oxc-project/oxc-resolver

oxc-project

oxc-resolver

Blog Docs Changelog

Performance History

Latest Results

feat: record tsconfig files loaded through the `extends` chain

shulaoda:07-14-feat_record_tsconfig_files_loaded_through_the_extends_chain

3 hours ago

chore(deps): update dependency @napi-rs/cli to v3.7.3

renovate/napi

20 hours ago

main

1 day ago

chore(deps): update napi

renovate/napi

1 day ago

perf: store canonicalization state only where it differs from the path Canonicalization cached a `(Weak, Box<Path>)` result on every path level and interned the whole canonical-side chain alongside the logical one. For a single pnpm package both `node_modules/pkg/*` and `node_modules/.pnpm/pkg@x/node_modules/pkg/*` were fully interned, and every node of both chains stored a canonical mapping — on a path that is its own canonical form (the large majority) that is a self-referential `Weak` plus a byte-copy of the node's own spelling. Keep canonicalization state only where it is not redundant: - A path that is its own canonical form (no ancestor is a symlink — the large majority) carries a one-bit `canonical_is_self` flag and no heap. - A symlink stores its resolved target (live `Weak`, used by `followed_metadata`); a path below a symlink caches its derived canonical form so re-resolving it stays O(1). `Cache::canonicalize` scans the cached parent chain bottom-up to the nearest node whose canonical form is known, verifies via the cached `lstat` bits that the suffix contains no symlink, then folds the suffix onto the prefix in one allocation. Measured over the bench-pm workload (16 requests, 8 layouts): - Cache entries carrying a canonical mapping: 50-71 -> 5-21 per combo. - Path key + mapping heap bytes: -40% per combo. - Cold allocations: -6 to -9% (npm-flat 1086->1020, pnpm-isolated 1211->1127). - Filesystem syscalls: byte-identical (per-op-type and total). - Warm resolution: at parity (allocations identical; wall time within noise on resolver_memory / resolver_real single-thread and resolve-from-symlinks). Adds tests pinning that self-canonical paths store no mapping and that dirty spellings inside node_modules canonicalize correctly. Closes #852

perf/canonicalize-symlink-mappings

2 days ago

perf: store canonicalization state only where it differs from the path Canonicalization cached a `(Weak, Box<Path>)` result on every path level and interned the whole canonical-side chain alongside the logical one. For a single pnpm package both `node_modules/pkg/*` and `node_modules/.pnpm/pkg@x/node_modules/pkg/*` were fully interned, and every node of both chains stored a canonical mapping — on a path that is its own canonical form (the large majority) that is a self-referential `Weak` plus a byte-copy of the node's own spelling. Keep canonicalization state only where it is not redundant: - A path that is its own canonical form (no ancestor is a symlink) carries a one-bit `canonical_is_self` flag and no heap. - A symlink stores its resolved target (live `Weak`, used by `followed_metadata`); a path below a symlink caches its derived canonical form (dangling `Weak`) so re-resolving it stays O(1). `Cache::canonicalize` scans the cached parent chain bottom-up to the nearest node whose canonical form is known, verifying via the cached `lstat` bits that the suffix contains no symlink, then folds the suffix onto the prefix in one allocation. Measured over the bench-pm workload (16 requests, 8 layouts): - Cache entries carrying a canonical mapping: 50-71 -> 5-21 per combo. - Path key + mapping heap bytes: -40% per combo. - Cold allocations: -6 to -9% (npm-flat 1086->1016, pnpm-isolated 1211->1113). - Filesystem syscalls: byte-identical (per-op-type and total). - Warm resolution: at parity (allocations identical; wall time within noise on resolver_memory / resolver_real single-thread and resolve-from-symlinks). Adds a `pm-dense` benchmark (one cold resolver over every top-level `lodash-es/<file>.js`, serial) covering the dense regime the 16-request workload dilutes, and tests pinning that self-canonical paths store no mapping and that dirty spellings inside node_modules canonicalize correctly. Closes #852

perf/canonicalize-symlink-mappings

2 days ago

perf: store canonicalization state only where it differs from the path Canonicalization cached a `(Weak, Box<Path>)` result on every path level and interned the whole canonical-side chain alongside the logical one. For a single pnpm package both `node_modules/pkg/*` and `node_modules/.pnpm/pkg@x/node_modules/pkg/*` were fully interned, and every node of both chains stored a canonical mapping — on a path that is its own canonical form (the large majority) that is a self-referential `Weak` plus a byte-copy of the node's own spelling. Keep canonicalization state only where it is not redundant: - A path that is its own canonical form (no ancestor is a symlink) carries a one-bit `canonical_is_self` flag and no heap. - A symlink stores its resolved target (live `Weak`, used by `followed_metadata`); a path below a symlink caches its derived canonical form (dangling `Weak`) so re-resolving it stays O(1). `Cache::canonicalize` scans the cached parent chain bottom-up to the nearest node whose canonical form is known, verifying via the cached `lstat` bits that the suffix contains no symlink, then folds the suffix onto the prefix in one allocation. Measured over the bench-pm workload (16 requests, 8 layouts): - Cache entries carrying a canonical mapping: 50-71 -> 5-21 per combo. - Path key + mapping heap bytes: -40% per combo. - Cold allocations: -6 to -9% (npm-flat 1086->1016, pnpm-isolated 1211->1113). - Filesystem syscalls: byte-identical (per-op-type and total). - Warm resolution: at parity (allocations identical; wall time within noise on resolver_memory / resolver_real single-thread and resolve-from-symlinks). Adds a `pm-dense` benchmark (one cold resolver over every top-level `lodash-es/<file>.js`, serial) covering the dense regime the 16-request workload dilutes, and tests pinning that self-canonical paths store no mapping and that dirty spellings inside node_modules canonicalize correctly. Closes #852

perf/canonicalize-symlink-mappings

2 days ago

chore(deps): update rust crates (#1306) This PR contains the following updates: | Package | Type | Update | Change | |---|---|---|---| | [memchr](https://redirect.github.com/BurntSushi/memchr) | dependencies | patch | `2.8.2` → `2.8.3` | | [simd-json](https://crates.io/crates/simd-json) | dependencies | patch | `0.17.0` → `0.17.3` | --- ### Release Notes <details> <summary>BurntSushi/memchr (memchr)</summary> ### [`v2.8.3`](https://redirect.github.com/BurntSushi/memchr/compare/2.8.2...2.8.3) [Compare Source](https://redirect.github.com/BurntSushi/memchr/compare/2.8.2...2.8.3) </details> --- ### Configuration 📅 **Schedule**: (in timezone Asia/Shanghai) - Branch creation - "before 10am on monday" - Automerge - At any time (no schedule defined) 🚦 **Automerge**: Enabled. ♻ **Rebasing**: Whenever PR is behind base branch, or you tick the rebase/retry checkbox. 👻 **Immortal**: This PR will be recreated if closed unmerged. Get [config help](https://redirect.github.com/renovatebot/renovate/discussions) if that's undesired. --- - [ ] If you want to rebase/retry this PR, check this box --- This PR was generated by [Mend Renovate](https://mend.io/renovate/). View the [repository job log](https://developer.mend.io/github/oxc-project/oxc-resolver).  Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

main

2 days ago

Latest Branches

+2%

feat: record tsconfig files loaded through the `extends` chain#1309

3 hours ago

8a987ca

shulaoda:07-14-feat_record_tsconfig_files_loaded_through_the_extends_chain

+5%

chore(deps): update dependency @napi-rs/cli to v3.7.3#1308

20 hours ago

f02df2f

renovate/napi

+7%

perf: store canonicalization state only where it differs from the path#1298

2 days ago

a372fb0

perf/canonicalize-symlink-mappings

Home Terms Privacy Docs