Changelog

All notable changes to this project are documented here.

1.0.0-beta.1 - 2026-08-01

First public release, as a beta on npm’s beta dist-tag while the real-device pass finishes. Everything below ships in it.

Core

• Native <dialog> top layer, driven by a CSS scroll-snap-type: y mandatory track instead of a pointer-event physics simulation. Desktop mouse drags suspend snap for the session so the sheet tracks the pointer exactly.
• Four sides: bottom, top, left, right; side sheets opt out of the browser’s own edge swipe-to-navigate gesture while dragging.
• Multi-detent snap points, non-modal mode, asChild on Trigger/Close/Handle/Title/Description.
• Spring-to-linear() easing with interruptible WAAPI enter/exit legs.
• Imperative actionsRef (open()/close()/snapTo(detent)) for controlling a sheet without a controlled prop.
• Public --scrollsheet-stack-progress CSS var (0-1) on a receded parent panel, for a consumer’s own stacking depth effect.
• Desktop presentation defaults at 768px and up: bottom sheets dock right as a floating card with margins, side sheets square their lead edge; --scrollsheet-desktop-margin and --scrollsheet-max-inline tune it, --scrollsheet-radius restores rounding.
• Per-side safe-area handling: status-bar clearance for top and side sheets, home-indicator clearance for bottom sheets, landscape notch insets on side panels.
• Keyboard engine pinned to the visual viewport, with an inset estimate that survives iOS Safari’s focus scroll compensation.
• --scrollsheet-travel: none styling hook to switch off the enter/exit travel animation, for sheets whose entrance is carried by something else (a View Transition morph, say). Degrades like reduced motion: the phase machine and callbacks still run, in zero time.
• Scroll-driven animations: on Chrome/Edge 115+ and Safari 26+, the backdrop dim and --scrollsheet-progress run as CSS animation-timeline: scroll() animations, off the main thread entirely, instead of a per-frame onTravel write. Feature-detected once and stamped data-scrollsheet-sda on the dialog; the JS travel path is unchanged and stays the only path everywhere else, including Firefox’s still-flagged support. A stacked sheet’s parent recede and backgroundEffect stay JS-driven on every engine; both live outside the track’s own DOM subtree.
• fill prop on Sheet.Content stretches [data-scrollsheet-body] to the panel (flex column) instead of sizing to natural content height, promoting the docs “Full-height content” recipe into the library. The content detent measurement and content-morph both switch to the body’s first child when fill is set.
• onTravel’s third argument, TravelInfo, carries the resolved travel range and per-detent progressAtDetents (0-1, keyed by each detent’s resolved px height). The object is reused and mutated in place every travel frame, not allocated fresh. Read it synchronously, don’t store the reference.
• backgroundRef gives backgroundEffect an explicit target to scale or parallax, skipping the document-wide [data-scrollsheet-background] query. The implicit default’s ownership check is unchanged.
• data-scrollsheet-nested-scroll on a scrollable element inside the body hands its scroll back to the sheet: once the element sits at its top boundary during a touch that started inside it, the same swipe continues as sheet travel, even under handleOnly or disableDrag. Gating is per touch identifier, so an unrelated second finger never widens the drag surface; elements a content morph adds or removes mid-session are picked up and cleaned up without re-wiring.
• Marked nested scrollers now get the panel’s own scrollbar treatment: 'overlay' hides the native scrollbar and shows an auto-hiding thumb, 'hidden' hides it with no thumb, 'native' leaves it untouched. Discovery is delegated and observed the same way as the handoff above, so scrollers a morph adds or removes are picked up without re-wiring.
• closeThreshold now governs wheel-dismiss gestures on single-detent sheets, not just pointer drags. Native scroll-snap still drives every frame; a wheel-idle correction pass only overrides the resting position when the accumulated gesture disagrees with where native snap actually landed.
• injectStylesInto(shadowRoot, nonce?) adopts the core stylesheet into a shadow root via adoptedStyleSheets, for a sheet whose trigger and surrounding markup live behind shadow encapsulation. Falls back to a <style> tag scoped inside the root on engines without constructable stylesheets. The dialog itself still portals to document.body regardless.
• Closed-form spring solver: sampleSpringAt is O(1) with no integration error. spring() now throws instead of silently truncating a curve that never converges within its max duration.
• scrollsheet/motion (experimental): the spring/WAAPI/scroll-tween/geometry core as its own zero-React subpath, for driving your own motion off the same math the sheet runs on.
• keyboardExpands on Sheet.Root: a sheet resting at a short peek detent promotes to its tallest detent the moment the software keyboard actually appears (a measured nonzero inset, never focus alone, so a desktop click into an input expands nothing), and restores the peek detent when the keyboard closes unless a drag or controlled change moved the sheet in between. Off by default.
• Keyboard clearance for top/left/right sheets: --scrollsheet-keyboard is now tracked on every side, and edge-anchored panels turn it into bottom padding on [data-scrollsheet-body] plus scroll-padding on the panel, so a focused field near the bottom of a side drawer scrolls clear of the keyboard instead of sitting under it. The focus auto-reveal runs on every side too, previously bottom-only.
• Fixed: content-morph respects prefers-reduced-motion: a detent-geometry change now lands instantly (scroll position and snap stops synced in one frame) instead of tweening the morph.
• Fixed: Trigger now records itself as the opener on click, so the implicit backgroundEffect ownership check also fires on Safari, where WebKit never focuses a clicked button.
• Fixed: sideEffects now exempts **/*.css, so a webpack/Next.js production build keeps import "scrollsheet/styles.css" instead of tree-shaking it away.
• Fixed: the "use client" banner now lands on every dist chunk, not just the entry point. Rolldown drops it from shared chunks otherwise.
• shadcn/ui registry item (sheet) served from the docs site.
• Modal sheets set dialog.closedby where the engine supports it (Chrome/Edge 134+, Firefox 141+): closerequest when escapeDismissible is on, none when it’s off, so Esc and Android back route through the platform’s own close-request machinery. Never any: .scrollsheet-dialog always covers the full viewport, so there’s no “outside the dialog” for a native light-dismiss to detect, and it would only risk double-firing against the library’s own backdrop-click handling. Unsupported engines keep today’s manual dismissal, unchanged.
• Non-modal sheets (modal={false}) create a CloseWatcher (Chrome 144+) while open and dismissible, so Android back closes the sheet instead of navigating the page; a modal showModal() dialog already gets this for free via closedby, non-modal ones don’t. Touch-primary devices only (hover: none plus pointer: coarse): Esc is also a platform close signal wherever CloseWatcher exists, and a desktop watcher would close the sheet on Esc pressed anywhere on the page, where non-modal Esc is deliberately scoped to focus inside the sheet. Device-verification-pending: confirmed in a desktop feature-detection harness, not yet walked through on a real Android device.
• Considered and dropped: declarative command/commandfor invoker support on Sheet.Trigger. The sheet’s <dialog> is portaled and only mounts after the first open, so there’s no element for a commandfor reference to target before that first open, the exact case Trigger exists to handle. No amount of wiring fixes a target that doesn’t exist yet.

Vaul compat

• Drop-in vaul compatibility: import { Drawer } from 'scrollsheet' maps vaul’s props (direction, modal, shouldScaleBackground, snapPoints, fadeFromIndex, closeThreshold, onAnimationEnd, onClose) onto scrollsheet’s own API.
• closeThreshold is converted between conventions: vaul counts the fraction dragged away, scrollsheet counts the fraction still visible, so vaul’s 0.25 default maps to scrollsheet 0.75 and a migrated drawer dismisses after the same quarter-height drag it always did. Only live when snapPoints is unset, matching vaul.
• Drawer.Handle accepts preventCycle to suppress click-to-cycle; emits data-vaul-handle.
• Drawer.Content emits data-vaul-snap-points and warns once (dev-only) instead of leaking vaul/Radix-only interaction props onto the DOM.
• Drawer.Content supports asChild with a single element child, merging props and ref onto it. A Fragment child (which can never receive a ref) degrades to default rendering with a dev-only warning instead of breaking the panel silently.

Sonner compat

• Drop-in Sonner compatibility: import { Toaster, toast } from 'scrollsheet' ships toast()/.success/.error/.info/.warning/.loading/.message/.custom/.dismiss, toast.promise(), useSonner(), and <Toaster> on scrollsheet’s own primitives.
• visibleToasts caps how many toasts render at once (default 3, matching Sonner’s own). Past the cap, the oldest dismissible toast evicts to make room, onDismiss firing same as a manual close. A dismissible: false toast holds its slot; a new arrival queues, untouched in the store, until one clears.
• Animation parity: CSS-transition enter/exit, per-row swipe-to-dismiss (45px threshold, adapted from Sonner’s own), and a crossfade between the collapsed and expanded stack view. prefers-reduced-motion collapses every duration to zero and disables the CSS transitions and animations outright.
• <Toaster id> is the primary prop for routing toasts to a specific instance; toasterId still works as a deprecated alias (dev-only warn-once).
• toast.promise() matches real Sonner’s own resolve-path semantics: a resolved Fetch Response with !ok, or a resolved Error instance, both route to the error branch; success/error may be sync or async functions; the returned id carries .unwrap(), resolving with the settled value or rejecting with the settled reason.
• classNames slot-styling: a per-toast classNames object and a Toaster-level toastOptions.classNames default merge onto every rendered slot (toast, title, description, icon, loader, content, cancel/action/close buttons, and per-type).
• icons prop on <Toaster> overrides the default per-type glyphs, including the loading spinner; a per-toast icon always wins.
• injectToastStylesInto(shadowRoot, nonce?), a sibling of the core injectStylesInto for a <Toaster> mounted inside a shadow root.
• Documented gaps (not yet implemented): dir/RTL, hotkey, mobileOffset, object-form offset, toastOptions fields beyond what’s listed above, the “extended result object” form of promise()’s success/error, and React.isValidElement/function-form description handling inside promise().

Accessibility

• Focus lands inside the panel on open; Title/Description wire aria-labelledby/aria-describedby automatically.
• Multi-detent Handle exposes slider semantics (role, value, arrow/Home/End keys).

Size

• 19.4 kB gzipped, 17.2 kB brotli, zero dependencies, React 18+. Bundled the same way, vaul ships 21.6 kB with its required Radix tree.