Performance

Pre-Release

Verter is pre-release software. APIs may change between releases — see the API Stability document.

Verter's compiler is written in Rust and exposed to Node.js through NAPI-RS. The project treats performance as a measured release property, not a static marketing claim.

Interpreting Results

Benchmark results are valid only for the recorded source revisions, dependency versions, build profiles, platform, corpus, cache state, and work performed. The repository therefore does not publish a timeless "average speedup" table. Use the machine-readable artifact from a specific run when evaluating a change.

For a comparative result to support a release statement, it must:

• identify immutable Verter and comparison-compiler revisions;
• perform equivalent work, including source-map generation when maps are part of either measured path;
• validate behavior or output contracts before timing;
• attest cold, warm, and stateless cache behavior rather than assuming it;
• isolate workers and alternate ordering to limit cross-backend bias;
• use process RSS, not only JavaScript heap, for native-versus-JavaScript memory comparisons; and
• publish raw samples and corpus metadata, not only a ratio.

For Svelte, the CI rail pairs separate conformance and official-oracle behavior tests with a timed fence that validates clean mapped output, immutable identity, fresh stateless compilation, and isolated-process peak RSS. It is a regression rail for that explicit corpus, not evidence that Verter is universally faster or uses less memory than the official Svelte compiler.

Benchmark Families

• Compiler fixtures measure Vue compilation across a fixed set of component shapes. They are useful for local regressions but do not represent every Vue application.
• Svelte equal-work fence compares the experimental Verter client compiler with the pinned official compiler using source maps, behavioral validation, fresh stateless compilation, isolated workers, and peak RSS.
• Repository performance gate measures first-pass and warm-pass workloads with explicit corpus and statistical contracts.
• Component metadata and editor benchmarks measure their own public paths; their numbers must not be generalized to compiler throughput.

Running Locally

Run Rust microbenchmarks with:

cargo bench --package verter_bench

Run the JavaScript benchmark package with:

pnpm --filter @verter/benchmark bench
pnpm --filter @verter/benchmark bench:perf

The Svelte comparison requires a clean worktree at a full Git revision because the result records and verifies that identity:

pnpm --filter @verter/benchmark bench:svelte:compiler

See the benchmark package for fixture definitions, additional workloads, and result schemas.

Pre-Compilation for Faster Builds

Beyond raw compiler speed, Verter's @verter/unplugin supports a preCompile option that compiles all .vue files during buildStart before bundling begins. This front-loads compilation work and caches results, so subsequent transform() calls return instantly:

// vite.config.ts
import { defineConfig } from "vite";
import VerterVite from "@verter/unplugin/vite";

export default defineConfig({
  plugins: [
    VerterVite({
      preCompile: true,
    }),
  ],
});

See Cross-File Optimization for how preCompile enables whole-program analysis.

Next Steps

• Features -- Type safety features overview
• Cross-File Optimization -- Whole-program prop constness analysis
• Architecture -- How the compiler pipeline works