AI summary by TwinMind
Why Is Everyone So Mad?
Theo - t3․gg
18m 14s
Server Components Misconceptions
- Common misconception: "I can already do this with X (SvelteKit, Remix, tRPC, Relay)"
- Point isn't providing new functionality that couldn't exist before
- React's goal: make it easier to do the right thing simply and maintainably
- Server components make relationship between server/client simpler and better defined
- Default benefit: faster, simpler, better applications
- Streaming features are best in class
- Makes doing the right thing easier for users and developers
- Stop conflating "more features" with "more complex"
Complexity Analysis
- useEffect example: one thing but inherently complex
- Debugging is very complex
- Hard to figure out why useEffect misfires millions of times causing outages
- New model: complexity is shallower
- More things but less depth of complexity in each
- Shorter time from problem to solution, especially debugging
- Errors may say "component cannot render" or "import doesn't work"
- Small changes between before/after make issues identifiable
- Things break immediately in new model, not later
- Edge cases make errors clear about what changed
Breaking Changes and Migration
- Heavy investment in tRPC, React Query, styled components may not work well
- CSS-in-JS doesn't work well with server rendering
- "use client" directive works well for client-side behaviors
- React Query works inside client components
- Context works if only accessed inside client components
- Can wrap app in giant context, access deep in tree from client components
- Context value: builds relationship between DOM tree point and nodes beneath
- Not for prop drilling server-side data
- Can call same function 5 places, only executes once with cache/fetch
- If using context correctly, still works in interactive "use client" layer
Angular 2 Comparison
- People comparing to Angular 2 "Red Wedding" migration
- AngularJS → Angular 2: no real interop or backwards compatibility
- Complete rewrite of core functionality
- Right call but frustrated existing developers
- Server components different: backwards/forwards compatibility story
- Example: 2014 React component (create class) rendered as server component
- Can copy/paste old code into new app
- Single "use client" at route top makes most code still behave
- React's compatibility story is best in industry
- Server components are flag you turn on, still use all old code
- Progress being made for create-react-app users
Performance and Production Readiness
- Dev experience: hot reloading not as fast as before
- Focus on Turbo Pack for performance wins
- M2 Mac: hot refresh <1 second almost always
- Initial boot takes time but not too bad
- Older machines have bad dev experience on big app router apps
- Production performance: not just good, it's great
- Speed at which React app renders output HTML
- Granular request-level caching ability
- Some requests cached, others not
- Makes apps insanely performant in production
- Order of magnitude difference in application speed
- Combined with Suspense: some content immediate, more flows in naturally
- Micro-millisecond render differences not how you feel differences in serverless deployment
App Router Stability
- Comments: "app router isn't stable, just marketing, using unstable React"
- App router stability means syntax, behavior, experience is stable
- Server actions are unstable but optional (only using on one app)
- App router vs server component ecosystem distinction
- App router goal: rethink routing in Next applications
- Succeeded even without server components
- New routing system easier for complex applications
- React version: Next massages unstable React into stable consumer API
- If building own server rendering, React instability affects you
- Next handles React behavior changes so developers don't feel differences
- React going stable shouldn't affect Next development
- Shipping app router >6 months, feels very stable now
Vercel Relationship and FUD
- Transparency: pretty close with Vercel, they sponsor channel
- They have no idea about this video, wouldn't approve half the content
- Fear: React being "taken over" by Vercel
- Before Vercel: React was just Meta/Facebook
- One small team made every decision
- Not funded as well as should have been
- Could be thrown away at any point
- Current split: ~70-80% Meta, ~10% other, ~10% Vercel
- Small percentage of core team at Vercel
- People at Meta still run most of React, make most decisions
- Sebastian Markbåge went to Vercel for better server-side execution
- Working at PHP-using company not right for server-focused vision
- Went where made most sense for React/web dev impact
- React core team wants to expand beyond Meta's intended scope
- Tried working with Shopify, Remix, others
- Found Vercel best to work with, understood vision
- Hydrogen by Shopify used server components briefly, gave up (early/hard)
- Vercel doubled down with app router commitment
Next.js and React Ecosystem
- Weird to use another framework for full React benefits
- Also weird needing webpack to ship React originally
- React always embraced external dependencies (ESLint rules, JSX transpiler, Babel)
- For server components complexity, needed framework with buy-in
- Excited for future alternatives like webpack alternatives
- ESBuild, Snowpack, SWC, Turbo Pack ecosystem formed after React/bundler value shown
- Expect new frameworks from Astro, Remix teams, new creators
- Daishi has cool stuff coming
- Next.js "dog fooding" is generous
- Without Next, would have years of figuring out small crappy meta frameworks
- People stuck on create-react-app
- Initial adoption way less scary with stable, well-maintained framework
Server Cost Reality
- Misconception: Vercel's goal is selling servers
- Reality: run way less compute with new model
- Caching much easier
- Easy to opt out of many previous calls
- Cheaper infrastructure bills using new model
- Vercel's actual goal: make full-stack web dev so easy that deployment is the slow part
- If they're best deployment option, they win
- Always been their focus
- Features don't lock in more but make best performance deployment easy
- Moving from Node to Edge: single export variable in route file
- Next exports + Vercel deployment relationship powerful
- Goal: help full-stack developers build better things faster, more effectively, less server cost
- Lower cost of everything else makes paying for deployment handling sensible
- Epic Games model: empower game developers for success
- When working at big companies, lean on Epic's tools
- Vercel wants to be who understands web dev/tooling best
- Not about ramping up server count, about ramping up application deployment effectiveness
- Unfair reputation, making web dev better while making money doing it
