From 003f709eb239e816ebfc3a2a2bf5904af77d9609 Mon Sep 17 00:00:00 2001 From: dotCooCoo Date: Fri, 6 Nov 2026 22:28:23 -0800 Subject: [PATCH] Docs: give the open's kept-layout timings in CPU cycles on current main --- docs/target-app-parity.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/target-app-parity.md b/docs/target-app-parity.md index a082d30b5..0bed907fa 100644 --- a/docs/target-app-parity.md +++ b/docs/target-app-parity.md @@ -1354,7 +1354,7 @@ Moved from the ROADMAP's "Shipped so far" on 2026-10-10. Update it when a featur When a path object moves without changing shape, the renderer draws its stroke from the outline it made before the move, moved by the same offset. It outlines the stroke again when the path is scaled, rotated or edited, when the stroke changes, and when the move takes the view past the part of a long path it outlined. The pixels are the same. At 1600 × 1200, the frame after a 1 pt nudge takes 123 ms instead of 403 ms with an open path of 200,000 anchors in view (103 ms with nothing changed), and 13 ms instead of 26 ms with a closed path of 50,000 anchors whose stroke is dashed and has a width profile. When the stroke of a path moved this way has a gradient along or across it, the renderer also draws the gradient from the slices it made before the move, moved by the same offset. It makes the slices again when the path is scaled, rotated or edited, and when the stroke or its gradient changes. The pixels are the same. On Windows at 1600 × 1000, the frame after a 1 pt nudge of an open path of 200,000 anchors takes 75 ms instead of 1,237 ms with the gradient along the stroke and 69 ms instead of 1,420 ms with it across, and with an open path of 50,000 anchors 22 ms instead of 313 ms and 21 ms instead of 352 ms (medians of five interleaved runs). The renderer keeps the path of a path or compound path of 1,024 anchors or more by object, with the bounds of its segments and of its miter spikes in runs of 128 segments. When anchors move and the subpaths and anchor counts stay the same, the next frame changes only the segments that start or end at a moved anchor and measures again only the runs that hold them or the segments with length before them, instead of making the whole path again and measuring it whole for culling. The pixels and the cull bounds are the same. On Windows at 1600 × 1200, the frame after one anchor moves takes 8.2 ms instead of 30.1 ms with an open path of 200,000 anchors, 4.3 ms instead of 10.4 ms with a closed path of 50,000 anchors, and 6.6 ms instead of 11.9 ms with the filled closed path of 50,000 anchors and a dashed stroke (medians of five runs of `anchor_bench`). - Opening a document keeps the layouts it makes for its type objects, up to 128 MiB, and the first frame uses them instead of laying those objects out again. It does not keep a layout with bitmap or color glyphs or a text object with a gradient. The frame uses a layout only for a text object equal to the one laid out, and only while the fonts and the hyphenation exceptions have not changed. It uses each layout once. The renderer drops the rest when a frame that used one ends, and all of them after eight frames that use none. The frames of a threaded story are laid out while the file is read, and the first frame lays them out again. The pixels are the same. On Windows at 1600 × 1200 with other builds running, in medians of interleaved rounds of a local example, with 5,000 point type objects in view the first frame takes 242 ms instead of 280 ms (103 ms instead of 212 ms in the fastest round) and the open takes 237 ms instead of 178 ms (170 ms instead of 143 ms); with one area type object of 500 paragraphs the first frame takes 211 ms instead of 464 ms. + Opening a document keeps the layouts it makes for its type objects, up to 128 MiB, and the first frame uses them instead of laying those objects out again. It does not keep a layout with bitmap or color glyphs or a text object with a gradient. The frame uses a layout only for a text object equal to the one laid out, and only while the fonts and the hyphenation exceptions have not changed. It uses each layout once. The renderer drops the rest when a frame that used one ends, and all of them after eight frames that use none. The frames of a threaded story are laid out while the file is read, and the first frame lays them out again. The pixels are the same. On Windows at 1600 × 1200, in medians of seven interleaved rounds of a local example, counted in the main thread's CPU cycles because other builds were running: with 5,000 point type objects in view the first frame takes 409 million cycles instead of 894 million and the open 816 million instead of 665 million, so the two together take 1,225 million instead of 1,559 million; with one area type object of 500 paragraphs the first frame takes 697 million instead of 1,193 million and the open 535 million instead of 568 million. Type › Show Hidden Characters keeps the marks it places on each text object from one frame to the next. The canvas lays a text object out again when an edit changes it or its threaded story, or when the last frame that showed the marks did not draw it, and it lays out all type in view again when the fonts or the hyphenation exceptions change. With 1,000 two-line labels in view, drawing the marks takes 0.8 ms per frame in a release build instead of 19.7 ms. A selection with more than 100,000 points to outline shows its bounds instead of each object's outline and anchors. Paths count their anchors. Area type and type on a path count the anchors of their frame or path, and point type counts six: its box's corners and the ends of its baseline marker. Images, symbol instances, placed documents, and area type or type on a path in perspective count the four corners of their box, and gradient meshes and envelopes count their points and the ends of their mesh lines. The hover highlight leaves out an object past that count, and such an object shows its bounds as the key object. With 32,768 type objects, images, symbol instances, gradient meshes or envelopes selected, the selection's outlines and anchors add fewer than 100 vertices to the canvas's mesh (#2460). The canvas keeps the selected objects' outlines, anchors and handles from one frame to the next as one mesh. It builds an object's marks again when the object, its layer color or its selected anchors change. A pan moves the kept outlines and handles and places the anchor squares again. Each subpath and handle gets its triangles the first time it is in view. A frame with a new zoom or rotation draws the marks one shape at a time and keeps nothing, and the next frame at that view builds them. The pixels are the same, apart from two small differences: a kept outline's corner of exactly 90 degrees in a rotated view can keep its miter or cut corner after a pan, and kept vertices can move by rounding errors. On Windows at 1440 × 900, in a release build, with 2,000 selected circles of 50 anchors each in 10 fill colors, an idle frame takes 12.8 ms instead of 59.9 ms with the Selection tool and 29.8 ms instead of 192.1 ms with Direct Selection (medians of five runs). A frame during a zoom takes about what it did: 86.7 ms instead of 80.0 ms with the Selection tool.