window's picture. Drawing alone is unchanged:
+    200 presets gave identical results before and after. The price is paid
+    only when buffers are allocated: a full rebuild of the targets at
+    1920x1080 (feedback, blur and flash limiter) takes 8.6 ms instead of
+    3.3 ms, less than one frame.
+  - **Found on the way, smaller:** `sampler_randNN` picked from the texture
+    list in the order the file system returned — alphabetical on Windows,
+    arbitrary on Linux — so the same preset could get another texture on
+    another machine; the list is now sorted the way the measurement script
+    sorts it. The importer (#574) did not strip the filter/wrap prefix from
+    sampler names, so a preset asking for `sampler_fw_worms` did not bring
+    the `worms.jpg` beside it; now it does. The builtin presets module gave
+    Node nothing; the main process now reads their sources from it. The
+    light colour sampler (#589) read its 64x16 canvas with `getImageData`
+    about 30 times a second and Chromium warned about it — the smoke test
+    caught the warning once lights were set to take MilkDrop's colours; the
+    downscale stays on the GPU and only the 4 KB result is read back from a
+    canvas made for reading.
+  - **Found on the way — the smoke test could drive real lights.** The smoke
+    test and the screenshot tool run on the user's own profile. With
+    Dynamic Lighting on in the settings, the app drove the lights at
+    start-up in those runs too, and the smoke test itself turns OpenRGB on
+    in the panel. As with the camera, which automation never opens, the
+    physical outputs are now off in these runs: every Dynamic Lighting
+    setting goes through one function that sends it switched off, the
+    device scan is skipped, and OpenRGB and Art-Net are never started.
+    Diagnostics (`--diag`) still drives the lights, since that is what it
+    is for.
+  - **Measured** in an isolated copy with 120 corpus presets and the panel
+    driven: the first thumbnail after 1.0 s, the 20 visible cells full in
+    10–17 s. With a visualizer window drawing MilkDrop at 75 Hz, 20 s idle
+    against 20 s while 31 thumbnails were drawn: frame time p99 13.5 against
+    13.6 ms, longest frame 17.2 against 26.7 ms, none over 33 ms either way
+    — so no pause between jobs was added. Asking for 40 cached thumbnails
+    took 2.9 ms (median round trip from the panel) with no texture folder
+    and 4.2 ms with the corpus texture pack chosen, so the texture listing
+    is not cached. The English UI was checked.
+  - **Not done:** thumbnails in the web remote and in the list view;
+    drawing every thumbnail ahead of time (only what is looked at is drawn;
+    10,000 presets would take over an hour); a thumbnail shows the first two
+    seconds, not a preset's later look. Thumbnails always use the engine's
+    defaults (MilkDrop fidelity on, mesh 64, smooth lines, flash limiter
+    on): with other settings the live picture differs a little.
+  - **Tests.** 22 tests: recipe and page files, key and texture signature,
+    the queue (cache, one job per key, newest first, failure, staleness,
+    cap, clear), cleanup, the protocol, the page driver with a fake engine
+    (new instance per job, seeding, texture wait and timeout, no or lost
+    context, a frame not drawn), the builtins
+    in Node, the main-process pins, and the panel's grid with a fake DOM.
+    The smoke test draws a builtin preset's thumbnail through the whole
+    chain — the recipe files read from the package, the hidden window, the
+    protocol and the panel's CSP — into a temporary folder, not the user's
+    data. One engine test drives `_makeTarget` against a fake GL: a zero
+    array of the right type and length for both formats, the unpack
+    alignment set first, the clear kept, the buffer shared and grown. One
+    checks that the light sampler reads only the small canvas made for
+    reading, and three check that every path to the lights passes the
+    automation flag. 57 of 58 mutations are caught; the survivor removes
+    the "no context" check, which the frame counter already makes (without
+    a context no frame is drawn).
 
 ## v3.1.6 — Comprehensive video export
 

src/admin/index.html

@@ -2,7 +2,7 @@
 <html lang="tr">
   <head>
     <meta charset="UTF-8" />
-    <meta http-equiv="Content-Security-Policy" content="default-src 'self'; img-src 'self' data: blob: sv-logo:; connect-src 'self' sv-logo:; media-src 'self' sv-media: blob: data:; style-src 'self' 'unsafe-inline';" />
+    <meta http-equiv="Content-Security-Policy" content="default-src 'self'; img-src 'self' data: blob: sv-logo: sv-thumb:; connect-src 'self' sv-logo:; media-src 'self' sv-media: blob: data:; style-src 'self' 'unsafe-inline';" />
     <title>Ses Görselleştirici — Yönetici Paneli</title>
     <link rel="stylesheet" href="admin.css" />
     <link rel="stylesheet" href="studio.css" />