fix(fab): fan out the first fab-flower action when no main-action is set

[JamBalaya56562]

Fixes #4531

Problem

In .fab-flower the center pin was > *:nth-child(-n + 2) { --position: 0rem; }, which pinned both the trigger (1st child) and the first action (2nd child) to the center. So when no .fab-main-action / .fab-close is present, the first action button overlaps and "replaces" the trigger instead of fanning out into the quarter-circle arc.

As reported by @topherfangio, this should only happen when a button is explicitly marked .fab-main-action. (@saadeghi: "You're right. Will be fixed in v5.6 soon.")

Fix (minimal / additive)

Following review (@pdanpdan, @saadeghi) the change is kept as small and additive as possible:

1. Center by identity, not position — pin the trigger plus .fab-main-action / .fab-close (> *:nth-child(1), > .fab-main-action, > .fab-close).
2. The existing arc blocks (:has(:nth-child(3..6))) are left untouched — Mode A (with a main-action/close) is byte-identical to v5.6, so there is no regression risk.
3. A single block is appended for the no-main-action case: &:not(:has(.fab-main-action, .fab-close)), mirroring the existing degrees shifted by one child (the arc starts at the 2nd child) and hiding :nth-child(n + 6) so it caps at the same 4 actions as Mode A.

Net diff vs v5.6 is just the identity-pin line + the appended :not(:has()) block. (An earlier revision wrapped the existing rules in :has(.fab-main-action)/:has(.fab-close); that wrapping has been reverted.)

Playground

▶️ Before / after, side by side: https://play.tailwindcss.com/LifJWa8hZG

Rows are action counts, left column is the released (buggy) daisyUI, right column is this PR. Heads-up: Play loads the released daisyUI, so the right column re-creates the fix's result with a clearly-labelled preview shim on top of it — that shim is more verbose than this PR (it has to override the released @layer rules). The real CSS is the small diff above.

Verification

bun run build succeeds (lightningcss accepts the :not(:has()) selectors) and bun test passes (86/86).

Headless-Chromium measurement of each action's center offset from the trigger (button = 48px), opening each FAB via :focus-within:

case	trigger	center button	arc actions (dist @ angle)
no main / 1	0px	—	A 67px @135°
no main / 2	0px	—	A, B 67px @165° / 105°
no main / 3	0px	—	A, B, C 86px @180° / 135° / 90°
no main / 4	0px	—	A, B, C, D 106px @180° / 150° / 120° / 90°
fab-main-action / 3	0px	0px (centered)	A, B, C 86px @180° / 135° / 90°
fab-close / 3	0px	0px (centered)	A, B, C 86px @180° / 135° / 90°

• Without a main-action the first action goes from overlapping the trigger (dist 0) to fanning out (dist > 0) at the correct angles.
• With fab-main-action / fab-close the center button stays on the trigger and the arc matches the unchanged v5.6 geometry.

Notes

• Single-component change in packages/daisyui/src/components/fab.css, plus a docs example (fab-flower without a main action) and a capacity note.

🤖 Generated with Claude Code

[pdanpdan]

Hello.

I would say that the solution is a little bit convoluted.
I would go with:

• apply --position: 0rem; only to the FAB button and .fab-main-action and .fab-close (like in PR)
• keep the current CSS for flower
• at the end add the styles for &:not(:has(.fab-main-action, .fab-close)) in the same way they are done above - only change the number of children
• also for this case > :nth-child(n + 6) should be hidden

[saadeghi]

Thanks for the PR.
I would appreciate if the CSS rule changes stay as minimum as possible. When this many lines are changing in a PR, we have to predict and test every possible situation to make sure it's not breaking existing apps in the next update.

[JamBalaya56562]

Thanks @pdanpdan and @saadeghi — updated to keep the change minimal and additive, exactly along the lines you suggested:

• Kept the identity-based center pin (> *:nth-child(1), > .fab-main-action, > .fab-close).
• Reverted the wrapping of the existing arc blocks — they are now byte-identical to v5.6, so Mode A (with a fab-main-action / fab-close) is completely untouched and carries no regression risk.
• Appended a single &:not(:has(.fab-main-action, .fab-close)) block, mirroring the existing degrees with the child numbers shifted by one, plus > :nth-child(n + 6) hidden (so the no-main case caps at the same 4 actions as Mode A).

So the net diff vs v5.6 is now just the identity-pin line + that one appended block. Pushed as a new commit, and the description has the headless before/after numbers (no-main fans out, with-main/close stays centered and unchanged). 🙏