next.js

753e102c - turbopack: use reverse postorder priority in collect_graph (#100429)

turbopack: use reverse postorder priority in collect_graph (#100429) ## Summary `collect_graph` computes per-module entry membership with the same fixed-point traversal and BFS-depth priority as `compute_chunk_group_info`. This PR switches it to the reverse-postorder rank added in the PR below. That removes its own BFS depth pass, and a module is now visited only after all of its importers (except across cycles). `TraversalPriority` has no remaining users, so it is removed. Stacked on #100428, which introduces `ModuleGraphSnapshot::reverse_postorder_ranks`. ### Benchmarks **Setup:** - Same as the PR below: Turbopack `next build --experimental-build-mode compile`. - `next-swc` release builds (LTO disabled, 16 codegen units). - Binaries interleaved per run, 3 runs each, clean `.next`. - Time is the depth or rank pass plus the fixed-point traversal, measured with local-only instrumentation. Values are medians with min–max ranges. - "Before" is the PR below, which already includes its entry deduplication. | app | before | after | |---|---|---| | generated app router app (300 routes, ~12k modules) | 14.5 ms (14.4–16.5), 12,030 visits | **9.5 ms** (8.3–11.0), 11,938 visits | | `bench/nested-deps-app-router-many-pages` (~31k modules) | 422 ms (417–426), 30,869 visits | **197 ms** (193–205), 30,777 visits | For reference, `collect_graph` on many-pages takes 534 ms (496–546) on canary. ## Verification - `cargo test -p turbopack-core --lib` - `cargo test -p turbopack-tests --test snapshot`: passes with no snapshot changes. - `cargo test -p turbopack-tests --test execution -- collect` (emit/collect execution tests). <!-- NEXT_JS_LLM --> <!-- fleet b56269ed-2145-4f30-b0fe-dc9762520521 --> Co-authored-by: vercel-fleet-prod[bot] <318278635+vercel-fleet-prod[bot]@users.noreply.github.com> Co-authored-by: Tobias Koppers <1365881+sokra@users.noreply.github.com>