Skip to content

fix(chat): give the reply's close button its own ground - #732

Merged
bmc08gt merged 1 commit into
mainfrom
fix/chat-reply-dismiss-contrast
Sep 6, 2026
Merged

fix(chat): give the reply's close button its own ground#732
bmc08gt merged 1 commit into
mainfrom
fix/chat-reply-dismiss-contrast

Conversation

@bmc08gt

@bmc08gt bmc08gt commented Sep 6, 2026

Copy link
Copy Markdown
Collaborator

Follow-up to #730, which made the reply quote Liquid Glass over the transcript. That change moved the ground under the quote's dismiss control from an opaque slab to glass sampling whatever is behind it, and the control was still drawn for the slab: an 18pt .light xmark in textSecondary, the least dense thing on the bar. Its contrast now varied with which message had scrolled past.

Drawn as xmark.circle.fill instead, so the glyph carries its own ground.

The disc is a plain translucent fill (textSecondary at 35%), not a second glass layer. Nesting glass in glass muddies the layering; a control on a glass platter belongs on it, not above it. CancelEditButton in the same bar does get its own glassBackground, and that stays correct — the bar's surface under it is opaque backgroundMain, so that is one glass level rather than two.

Palette rendering rather than monochrome: a monochrome fill knocks the ✕ out as a transparent hole, which over glass fills with the transcript — the exact thing the disc is there to prevent.

SystemSymbol gains closeCircle so the strip keeps using the app's symbol vocabulary rather than a raw string. close is untouched and still used by CancelEditButton.

Both candidates were built and compared on device before picking this one; the alternative was matching CancelEditButton's 17pt semibold xmark in textMain.

The quote above the composer is now Liquid Glass sampling the transcript, so
the dismiss control's contrast moved with whatever message scrolled behind it.
A hairline ✕ at `.light` in `textSecondary` was the least dense thing on the
bar and had the least to spare.

Draw it as `xmark.circle.fill` instead. The disc is a plain translucent fill,
not a second glass layer — glass nested in glass muddies the layering, and on
a glass platter controls are meant to be drawn on it rather than above it.
Palette rendering rather than monochrome, because a monochrome fill knocks the
✕ out as a hole that the transcript then shows through.
@bmc08gt bmc08gt self-assigned this Sep 6, 2026
@bmc08gt
bmc08gt merged commit 44c0b55 into main Sep 6, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant