turbo

8e431052 - perf: Reuse input hashes and trim pnpm lockfile allocations (#12199)

perf: Reuse input hashes and trim pnpm lockfile allocations (#12199) ## Summary - reuse repo-index blob OIDs for clean tracked files matched by explicit `inputs`, including parent-path matches, instead of sending them back through `hash_objects` - reuse the same repo-index split for explicit include literals/globs in the mixed default-plus-input path - reduce pnpm lockfile parser allocator churn by preallocating the dependency index and combined snapshot dependency maps - document the explicit-input repo-index reuse in `ARCHITECTURE.md` ## Verification - `cargo fmt --all` - `cargo test -p turborepo-scm package_deps::tests` - `cargo test -p turborepo-scm repo_index::tests` - `cargo test -p turborepo-scm git_index_regression_tests::test_inputs_without_defaults_match_no_index_for_tracked_and_parent_files` - `cargo test -p turborepo-lockfiles pnpm` - `cargo test -p turborepo-repository package_manager` - `cargo check -p turborepo-lockfiles -p turborepo-repository -p turborepo-scm -p turborepo-task-hash -p turborepo-lib` - `cargo build --profile release-turborepo -p turbo` ## Profiling Release benchmark command: `target/release-turborepo/turbo run build --filter=@turbo/cli --skip-infer --force --profile=... --ui=stream` Because the child `cargo build --package turbo` hits a nightly incremental compiler ICE on this machine, the warm post-change benchmark runs used `CARGO_INCREMENTAL=0` to keep the task execution path stable. The Turbo-owned hashing and lockfile work measured here happens before the child build and is unaffected by that setting. Saved baseline trace before this PR: - duration: `934.0ms` - `hash_scope`: `13.2ms` - `get_package_file_hashes_from_inputs`: `13.0ms` total - `hash_objects`: `4.5ms` - `parse_lockfile`: `18.6ms` Post-change warm traces: - `pr3-after-lockfile-warm.trace.md`: duration `815.0ms`, `get_package_file_hashes_from_inputs` `9.3ms`, `parse_lockfile` `14.6ms` - `pr3-after-lockfile-warm-2.trace.md`: duration `766.5ms`, `get_package_file_hashes_from_inputs` `7.9ms`, `parse_lockfile` `16.7ms` - `hash_objects` no longer appears in the warm hotspot table after the explicit-input hashing change