The body below the header had a 160-point floor, so with the minimum capped by a screen lower than 220 points the content ran past the window. It now takes the height left below the header (the table scrolls). Test on_a_short_screen_the_variables_window_fits_inside_it (red before: the window ran to y 206 on a 180-high screen); variables 45 passed, clippy clean (CodeRabbit on #2363).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The minimum (600 x 240) is capped at the screen less its margin, so on a screen narrower than that the window stays inside it. The window is pinned at its top-left corner and opens centred, so dragging one edge leaves the opposite one in place (the centre pivot moved both). Tests on_a_narrow_screen_the_variables_window_fits_inside_it (red before: the window ran from x -26 to 576 on a 550-wide screen) and the top-edge check in the_variables_window_grows_when_dragged_taller (red before: 210 -> 190); variables 44 passed, clippy clean (CodeRabbit on #2363).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
At 480 points the table was 279 wide, leaving 11 points next to the 220-point search field for a library collection's Extend collection button. The minimum is now 600 x 240 (table 399 wide); the default size's clamp uses the same minimum. variables 43 passed, clippy clean (CodeRabbit on #2363).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The window was fixed at 880 x 480. It now resizes from every edge and corner, down to 480 x 240 and up to the screen less a margin, and its header, sidebar and table follow its current size. Egui's memory keeps the size and place while the app runs.
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
Both moves had names of the same length, so which one ran first came
from the order of the variables. The first one now has a longer name,
which the longest-first order always moves first, so the second one
always collides after a move that has to be undone.
The collision test's two moves had names of the same length, so the
first one already collided and nothing was left to undo. With the
existing name on the second move, the first moves, the second collides,
and the test checks that the batch puts the first one back.
Renaming a group in the Local variables window replaced every slash in the
typed name with a hyphen, so "Palette/Orange" gave a flat Palette-Orange
group and there was no way to move a group under a new parent.
The typed path now replaces the group's last part, so a slash nests the
group as in Figma: Orange/100 becomes Palette/Orange/100 and the sidebar
shows Orange under Palette. Each part is trimmed and empty parts are
dropped; a name that comes out empty or unchanged does nothing. The
renames go deepest name first, so a group moved into its own new subgroup
(Orange to Orange/Warm) never collides with a variable that is still to
move. A rename onto a variable name that stays is refused by
variable.update and the batch rolls back; it stays one undo step.