Dispatch is a weekly letter for engineers who read source code the way other people read novels. Every Tuesday: the pull requests that shifted a paradigm, the RFC debates worth your lunch break, and the quiet library releases nobody tweeted about.
One email. Every Tuesday. Unsubscribe in one click.
This week's table of contents
+ 6 more items in the full edition →
Trend observation
37% of new Vite configurations audited this month now ship with Rolldown as the default bundler — up from 4% in Q3 2025. The ecosystem is moving faster than the migration guides.
The auto-memoisation story is subtler than the blog post implies. Three engineers dissect the generated output and find the edge cases.
Rolldown is faster, but the plugin API surface changed in ways that will quietly break a surprising number of community plugins.
Two years of polyfill purgatory ends. The spec's coordinate model is worth re-reading now that it's actually shippable.
Deno, Bun, and two cloud providers now share a common API surface for fetch, streams, and crypto. What this means for 'write once, deploy anywhere.'
This week's lens
Partial hydration, TC39 Stage 3 proposals, and the V8 compiler tier that framework authors are only now beginning to exploit. Combined reading time: ~54 minutes. Worth every one.
It started as a Google Research preprint in 2023. Now it's shaping two major frameworks' architectures in divergent ways. We read both implementations.
"Partial hydration isn't a feature — it's a renegotiation of where computation lives."
Editorial
The committee voted to advance, but two delegates filed objections to the exhaustiveness semantics. The next meeting will be the one to watch.
Editorial
Maglev sits between Sparkplug and Turbofan. For React's compiler output, it changes which optimisations are worth doing at the framework level.
Guest: Tao Zhang, V8 team
Release activity
12 notable releases this week. 3 carry breaking changes. The formatter wars have a new entrant, and TanStack Query quietly shipped the most significant default behaviour change in two years.
v1.9.0The first formatter that respects specificity intent — it will not reorder your declarations unless they are genuinely equivalent.
v6.0.0-beta.4BreakingThe new default behaviour cuts unnecessary re-renders by 40-60% in benchmarks. The migration guide is two paragraphs.
v3.1.2The Effect team published a Zod compatibility layer. It is not a perfect shim, but it covers 90% of real-world schema usage.
v1.43.0Two years of experimental status ends. The Vite plugin API is stable enough that Playwright now ships an official adapter.
Sentiment scan
GitHub discussion threads this week averaged 4.2× more substantive technical replies than equivalent Twitter threads. The conversation has moved. The readers who followed it are better informed.
The signal-vs-store debate resolved in a way that surprised even the core team. One comment from a Preact contributor shifted the final API.
HypeScript, type-level Tetris, compile-time SQL parsers — the people doing this aren't showing off. They're finding the edges of what inference can do.
A measured response to the perennial complaint. The fatigue is real but the diagnosis is usually wrong. What actually exhausts senior engineers.
Evergreen depth
Issues that hold up. The reader who arrives here has earned the longer view — 47 editions of the web's quieter arguments, the ones that mattered more than the ones that trended.
When a paradigm stops generating controversy it has usually won.
The patterns that emerged were not the ones the proposal authors anticipated.
The conceptual reframe that makes the mental model click.
A proposal that shaped the web without ever shipping.
The post-mortem the web platform community has not had yet.
The founding document. Still true.