zimmerframe

Summary Adds an optimized export for analysis passes that only inspect the tree without transforming it. What's different from No mutation tracking — no object, no checks, no calls No return values — visitors return , context.visit/next return No object spreads in the universal visitor path — when both and a specialized visitor exist, the context object is mutated in-place instead of creating spread copies Shared path array — single array across the entire walk instead of passing through every recursive call Types New exports: , , These mirror the existing // types but with return types, making it a type error to accidentally return replacement nodes from a read-only walk. Benchmark (in Svelte compiler) Switching 7 analysis-phase files from to in the Svelte compiler gives ~20% faster compilation: Tests 6 new tests covering: Basic traversal Universal + specialized visitor interaction State passing through universal visitor Path correctness with Path correctness with halting traversal All 17 tests pass (11 existing + 6 new). 🤖 Generated with Claude Code

Some tweaks to make traversal faster in case there's no mutations, and prevent allocations in various other places or make them slightly faster. Added a benchmark which I used to implement this. On my machine it's ~50% faster, e.g. "default read-only" goes from 11400 ops/s to 16900 ops/s

A bit more on top of #36: skip creating the per-node closures entirely when no visitor matches, hoist to walk scope, and clone mutated arrays lazily instead of the index record + . "default read-only" goes from ~13700 to ~19800 ops/s, "manual visit" from ~13800 to ~23500.