Docs: qualify the eyedropper row for a zoomed-out image (#2192 review)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Model-Effort: Opus 5.5 drove; Opus 5.5 briefed; Sonnet 5.5 wrote at medium

docs/ui-parity.md

@@ -100,7 +100,7 @@ interaction feel not yet compared.
 | Text: wrap style (Auto, Balance, Pretty) | done | per text layer (`wrapStyle` on `node.set`, `text.wrapStyle`, the Type settings control under Vertical trim, inspect, CSS `text-wrap: balance` or `pretty`): Auto breaks greedily; Balance narrows each paragraph as far as it goes without another line, so its lines come out close to the same length (what browsers do for `text-wrap: balance`); Pretty keeps lines full and, when the last line is one word, narrows one of the last four lines just enough to give a word to the lines after it (Explore text properties: "Auto: Figma fits as many words as possible on each line before wrapping to the next", "Balance: Figma distributes text as evenly as possible across every line, so each line is close to the same length", "Pretty: Figma keeps each line full, but adjusts the last four lines to avoid an orphan"). Saved line breaks from a file still win. Not yet included: per-paragraph settings, text-style support and the `.fig` import/export mapping (#1827). Test: `text_behaviour_tests::balance_and_pretty_wrap_styles_break_lines_as_documented` |
 | Text: direction override (Text ▸ Text direction) | done | per paragraph (`ParaList::direction`: auto, ltr, rtl): Text ▸ Text direction ▸ Auto / Left to right / Right to left and `text.direction {direction, ids?}` set the edited paragraphs while editing, else every paragraph of the selected texts; `node.set textDirection` the layer's (or the chars `start`/`end` name); the Type settings control appears once a right-to-left script is in the text or a direction is forced; inspect reports `textDirection` when the paragraphs agree and `lineDirections` otherwise; Developer Mode's CSS writes `direction: ltr` or `rtl` when every paragraph is forced the same way (with `text-align: left` beside `rtl`, as Left is absolute on the canvas); `.fig` import and export carry each line's `sourceDirectionality`, read up to the paragraph count. Layout gives the paragraph its base level, so a forced right-to-left paragraph of Latin letters starts at the right, mirrors its list gutter and indent and orders its runs right to left, and a forced left-to-right paragraph of Hebrew the other way ("By default, Figma automatically handles text direction based on language detection. However, controls are available to override text direction per paragraph", "Go to Text > Text direction. Select Left to right or Right to left", Add right-to-left text). Not included: a direction in text styles (the article names none). Tests: `text_behaviour_tests::text_direction_is_forced_per_paragraph`, `fig::text_direction_tests` |
 | Text: superscript and subscript | done | `sups` / `subs` use the font's own glyphs where it has the feature and every letter or digit of the run gets one; otherwise the standard glyphs are drawn shrunk and raised or lowered by the font's OS/2 superscript and subscript metrics (0.65 em at 0.35 em up or 0.14 em down without them), and a run mixing available and unavailable glyphs is synthesised whole, as Explore text properties says ("Figma draws a synthesized version by using the font's standard version of the character, shrinking it, and positioning it accordingly"; mixed glyphs "all update to the synthesized form for consistency"), #1828; tests `text::tests::superscript_native_or_synthesised_and_never_mixed`, `text_behaviour_tests::superscript_and_subscript_are_synthesised_without_glyphs_for_them` |
-| Eyedropper (I) | done | picks the colour seen under the pointer (opaque) into the selection's fill and the recent colours, one undo step (`fx_behaviour_tests`): one screen pixel centred on the pointer, sampled as the canvas draws it, so an image magnified past 2x gives the pixel under the pointer, not a blend with its neighbours (#2189, `eyedropper_image_tests`); with Cmd/Ctrl+Shift held it makes a colour style or variable of it instead, as the reference app's eyedropper does: a Create from color window asks which, its name (the next free "Color n") and a variable's collection (the file's first, or a new one made with it in the same undo step), and leaves the selection's fills alone (`eyedropper_create_tests`, `sampled_color_tests`) |
+| Eyedropper (I) | done | picks the colour seen under the pointer (opaque) into the selection's fill and the recent colours, one undo step (`fx_behaviour_tests`): one screen pixel centred on the pointer, sampled as the canvas draws it when zoomed in, so an image magnified past 2x gives the pixel under the pointer, not a blend with its neighbours; zoomed out it samples the full-resolution image (the canvas may show a smaller copy, which the eyedropper does not use), so it can differ slightly from the shown colour (#2189, `eyedropper_image_tests`); with Cmd/Ctrl+Shift held it makes a colour style or variable of it instead, as the reference app's eyedropper does: a Create from color window asks which, its name (the next free "Color n") and a variable's collection (the file's first, or a new one made with it in the same undo step), and leaves the selection's fills alone (`eyedropper_create_tests`, `sampled_color_tests`) |
 | Comment (C), Shift+C to show/hide | done | local only |
 | Text on a path (Figma Draw) | done | the Text on a path tool in the Text flyout (`tool.textOnPath`) shows the I-beam over a shape's or vector's path, and a click there puts new text on the path, starting at the click, with the path's effects and its fill when one shows; the path layer is removed, and the path comes back if nothing is typed. Also from the Design panel or `text.onPath`; the path is edited by resizing only |
 | Stretch and scatter brushes (Figma Draw) | done | eight presets drawn in code; the Pencil's brush is set by `tool.pencilBrush`, not from the toolbar |
@@ -188,7 +188,7 @@ in the table. `crates/ui-egui/src/shortcut_keys_tests.rs` drives the keys a look
 
 | Date | Change | Summary |
 |---|---|---|
-| 2026-11-09 | patch | M3, #2189: the eyedropper picks the colour the canvas shows. It sampled a 1x1 render without the canvas's magnified-image pixels, so on an image magnified past 2x it blended a 1 px line with its neighbours (#3D444D picked as #343B44 beside #151B23), and the pixel it sampled began at the pointer instead of being centred on it. Test: `eyedropper_image_tests` |
+| 2026-11-09 | patch | M3, #2189: the eyedropper picks the colour the canvas shows. It sampled a 1x1 render without the canvas's magnified-image pixels, so on an image magnified past 2x it blended a 1 px line with its neighbours (#3D444D picked as #343B44 beside #151B23), and the pixel it sampled began at the pointer instead of being centred on it. It now matches the canvas when zoomed in; zoomed out it still samples the full-resolution image, not the smaller copy the canvas may show. Test: `eyedropper_image_tests` |
 | 2026-11-11 | patch | M5, gap 4: marquee selection as measured in Figma's web app (Select layers and objects, Selection marquee): a partly covered top-level frame or section gives the layers in it the marquee touches (it was selected only when enclosed, so nothing came), an empty frame is selected on touch, a rotated shape is touched by its outline and not its box, a drag from inside a frame selects that frame only when the marquee also takes another top-level frame (then whole), and also takes the top-level layers it touches outside it, and selecting a frame whole keeps the marquee at the top level (a partly covered frame then comes whole, not its layers). Flipped: `marquee_selects_what_it_touches`, `cmd_marquee_selects_nested_layers`, `a_box_selects_like_a_marquee`. Tests: `marquee_behaviour_tests`. Counts unchanged |
 | 2026-11-08 | minor | Vector edit mode, #1875, measured in Figma (web, real keys): Copy with segments selected takes only those (with just a point selected, the segments touching it) and a paste outside edit mode is a new vector layer in place; Cut removes them, the path splitting into subpaths in the same layer with edit mode staying on, and a paste outside gives a new layer with the cut segments; Paste while editing a vector adds the clipboard's segments, or a copied layer's outline, to the same network in place, selecting the pasted points, with no new layer and the bounds growing. FigCraft copied the whole layer, Cut left nothing to paste and Paste did nothing. `edit.copy`, `edit.cut` and `edit.paste` act so with no `ids`/`parent` given (Ctrl+X is the real Cut again; Shift+Delete still heals); the copy carries the layer's look (fills, strokes, effects, opacity, blend; the width profile only when every segment is copied); layers with no outline are skipped. Tests: `vector_clipboard_tests` |
 | 2026-11-08 | patch | Selection: Escape with layers selected clears the selection instead of stepping out to the parent first. Measured in the Figma web app with a rectangle in a group in a frame, selected from the canvas, from the Layers panel and after Tab: one Escape selects nothing. Editing, gestures and the pen still end first |