weft
ir 2.6.0 · warp 1.8.0

Every surface at once

Nothing here fails on an old engine. Every missing capability costs a wire form, a fill mechanism, or an animation.

chromium23 ms
firefox24 ms
webkit23 ms, a desktop proxy
It never aggregates engines, and it says “not measured” with a reason instead of reporting a zero.
✓ Valid Three engines, measured directly

Chromium, Firefox and WebKit, never aggregated, with WebKit labelled a desktop proxy rather than an iOS number.

✕ Refused --engines ios with no device attached
what you wrote→a desktop proxy has nothing honest to fall back on→refuses by name

It says “not measured” with a reason instead of reporting a zero.

In plain terms

A browser engine on a laptop is not a phone. Three engines are measured directly; iOS and Android are real lanes that refuse to run until a device is attached, because a desktop proxy has nothing honest to fall back on.

The target is every surface at once: desktop browsers, mobile browsers, embedded webviews and old devices. Nothing in this design fails on an old engine — every missing capability costs a wire form, a fill mechanism, or an animation, and never the page.

SurfaceEngineWhat it costs
Chrome, Edge, ElectronChromiumNothing. The Chromium-only bonuses are available.
Safari, iOS webviewsWebKitNo compression dictionaries and no speculation rules.
FirefoxGeckoNo cross-document view transitions in stable.
An old Android webviewAn old ChromiumIn-order streaming and full navigations, which is how the web worked.

A degradation ladder, not a feature check

Every mechanism on this site has a rung below it that the browser already implements. Out-of-order fill needs a 330-byte inline script; without it the page streams in order. A delta needs a resident template; without one the region arrives as html. Navigation staging needs an epoch; without it a click is a navigation.

Which means the bottom rung of every ladder is a page that works with JavaScript switched off.

The device tier is a read, so it is an axis

Low cardinality is what lets a read become an ahead-of-time permutation rather than a branch taken per request.

output

This deployment thinks you are on a device, so the plan reachable from here is the one built for that tier.

app/fragments/examples/device.tsx
import { fragment, type Ctx } from '@weftjs/core'

/**
 * A read with three possible values, which is what makes it an ahead-of-time permutation.
 *
 * `ctx.device()` taints `device`, and the plan can carry a branch per value rather than a branch
 * taken per request: low cardinality is the property that lets a read become an axis. The fragment
 * does not know which one it will be rendered for, and does not need to.
 */
export default fragment(async (ctx: Ctx) => {
  const tier = ctx.device()
  return (
    <p class="device">
      This deployment thinks you are on a <strong>{tier}</strong> device, so the plan reachable from here is
      the one built for that tier.
    </p>
  )
})
What the compiler knows
Holes1
Escape elided0
Effect setdevice
Wire formshtml · bundle · split · patch · delta

Holes

BindingHoleEscape
tiertextescape

What the client wires on adoption

No wiring: nothing on this template changes after it is painted, so the client attaches nothing to it.

Sealed templates
1
Version
63c7968954e0ae5445e34cdda1d5153e

The facts below show device in the read set. The plan can carry a branch per value; the fragment does not know which one it is rendered for.

Three values is low cardinality, and low cardinality is what lets a read become an ahead-of-time permutation instead of a branch taken per request. The plan can carry a branch per tier; the fragment never learns which one it was rendered for.