The column moved sideways the moment segmentation finished. It takes the
widest minimum any panel declares, each panel publishes
`layout.preferred-width` as that minimum, and `MaskPanel` gained rows
whose width came from the model's output -- so a photograph the user was
looking at jumped because a label said "traffic light".
`overflow: elide` did not prevent it and was never going to. Eliding is
what a `Text` does when it draws; at layout time it still asks for the
width of its whole string, and it is the asking that reaches the column.
Both the subject rows and the mask entries are bounded, because a mask is
itself a segmentation result -- without the second half the column moved
when a subject was clicked instead of when one was found.
The bound is stated for the reason `ChipGrid` declares its width from its
column count rather than from its options (ff6c313): what a panel asks
for must follow from its structure, never from its data. 160px is the
same judgement as that commit's 88px chip, and it is a policy rather than
a measurement -- there is no harness in the tree that measures a panel's
width, and the 482-to-351 figure in ff6c313 was read off the running app.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
dtourolleandClaude Opus 5
3e19628324