We spent two weeks arguing about Taro versus uni-app versus just writing WXML by hand. Picked Taro, shipped on WeChat, and quietly congratulated ourselves on being cross-platform. Eight months later a second host came along and we rewrote login, payments, subscription messages, the review checklist and half our analytics. The UI compiled fine. Everything that made money didn't.
That's the part the framework comparison tables tend to leave out.
What a compiler actually reuses
Taro and uni-app are good at what they do: let you write React or Vue once and emit the markup, lifecycle and component tree each host expects. For a team that would otherwise maintain three copies of every screen, that's a real saving. Styles, conditionals, list rendering — all handled.
What a compiler can't do is invent an API the host doesn't have. Login is different. Payment intents are different. Subscription messages have different consent models. Review policies, package limits and update mechanisms are different. Every one of those sits below the view layer, which means every one of them is still yours to adapt.
| Approach | What gets reused | Where it first breaks | Good fit |
|---|---|---|---|
| Plain WXML + host SDK | Nothing, but nothing surprises you | The second host, immediately | One host, small team |
| Taro / uni-app | Components, styles, view logic | Login, payments, notifications | Many screens, one host today |
| Home-grown adapters | Whatever you bothered to abstract | Maintenance, quietly, forever | Teams with platform engineers to spare |
| Shared runtime | Identity, payments, review, updates | You give up host-specific polish | Several hosts, several markets |
The lowest common denominator trap
Compile-to-everywhere frameworks have an awkward habit: the moment a feature exists on only one host, you either hide it or start writing platform branches. Six months in, the codebase is full of if (host === ...) and the "write once" promise is a comment nobody believes. It's not a bug in the framework. It's what happens when the abstraction sits at the view layer instead of the platform layer.
We'd still use Taro again for a UI-heavy product. But we'd stop calling it cross-platform and start calling it what it is — a very good way to avoid copying screens.
The honest recommendation
If you're building for one host, skip all of this. Use the native SDK, ship, and don't add an abstraction you won't collect on. We've watched small teams burn a quarter building a portability layer for a second market that never launched.
The calculus changes when there really are several hosts, or several regional super apps you don't control. Then the question isn't which framework compiles best. It's which layer you refuse to rebuild. Identity, payments, review and updates are the ones that cost real weeks and carry real compliance risk; view code is the cheapest thing to regenerate. Put the abstraction where the cost is.
That's the appeal of a shared runtime: the host-specific parts live in one place, and the same mini-program can target a messaging app in one market and a payments app in another without rewriting the platform layer. It's slower on day one and much cheaper on day three hundred.
Sources
- Taro documentation — the compile-to-multi-host model and its platform differences.
- uni-app documentation — Vue-based multi-end output and conditional compilation.
- WeChat Mini Program framework docs — login, payment and subscription APIs.
- Alipay Mini Program docs — a concrete example of divergent platform capabilities.
- MDN — Web Components — why view-layer portability is the easier half of the problem.