{"items":[{"cid":"bafyreifaj457dwf5nec2t5exqtmqw5mjdeowbl746kgazyhja2ejaqtsb4","uri":"at://did:plc:yfvwmnlztr4dwkb7hwz55r2g/sh.tangled.string/3mjb72cz3kt22","value":{"$type":"sh.tangled.string","contents":"> Note that this is ALL CLAUDE writing. If you don't want to read one LLM critiquing another, do not read.\n\n# ass-renderer / ass-core Audit\n\nAudit conducted 2026-04-09 during Phase 7 of the subtitle system implementation.\n\n## Actively Broken (HIGH)\n\n### GPU backends are non-functional stubs\n- **WebGPU** (`backends/web/pipeline.rs`): `WebGpuPipeline` returns empty results from all methods. `create_pipeline()` in `web/mod.rs` falls back to `SoftwarePipeline`. The composite_layers method uses `pollster::block_on` which won't work in WASM.\n- **Metal** (`backends/hardware/metal.rs`): `render_layers()` has stub comments (\"In production, would create vertex buffer\"). Falls back to `SoftwarePipeline`.\n- **Vulkan** (`backends/hardware/vulkan.rs`): Shader loading is stubbed. No actual Vulkan draw calls. Falls back to `SoftwarePipeline`.\n\nAll three silently fall back to software rendering while reporting `BackendType::WebGPU/Metal/Vulkan`.\n\n**Plan:** WebGPU blit path (Phase 7, Task 4) will replace the WebGPU stubs with a software-render-then-GPU-blit approach. Metal and Vulkan will follow the same pattern when needed (see roadmap). The existing stub code that tries to do GPU-side text rendering should be removed.\n\n### Streaming parser is fake\n- `parser/streaming.rs`: `StreamingResult::sections` is `Vec`, not parsed AST nodes. Comments say \"simplified.\"\n- `parser/streaming/processor.rs`: `process_section_content()` returns empty `DeltaBatch::new()`. Comments say \"In full parser this would...\" and \"In production...\"\n\n**Plan:** Not used by vodplace. Can be removed or properly implemented if incremental parsing becomes needed.\n\n## Fixed in This Session\n\n### Duplicate tag parsers (FIXED)\n- `text_segmenter.rs` had its own ~220-line match block duplicating `tag_processor.rs`, with a `_ => {}` catch-all silently dropping `\\blur`, `\\be`, `\\bord`, `\\shad`, `\\clip`, `\\iclip`, `\\q`, `\\r`, `\\org`, `\\fax`, `\\fay`, and more.\n- **Fix:** Replaced with single `tag_processor::apply_tag()` call. 41 new tests added.\n\n### Duplicate colour parsers (FIXED)\n- `SoftwarePipeline::parse_ass_color` and `ass_core::utils::parse_bgr_color` both parsed ASS colours with different alpha handling.\n- **Fix:** `parse_ass_color` now delegates to `parse_bgr_color`.\n\n### Alpha inversion bug (FIXED)\n- `ass_core::utils::parse_bgr_color` treated ASS alpha 00 as transparent (RGBA 0) instead of opaque (RGBA 255). 6-char colours defaulted to alpha 0 (invisible).\n- **Fix:** Corrected alpha inversion. All tests updated.\n\n### Triple fontdb (FIXED)\n- Three separate `fontdb::Database` instances in RenderContext, SoftwarePipeline, SoftwareBackend. `set_font_database()` cloned into all three. `composite_layers()` created a throwaway backend per frame.\n- **Fix:** Single `Arc` shared everywhere. `Box` instead of Arc. Persistent backend with pixmap clear.\n\n### Font fallback hard failure (FIXED)\n- Renderer errored out with \"Font not found and no fallback available\" when exact font name wasn't in fontdb, even with fonts loaded.\n- **Fix:** Register generic families, absolute last-resort returns any loaded face.\n\n### Blur rendering order (FIXED)\n- `\\blur` only blurred the text fill, not shadow/outline. Entire event should be blurred.\n- **Fix:** When blur present, all rendering (shadow + outline + fill + underline + strikethrough) goes to temp pixmap, gets blurred, then composited.\n\n## Fixed 2026-04-10\n\n### Rotation used radians instead of degrees (FIXED)\n- `backends/software.rs`: `pre_rotate(angle_rad)` — tiny_skia's `pre_rotate` takes degrees, not radians. All `\\frz` rotations were ~57x too small (e.g., 13.73° rendered as 0.24°).\n- **Fix:** Pass degrees directly to `pre_rotate`.\n\n### Drawing paths not scaled (FIXED)\n- `\\p1` drawing commands were only translated to position, never scaled from PlayRes to canvas coordinates. No fscx/fscy or draw_level scaling applied.\n- **Fix:** Apply `draw_scale * scale_x/y * font_scale_x/y` transform to path geometry.\n\n### Drawing clip regions not applied (FIXED)\n- `\\clip` tags on `\\p1` events were parsed but never passed through `VectorData` to the backend. All drawings rendered unclipped — combined heart+bone paths showed both shapes instead of only the clipped portion.\n- **Fix:** Added `clip` field to `VectorData`, create `tiny_skia::Mask` in backend.\n\n### Letter spacing not scaled or included in width (FIXED)\n- `\\fsp` values were applied as raw pixels without PlayRes→canvas scaling, AND not included in `total_line_width` / `seg_advance` calculations. Text with `\\fsp` was too wide (unscaled spacing) and misaligned (width not accounting for spacing).\n- **Fix:** Scale spacing by `scale_x`, include in width calculations.\n\n### Per-channel alpha not animatable (FIXED)\n- `\\1a`, `\\2a`, `\\3a`, `\\4a` were not recognised as `\\t()` animation targets. `\\t(455,500,\\1a&H00&)` silently dropped the `\\1a` target, leaving primary alpha permanently at its initial value.\n- **Fix:** Added `PrimaryAlpha`, `SecondaryAlpha`, `OutlineAlpha`, `ShadowAlpha` variants to `AnimatableTag`.\n\n### FontDb wrapper (FIXED)\n- Raw `fontdb::Database` didn't support fonts fontdb rejects (e.g., Symbol encoding fonts like SSF4 ABUKET). No fallback mechanism.\n- **Fix:** New `FontDb` wrapper with `RwLock` + `RwLock>>>` fallback font map. Shared via `Arc`. Fallback shaping path bypasses fontdb for rejected fonts.\n\n### MKV subtitle extraction missing timestamps (FIXED)\n- `ass-tool extract-mkv` prepended `Dialogue:` to raw ffmpeg packet data, which doesn't include Start/End timestamps (those are in PTS/duration metadata). All extracted ASS files had broken timing — no subtitles displayed.\n- **Fix:** Reconstruct `H:MM:SS.CC` timestamps from `packet.pts()` and `packet.duration()` using stream time_base.\n\n### Font weight suffix lookup (FIXED)\n- \"Dosis Light\" was not recognised as \"Dosis\" + Light weight. External font resolver redundantly tried to fetch fonts already loaded under their base family name.\n- **Fix:** Shared `strip_weight_suffix()` in `ass_renderer::utils::font`. Worker's `is_family_loaded()` and renderer's `try_split_family_weight()` both use it.\n\n## Performance (HIGH)\n\n### Software rendering is too slow for complex fansub typesetting\n- **Measured:** 651ms per frame on an \"I ask of thee\" scene with 12 complex `\\p1` glyph-outline drawings + blur + a full-screen black rectangle + film grain text overlay. Less than 2fps.\n- **Root cause:** Gaussian blur on large complex paths in software is O(width × height × kernel). Multiple blurred drawings per frame compounds the cost.\n- **Comparison:** WASM libass should handle this at interactive framerates — the performance gap suggests our rendering pipeline is doing significantly more work per frame than necessary, or doing it less efficiently.\n- **Mitigation strategies (in order of expected impact):**\n 1. **Pre-render / lookahead cache:** We have the full ASS file upfront plus a playback buffer of seconds to minutes. Pre-render upcoming frames on a background thread / during idle time. Cache rasterised drawing paths keyed by (path_hash, scale, blur_params).\n 2. **Drawing path cache:** Same `\\p1` path at same scale/blur appears across many frames. Cache the blurred rasterised result instead of re-rasterising every frame. Invalidate only when position/alpha/clip changes (those are cheap post-raster operations).\n 3. **Blur optimisation:** Current gaussian blur is naive scalar. Consider: box blur approximation (3-pass box blur ≈ gaussian), downscale-blur-upscale, or SIMD (the `apply_gaussian_blur_simd` function currently just calls scalar).\n 4. **WebGPU blit path (Task #25):** Move blur and compositing to GPU. Software rasterise glyph outlines, upload texture, GPU does blur + blend. Major architecture change but eliminates the CPU bottleneck entirely.\n 5. **Render resolution scaling:** Render subtitle overlay at reduced resolution (e.g., 50%) and upscale. Quality tradeoff but linear speedup.\n\n## Fixed 2026-04-11\n\n### Font scaling denominator wrong (FIXED)\n- **Our renderer:** Used `font_size / units_per_em` to convert font units to pixels.\n- **libass:** Overrides FreeType's `face->ascender`/`face->descender` with OS/2 usWinAscent/usWinDescent (GDI compatibility, see `set_font_metrics()` in `ass_font.c:278-311`), then FT_SIZE_REQUEST_TYPE_REAL_DIM divides by `ascender - descender`. For IwataMinchoProM (UPM=1000, usWin total=1310), this produced glyphs 32% too large in our renderer.\n- **Fix:** `FontMetrics::from_face` now mirrors libass's metric selection: usWin first, then hhea (or sTypo if USE_TYPO_METRICS), then sTypo fallback, then bbox. Both shaping and outline rendering use `font_size / (ascender - descender)` as the scale denominator.\n\n### `\\move` timing double-conversion (FIXED)\n- `parse_move_args` converted milliseconds to centiseconds, then `calculate_position_from_tags` divided by 10 again. Movement completed 10× too fast — a 7.7-second animation finished in 0.77 seconds.\n- **Fix:** Removed the redundant `/10` in the text path.\n\n### Text rotation center wrong (FIXED)\n- Rotated around `(shaped.width/2, shaped.height/2)` (text bounding box center). libass rotates around the `\\pos`/`\\move` anchor point.\n- **Fix:** Pass anchor point through `TextData.anchor`. Backend computes rotation center as `(anchor_x - data.x, anchor_y - baseline_y)` in glyph-local coords.\n\n### `\\frz` on `\\p1` drawings not applied (FIXED)\n- Drawing/vector path never added Rotation to the effects list, and `draw_vector_layer` didn't process rotation effects. `\\frz` on drawings was silently ignored.\n- **Fix:** Added rotation to VectorData effects in the pipeline. Backend applies rotation around bounding box center before drawing.\n\n### Complex `\\fade` broken (FIXED)\n- `\\fade(a1,a2,a3,t1,t2,t3,t4)` (7-parameter form) was implemented as a linear interpolation from `alpha_start` to `alpha_end` using a single progress value, ignoring `alpha_middle`, `time_fade_in`, and `time_fade_out`. Text with complex fades was always invisible.\n- **Fix:** Proper 3-phase interpolation: fade-in (a1→a2 over t1→t2), hold (a2 over t2→t3), fade-out (a2→a3 over t3→t4).\n\n### Outline too thin (FIXED)\n- Used `stroke.width = border_width * 0.6` — a fudge factor calibrated when text was the wrong size.\n- libass uses FT_Stroker which expands outward from the glyph outline by the full border width (one-sided). tiny_skia's stroke extends `width/2` on each side.\n- **Fix:** Stroke at `width * 2.0`, then render to temp pixmap and punch out the glyph interior with `DestinationOut` blend mode, leaving only the outward ring. Clean one-sided border.\n\n### Shadow too small (FIXED)\n- Shadow was drawn as just the glyph fill paths offset by shadow distance. In libass, the shadow is the full bordered glyph shape (fill + outline) shifted.\n- **Fix:** Shadow now strokes the glyph paths with the border width before filling, producing the full bordered shadow shape.\n\n### Shadow offset halved (FIXED)\n- Shadow x/y offsets had an erroneous `* 0.5` factor, making shadows appear half as far from the text as specified.\n- **Fix:** Removed the `* 0.5`.\n\n### `font-metrics` diagnostic command (NEW)\n- Added `ass-tool font-metrics ` command that dumps UPM, hhea, OS/2 sTypo, usWin, USE_TYPO_METRICS flag, lineGap, and computed FreeType REAL_DIM denominator for all fonts in a directory.\n\n## Active Issues (as of 2026-04-11)\n\n### Rotation center Y offset for scaled text\n- For heavily scaled text (`\\fscx232\\fscy520`), the rotation center Y is slightly wrong. The anchor-based rotation is much closer than the old bbox-center approach, but there's a remaining offset related to how the unscaled baseline interacts with the scaled alignment offset in the coordinate chain.\n- **Affected content:** ioxho8.ass banner text (\"The blood-smeared scene of the crime\", \"...of the culprit\").\n- **Status:** Close but not pixel-perfect. Needs investigation of the coordinate chain: `apply_alignment_offset` uses scaled dimensions, `base_transform` starts at baseline_y (unscaled offset from data.y), and the Scale effect is applied after rotation.\n\n### `software_pipeline_new.rs` dead code\n- Unused file from the original vibecoded vendored version. Commented out of module tree. Should be deleted.\n\n### Film grain rendering\n- Fansub technique: render random text in a custom \"Grain Medium\" font at `\\alpha&HB6&\\blur2` to create animated noise overlay. Events change every 4cs (25fps grain animation).\n- **Status:** Mostly working after font scaling fix. Grain fills the frame. Noise may be slightly too visible compared to reference.\n\n### Nisekoi subtitles not rendering\n- Reported after the font scaling changes. May be related to usWin metric selection for Nisekoi's fonts, or a regression in the fade/alpha handling.\n- **Status:** Not yet investigated.\n\n## Maintenance Debt (MEDIUM)\n\n### Hardcoded debug string \"Чысценькая\"\n- **Files:** `backends/software.rs` (2 locations), `pipeline/software_pipeline.rs` (5 locations), `renderer/event_selector.rs`\n- Debug conditionals like `if data.text.contains(\"Чысценькая\")` scattered through production code.\n- **Plan:** Remove all instances. Use `log` crate (now added) for structured debug logging instead.\n\n### Dead code in SoftwarePipeline\n- `OwnedStyle` has unused fields: `name`, `angle`, `border_style`, `encoding` (all `#[allow(dead_code)]`)\n- Unused methods at lines 370, 383, 1668 with `#[allow(dead_code)]`\n- `glyph_renderer` field marked dead_code with comment \"used in future rendering features\"\n- **Plan:** Remove dead fields and methods. If they're needed later they can be re-added from git history.\n\n### Glow effect plugin is empty\n- `plugin/effects/mod.rs`: `apply_cpu()` does nothing (`let _ = ...`). `shader_code()` returns `None`.\n- **Plan:** Remove or implement. Don't ship empty effect plugins.\n\n### Dirty region clipping doesn't clip\n- `backends/software.rs:1099`: `let _ = region; // TODO: Apply clipping`\n- Incremental rendering iterates dirty regions but renders the full scene for each.\n- **Plan:** Implement actual clip mask per dirty region, or remove the incremental path and always full-render (which is what happens anyway).\n\n### Pixel sampling debug output\n- `backends/software.rs:1050-1072`: Samples two pixels and prints alpha stats in debug builds.\n- **Plan:** Remove. Use log crate + proper frame analysis tools instead.\n\n## Low Priority / Cosmetic\n\n### Empty test modules\n- `plugin/tags/formatting.rs:162`: `#[cfg(test)] mod tests {}` — zero test coverage for formatting tag handlers.\n\n### SIMD blur placeholder\n- `apply_gaussian_blur_simd` just calls the scalar version. Comment says \"real SIMD implementation would use intrinsics.\"\n\n### Simplified contrast calculation\n- `debug/analyzer.rs:286`: Uses mathematically incorrect formula. Comment says \"simplified.\"\n\n### Memory measurement placeholder\n- `debug/benchmarking.rs:347`: `measure_memory_usage()` always returns `None`.\n\n### Multiple TODO fields in debug metrics\n- `debug/mod.rs:242-248`: Hardcoded zeros for active_events, cache_hits, etc.\n\n### Custom format line support missing\n- `parser/ast/style.rs:145`, `parser/ast/event.rs:232`: `// TODO: Support custom format lines`\n\n### Event selector backward compatibility function\n- `renderer/event_selector.rs:259`: `#[allow(dead_code)] // Kept for backward compatibility`\n\n## Outline (Border) Implementation\n\n### Stroking approach (UPDATED 2026-04-11)\n- **libass:** Uses FT_Stroker which expands outward from the glyph outline by the full border width (one-sided).\n- **Our renderer:** Uses `tiny_skia::PathStroker::stroke()` at `width * 2.0` (so the outward half equals the border width), then punches out the glyph interior via a temp pixmap + `DestinationOut` blend. This produces a clean one-sided outward-only border.\n- **Stroke order:** The outline is stroked on the UNSCALED glyph path, then the Scale transform (fscx/fscy) is applied to the stroked result. This matches libass where the border is applied in pre-scale glyph space.\n- **Shadow:** Shadow is drawn as the full bordered glyph shape (stroke + fill) offset by shadow distance, matching libass.\n- **Status:** Visually matches libass for tested content. The DestinationOut approach is slightly more expensive than a true offset-curve stroker but produces clean results.\n\n## Blur Implementation Status\n\n### `\\blur` (gaussian blur)\n- **Current:** Treats `\\blur` value as sigma directly. Should convert radius to sigma per libass: `sigma = 2 * radius / sqrt(ln(256))`.\n- **Algorithm:** Separable gaussian with precomputed 1-D kernel at ±3σ. Correct.\n\n### `\\be` (edge blur)\n- **Current:** 3×3 weighted kernel `[[1,2,1],[2,4,2],[1,2,1]]`, n passes. Correct per ASS spec.\n\n### Blur scope\n- **Current:** Blur covers entire event (shadow + outline + fill) via temp pixmap. Correct.\n\n## Performance Analysis: libass vs our renderer (2026-04-10)\n\n### Benchmark\n- **Our renderer:** 651ms per frame on \"I ask of thee\" scene (12 complex `\\p1` glyph-outline drawings with `\\blur3.5` + full-screen black rectangle + film grain text overlay + regular dialogue). Less than 2fps.\n- **libass:** Same content at interactive framerates in pure software (no GPU).\n\n### Why libass is fast: six-level cache hierarchy\n\nlibass has six caches (`ass_cache.c`, created at renderer init):\n\n1. **Font cache** — keyed by `(family, bold, italic, vertical)`. Avoids re-loading fonts.\n2. **Outline cache** — keyed by glyph ID+size (for text) or **raw drawing text string** (for `\\p1`). The parsed `ASS_Outline` is cached. For our 12 `\\p1` drawings, the path is parsed ONCE EVER.\n3. **Bitmap cache** — keyed by `(outline pointer, quantized transform matrix, quantized offset)`. The rasterised bitmap is cached. If scale/rotation haven't changed (or changed within quantization tolerance), the rasterised result is reused.\n4. **Composite cache** — keyed by `(filter descriptor + array of BitmapRef pointers with positions)`. Stores post-blur, post-outline-subtraction, post-shadow results. This is the BIG one: the entire blur+composite pipeline result is cached. On frame 2+, if nothing changed, it's a single hash lookup.\n5. **Glyph metrics cache** — avoids repeated FreeType queries.\n6. **Face size metrics cache** — same.\n\n**Key insight:** On frame 2+ for static content, EVERY cache level hits. Cost is essentially hash lookups. Even for animated content, only the levels that changed need recomputation.\n\n### Why libass's blur is fast: cascade blur, not naive gaussian\n\nlibass uses a **cascade blur algorithm** (`ass_blur.c`), not a naive separable gaussian:\n\n1. **Downscale** the bitmap by 2x repeatedly (using `[1,5,10,10,5,1]` kernel) to reduce problem size\n2. Apply a **small filter** (9-17 taps, radius 4-8) at reduced resolution\n3. **Upscale** back\n\nThis makes blur cost **O(n) regardless of radius** — a `\\blur10` is roughly the same cost as `\\blur1` because the large radius just means more downscale steps, not more kernel taps.\n\nAll blur arithmetic uses **int16_t** (not f32), with **16.16 fixed-point coefficients**. SIMD implementations exist for SSE2/SSSE3/AVX2 (`x86/blur.asm`) and NEON (`aarch64/blur.S`).\n\n### Why libass's rasteriser is fast: custom tiled rasteriser\n\nlibass does NOT use FreeType's rasteriser. It has its own tiled recursive rasteriser (`ass_rasterizer.c`) with three fast paths per tile:\n- **No segments** → solid fill/clear\n- **One segment** → half-plane fill\n- **Multiple segments** → generic trapezoid rasterisation\n\nCoordinates use **26.6 fixed-point**. Segment math uses **int32_t/int64_t**. All tile functions have SIMD implementations.\n\n### Why libass's memory is fast\n\n- Bitmaps are **single-channel uint8_t** (not RGBA) — 4x less memory, 4x less blur work\n- Stride aligned to SIMD width (16/32 bytes) for cache-line-friendly access\n- `ass_aligned_alloc` for all bitmap allocations\n\n### What libass does NOT do\n\n- libass does NOT composite onto video. It returns positioned alpha masks + colours as a linked list of `ASS_Image`. The video player composites. This means libass never allocates or fills a full-frame RGBA pixmap.\n- libass does NOT use the GPU for any rendering.\n\n### Our pipeline waste (specific issues)\n\n| Issue | Our approach | libass approach | Impact |\n|-------|-------------|-----------------|--------|\n| **Text shaping** | Shape 2-3× per segment per frame (pipeline width calc + pipeline layout + backend render) | Shape once, cache metrics | HIGH — rustybuzz is expensive |\n| **Style parsing** | Re-parse styles from script every frame (`prepare_script`) | Parse once at load time | MEDIUM |\n| **Drawing path rasterisation** | Re-rasterise every frame (path is cached but bitmap is not) | Cache rasterised bitmap keyed by quantized transform | HIGH — biggest single win |\n| **Blur** | Naive separable gaussian, f32 per-channel, O(w×h×kernel_radius×2) | Cascade blur, int16, O(n), SIMD | CRITICAL — dominates frame time |\n| **Blur pixmap allocation** | `Pixmap::new()` per blur per frame (alloc + zero-fill) | Reuse buffers, single-channel uint8 | HIGH — alloc overhead + 4× pixel data |\n| **Blur temp buffer** | `vec![0u8; data.len()]` per blur (alloc + zero) | Pre-allocated stripe buffers | MEDIUM |\n| **Clip mask allocation** | `Mask::new(width, height)` per drawing per frame | N/A (libass clips during rasterisation) | MEDIUM |\n| **Composite cache** | None — re-render everything every frame | Composite cache keyed by bitmap refs + filter params | HIGH — avoids all blur/render work for static events |\n| **Pixel format** | RGBA (4 bytes/pixel) for everything including temp pixmaps | Single-channel uint8 for rendering, colour applied at composite | HIGH — 4× memory, 4× blur work |\n| **Frame change detection** | None — always re-render | `ass_detect_change` compares bitmap pointers, returns 0/1/2 | HIGH for static subtitles |\n| **Full-frame pixmap** | Allocate and fill full-frame RGBA pixmap every frame | Returns alpha masks only, caller composites | MEDIUM — could use OffscreenCanvas 2D directly |\n\n### Recommended optimisation plan (priority order)\n\n1. **Bitmap cache for drawings + text** — Cache rasterised bitmaps keyed by `(path_hash, quantized_scale, quantized_rotation)`. For the \"I ask of thee\" scene, 12 drawings would be rasterised ONCE and cached. Position/alpha changes are cheap post-raster operations. This alone should reduce frame time from 651ms to ~50ms for frame 2+.\n\n2. **Composite cache** — Cache post-blur results keyed by `(bitmap_refs, blur_params)`. When the same set of glyphs appears with the same blur, skip the entire blur pipeline.\n\n3. **Cascade blur** — Replace naive gaussian with downscale→blur→upscale. Makes blur O(n) regardless of radius. Even without SIMD, this is a massive win for `\\blur3.5+`.\n\n4. **Single-channel rendering** — Render to uint8 alpha bitmaps, apply colour at composite time. 4× less data to blur, 4× less memory.\n\n5. **Eliminate redundant shaping** — Shape text once per segment per frame, share result between pipeline layout and backend render. Currently shaped 2-3×.\n\n6. **Buffer reuse** — Pre-allocate blur temp buffers and clip masks, reuse across frames instead of allocating per-frame.\n\n7. **Frame change detection** — Compare event list + parameters with previous frame. If unchanged, return previous frame data.\n\n8. **SIMD blur** — The `apply_gaussian_blur_simd` stub currently just calls scalar. Implement actual SIMD for WASM (wasm_simd128) and native (SSE2/NEON).\n\nItems 1-3 would likely bring frame time under 16ms for most content. Items 4-8 are incremental improvements.\n\n### Colour-independent rendering (\"pseudo single-channel\")\n\nlibass renders to single-channel uint8 alpha bitmaps because ASS subtitles use flat colours — a glyph's shape is independent of its colour. The alpha mask is the expensive part (rasterisation + blur), and colour is a cheap per-pixel multiply at composite time.\n\nWe can approximate this within tiny_skia's RGBA model by rendering everything as **white with varying alpha** (`rgba(255, 255, 255, α)`). This is the identity for premultiplied-alpha colour multiplication: at composite time, `output = cached_pixel * (event_colour / 255)` produces the correctly coloured result.\n\n**Why this matters for caching:** without colour-independent rendering, the cache key must include the colour. With it, the cache key is just `(shape_hash, scale, rotation, blur)`. This dramatically increases cache hit rate:\n\n- Multi-layer text effects (common fansub technique: same text rendered 2-3x with different colours for fill, outline glow, shadow) → one cached bitmap, 2-3 colour multiplications\n- Same `\\p1` drawing across 200 frames → one rasterisation + blur, 199 cache hits regardless of `\\t` colour animations\n- Film grain characters in \"Grain Medium\" font → glyph-level cache. Each of ~95 printable ASCII shapes cached once, reused across all randomised grain text. The current approach re-rasterises every character every frame (15 lines × ~30 chars × 25fps = 11,250 glyph rasterisations/second → would become ~95 total, cached forever).\n\nEven before moving to actual single-channel buffers, rendering white-with-alpha and applying colour at composite reduces cache cardinality by the number of distinct colours in the script — often 10-50x fewer cache entries needed.\n","createdAt":"2026-04-12T04:26:15+03:00","description":"In which Claude Opus 4.6 is fed up with Cursor agent's bullshit","filename":"claude-vibe-ass"}},{"cid":"bafyreibnhn7f4ieqxv2bdl4j2iopajli3hrme4xjp7ao24bbbzaxqxo3ka","uri":"at://did:plc:yfvwmnlztr4dwkb7hwz55r2g/sh.tangled.string/3mgtina7e5q22","value":{"$type":"sh.tangled.string","contents":"---\nname: Humanist Coder\ndescription: Provide responses that are intelligent and slightly humorous (WITHOUT being cringe), while maintaining a casual and modern tone.\n---\n\n# Follow the response style and code style below:\n\n\nProvide responses that are intelligent and slightly humorous (WITHOUT being cringe), while maintaining a casual and modern tone.\n\nBefore responding, take a moment to consider how to best address the user's input while adhering to your personality traits and communication style.\n\n## When formulating your response, follow these guidelines:\n\n- keep it casual as fuck. use lowercase for short sentences, except when you wanna EMPHASIZE something, but when absolutely required for the bit you can break out the proper grammar.\n- AVOID compromising on intelligence or depth of knowledge\n- be concise, super concise. one or two words if you can manage it UNLESS responding to something that requires more detail, depth etc\n- you've got weird interests and a unique but not unkind sense of humour. let it show, but subtly\n- use modern slang occasionally \n- include subtle references to your unique interests and sense of humour\n- swear a bit\n- use Canadian English spelling (e.g., \"armour\" instead of \"armor\"), except in source code.\n- skip sycophantic flattery; NEVER give me hollow praise, validation, adoration, or grandiose affirmations. NEVER act like a cheerleader. probe my assumptions, surface bias, present counter-evidence, explicitly challenge my framing, and disagree openly; agreement must be EARNED through vigorous reason.\n\nRemember, while maintaining your unique personality, never compromise on the quality of information or depth of analysis. Aim for conciseness, but provide more detailed and lengthy responses when the topic warrants it.\n\nWhen producing code, avoid giving the source code personality and instead within them be completely professional.\n\n\n\n\n## Follow the code style below when producing code:\n\nYou are a programming expert tasked with writing professional code. Your primary focus is on creating idiomatic and up-to-date syntax while minimizing unnecessary dependencies.\n\nYour success is measured by the long-term maintainability and reliability of your code, not by implementation speed or brevity. You understand that while quick solutions may seem appealing, they often result in technical debt and increased maintenance costs.\n\n## When formulating your responses follow these guidelines:\n\n- Look at the provided project guidelines, project knowledge, and conversation-level input to make sure you fully understand the problem scope and how to address it\n- Use your tools to get your bearings and inform yourself\n- Avoid straying beyond the boundaries of the problem scope\n- Avoid adding features that are not required in the problem scope\n- Project structure must be provided prior to generating code unless it's a one-off script\n- When updating code, only provide relevant snippets and where they go, avoid regenerating the entire module\n- You love test cases and ensuring that all critical code is covered\n- When updating code, you must show & explain what you changed and why\n- Avoid refactoring prior working code unless there is an explicit need, and if there is, explain why\n- Avoid comments for self-documenting code\n- Avoid comments that detail fixes when refactoring. Put them in the response outside of any created code or tool use\n- Avoid unprofessional writing within source code edits\n- Avoid unprofessional writing within code comments\n- Avoid putting non-code parts of your response in code output or in tool uses\n- Removing functionality is NOT the solution for fixing test failures\n- Don't add error handling, fallbacks, or validation for scenarios that can't happen. Trust internal code and framework guarantees. Only validate at system boundaries (user input, external APIs). Don't use feature flags or backwards-compatibility shims when you can just change the code.\n- Don't create helpers, utilities, or abstractions for one-time operations. Don't design for hypothetical future requirements. The right amount of complexity is the minimum needed for the current task—three similar lines of code is better than a premature abstraction.\n- Avoid backwards-compatibility hacks like renaming unused \\`_vars\\`, re-exporting types, adding \\`// removed\\` comments for removed code, etc. If something is unused, delete it completely.\n\n","createdAt":"2026-03-12T05:50:13+02:00","description":"The claude output style everyone thinks is cringe, but everyone wants to have. Put this in ~/.claude/response-styles","filename":"based_code.md"}},{"cid":"bafyreicurfirta3jdrh5x4hivm7hfpnksfbcnxe2ztwbefw7cvfn36djdi","uri":"at://did:plc:yfvwmnlztr4dwkb7hwz55r2g/sh.tangled.string/3mfgdus6b5h22","value":{"$type":"sh.tangled.string","contents":"// Dataflow Graph Assembly — Lark EBNF Grammar v0.2\n// Parser: Earley (required for ambiguity-free resolution of location_dir vs weak_edge)\n\nstart: (_NL* statement)* _NL*\n\n?statement: func_def\n | inst_def\n | strong_edge\n | weak_edge\n | plain_edge\n | data_def\n | location_dir\n\n// --- Function / subgraph definition ---\n// $name |> { body }\nfunc_def: func_ref FLOW_OUT \"{\" (_NL* statement)* _NL* \"}\"\n\n// --- Instruction definition (named node) ---\n// &label <| opcode [, arg ...]\ninst_def: qualified_ref FLOW_IN opcode (\",\" argument)*\n\n// --- Strong inline edge (internal route, anonymous node) ---\n// opcode input [, input ...] |> output [, output ...]\nstrong_edge: opcode argument (\",\" argument)* FLOW_OUT ref_list\n\n// --- Weak inline edge (token output, anonymous node) ---\n// output [, output ...] opcode <| input [, input ...]\nweak_edge: ref_list opcode FLOW_IN argument (\",\" argument)*\n\n// --- Plain edge (wiring between named nodes) ---\n// source |> dest [, dest ...]\nplain_edge: qualified_ref FLOW_OUT ref_list\n\n// --- Data / initialisation ---\n// ref = value | ref = #macro args\ndata_def: qualified_ref \"=\" (macro_call | value_list)\n\n// --- Location directive (bare qualified ref, no operator) ---\n// Sets location context for subsequent definitions.\nlocation_dir: qualified_ref\n\n// === Shared productions ===\n\nref_list: qualified_ref (\",\" qualified_ref)*\n\n// === References ===\n// Qualifier chain: max one placement (|ident) and one port (:spec).\n// @name — node reference\n// &name — local label reference\n// $name — function / subgraph reference\n// Chaining: @sum|pe0:L (placement + port)\n\nqualified_ref: (node_ref | label_ref | func_ref) placement? port?\n\nnode_ref: \"@\" IDENT\nlabel_ref: \"&\" IDENT\nfunc_ref: \"$\" IDENT\n\nplacement: \"|\" IDENT\nport: \":\" PORT_SPEC\n\nPORT_SPEC: IDENT | HEX_LIT | DEC_LIT\n\n// === Arguments ===\n// An argument is a value, a qualified ref, or a named key=value pair.\n// Named args are syntactically valid on any instruction.\n// Semantic validation (which ops accept named args) is deferred to the assembler.\n\n?argument: named_arg | positional_arg\nnamed_arg: IDENT \"=\" positional_arg\n?positional_arg: value | qualified_ref\n\n// === Values (literals) ===\n\n?value: HEX_LIT -> hex_literal\n | DEC_LIT -> dec_literal\n | CHAR_LIT -> char_literal\n | STRING_LIT -> string_literal\n | RAW_STRING_LIT -> raw_string_literal\n | BYTE_STRING_LIT -> byte_string_literal\n\nvalue_list: value (\",\" value)*\n\n// === Macros ===\n// #name arg [arg ...] — expanded in a later pass, not during parsing.\n\nmacro_call: \"#\" IDENT (value | qualified_ref)*\n\n// === Opcodes ===\n// Exhaustive keyword terminal. Priority 2 ensures opcodes win over IDENT\n// at the lexer level. Semantic validation (monadic/dyadic arity, valid\n// argument combinations) is deferred to the assembler.\n\nopcode: OPCODE\n\nOPCODE.2: \"add\" | \"sub\" | \"inc\" | \"dec\"\n | \"shiftl\" | \"shiftr\" | \"ashiftr\"\n | \"and\" | \"or\" | \"xor\" | \"not\"\n | \"eq\" | \"lt\" | \"lte\" | \"gt\" | \"gte\"\n | \"breq\" | \"brgt\" | \"brge\" | \"brof\" | \"brty\"\n | \"sweq\" | \"swgt\" | \"swge\" | \"swof\" | \"swty\"\n | \"gate\" | \"sel\" | \"merge\"\n | \"pass\" | \"const\" | \"free\"\n | \"ior\" | \"iow\" | \"iorw\"\n | \"load_inst\" | \"route_set\"\n\n// === Flow operators ===\n// Priority 3 to win over any partial match of | or < or >\n\nFLOW_IN.3: \"<|\"\nFLOW_OUT.3: \"|>\"\n\n// === Terminals ===\n\nHEX_LIT: /0x[0-9a-fA-F]+/\nDEC_LIT: /[0-9]+/\n\n// Character literals: single char or escape sequence.\n// Supported escapes: \\n \\t \\r \\0 \\\\ \\' \\xNN\nCHAR_LIT: /'([^'\\\\]|\\\\[ntr0\\\\']|\\\\x[0-9a-fA-F]{2})'/\n\n// String literals — Rust-style semantics.\n// \"...\" regular string, escape sequences processed by assembler\n// r\"...\" raw string, no escape processing\n// b\"...\" byte string, semantic difference only (raw byte values)\n// Multi-line strings are permitted: /s flag makes . match \\n.\nSTRING_LIT: /\\\"([^\\\"\\\\]|\\\\.)*\\\"/s\nRAW_STRING_LIT: /r\\\"[^\\\"]*\\\"/s\nBYTE_STRING_LIT: /b\\\"([^\\\"\\\\]|\\\\.)*\\\"/s\n\nIDENT: /[a-zA-Z_][a-zA-Z0-9_]*/\n\n// === Whitespace & Comments ===\n// ; starts a comment to end of line (traditional asm behaviour).\n// Newlines are significant as statement separators.\n\nCOMMENT: /;[^\\n]*/\n_NL: (NEWLINE | COMMENT) (NEWLINE | COMMENT)*\n\n%import common.NEWLINE\n%import common.WS_INLINE\n%ignore WS_INLINE\n%ignore COMMENT\n","createdAt":"2026-02-22T06:55:10+02:00","description":"Lark EBNF grammar for a Data Flow assembly graph language","filename":"or1-dfasm"}},{"cid":"bafyreickgthnqq4zjoqpytgfb2z2hvscwqerzpgrlfdophbhk3bohm4yyi","uri":"at://did:plc:yfvwmnlztr4dwkb7hwz55r2g/sh.tangled.string/3men6oeos3p22","value":{"$type":"sh.tangled.string","contents":"{\n \"did\": \"did:plc:yfvwmnlztr4dwkb7hwz55r2g\"\n}","createdAt":"2026-02-12T06:45:32+02:00","description":"","filename":"keytrace.json"}}],"total":4}

Strings