Responsivity — how it works

Two small JavaScript engines let a component react to the space it is actually given — not the size of the window. That is the difference between a media query (asks the viewport) and a container query (asks the box). Pure Admin ships both a CSS-native path and a JS path, because some responsive moves (mounting a chart, moving a block into a menu) can't be done in CSS alone.

The two engines

1 · Fit — degrade a row to fit

Fit watches a horizontal container and, when the content can't fit, degrades the lowest-priority slot first, restoring as space returns. The navbar auto-inits; any other flex row opts in with pureAdmin.components.fit.init(el). You mark participants with attributes:

<div class="toolbar" data-pc-fit-default-priority="20">
  <span data-pc-fit="steps" data-pc-fit-priority="30">   <!-- full → smaller → gone -->
    <span data-pc-fit-step="0">Pure Admin</span>
    <span data-pc-fit-step="1">PA</span>
  </span>
  <span data-pc-fit="hide" data-pc-fit-priority="10">v2.9.0</span> <!-- drops first -->
  <button data-pc-fit-ignore>Save</button>                 <!-- pinned, never yields -->
</div>

• data-pc-fit="hide" — remove the slot when it must yield.
• data-pc-fit="steps" — show the largest ranked variant that fits (logo → wordmark → monogram).
• data-pc-fit="relocate" — move the slot somewhere else entirely (next section).
• data-pc-fit-priority — lower degrades first; data-pc-fit-auto folds in every child of a container; data-pc-fit-ignore pins one out.

2 · Relocation — fit decides, a sink places

A relocate slot doesn't just hide — it moves out of the row into a named destination and comes back on widen. Crucially, the Fit engine only decides a slot must yield; where it goes is a pluggable sink named by data-pc-fit-target. Two sinks ship built-in:

• target="floating-menu" — folds into a "•••" flyout panel. Self-contained; needs no sidebar.
• target="sidebar" — rebuilds the slot as a sidebar list item.
• your own — pureAdmin.components.fit.registerSink('name', { out, in }).

Live — drag to narrow the bar. The badge cluster is one relocate slot targeting floating-menu; when the bar runs out of room it folds into the "•••" panel, and returns when you widen. "New" is pinned with data-pc-fit-ignore.

<span data-pc-fit="relocate" data-pc-fit-target="floating-menu">
  <span class="pa-badge">Users 123</span> …
</span>

3 · The event & hands-off mode (Svelte / Phoenix)

Before it moves anything, Fit fires a cancelable pc:fit-relocate event on the slot (detail = { action: 'out' | 'in', target, container }). A plain HTML page lets the built-in sink do the DOM move. A reactive framework does the opposite: it owns placement itself.

Set data-pc-fit-managed (or call preventDefault()) and Fit performs no DOM surgery — it only hides the slot in-row and tells you it flipped. The framework then re-renders the block in its new home from state. That solves the staleness trap: a moved DOM node keeps its old Users: 123; a re-render is bound to the live store, so the relocated copy is always fresh.

// fit fires a DOM CustomEvent on the slot. In managed mode NO node is moved —
// you attach the listener the way your framework attaches DOM listeners.
// Plain JS (works everywhere):
el.addEventListener('pc:fit-relocate', (e) => {
  relocated = (e.detail.action === 'out');   // then RE-RENDER the block from state
});

// In Svelte you'd wire that in an action / onMount (NOT an `onpa:...` attribute —
// the colon isn't valid there), then render whichever home is current — one
// source of truth, always fresh, never a moved stale node:
{#if relocated}
  <Flyout><Stats {users} {rooms} {subs} /></Flyout>
{:else}
  <Stats {users} {rooms} {subs} />
{/if}

4 · Container Breakpoint — named modes at set widths

When a component restyles wholesale at set sizes — not a row shedding pieces, but a card that becomes a grid, then tabs, then icons — declare width thresholds and let the engine name the band. It's the JS counterpart to a CSS @container query, for the cases CSS can't reach (mount a chart, push to the server).

<div phx-hook="…" data-pc-breakpoints='{"icons":0,"tabs":34,"grid":64}'
     data-pc-breakpoint-initial="tabs">
  <div data-pc-show="grid">…rich grid…</div>        <!-- .d-none unless mode = grid -->
  <div data-pc-show="tabs icons">…compact…</div>
</div>

• Thresholds are rem by default (root font is 10px, so 34 = 340px, 64 = 640px); add data-pc-breakpoint-unit="px" for pixels.
• The engine reflects the band to [data-mode] (CSS can key off it) and toggles .d-none on data-pc-show pieces.
• It fires pc:breakpoint only on a flip — the hook for "mount on demand" (build the chart in grid, destroy it otherwise).
• A small hysteresis dead-band stops flip-flopping right at a threshold — important, because a flip mounts/unmounts, not just a CSS toggle.

Which one do I reach for?

Ready to see Fit degrade real cards and toolbars across several strategies? Head to Fit to Size for the worked, slider-driven examples.