You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 692498a
Browse filesBrowse the repository at this point in the historyBrowse files
Rename `followLast` to `followLink` and document it as the difference
between `stat()` and `lstat()`. Use `path.join()` for the mount name
prefixes purged on register and deregister, and explain why the hot
path in `resolveVFS()` concatenates instead.
Add a benchmark for the cost of the fs hooks on a real path and on a
layer path reached by id or by mount name. On Windows x64 (Release,
three runs of `statSync`, n=100000):
ops/s
vfs/fs-resolve.js target=real 80,929 82,063 82,618
vfs/fs-resolve.js target=id 67,669 66,850 63,834
vfs/fs-resolve.js target=name 65,911 64,155 60,767
Resolving through a mount name costs about 4% over resolving by id.
Against the merge base (dd5dfb5), with 30 interleaved runs each
and a Welch t-test, there is no significant difference:
main ops/s this ops/s change 95% CI p
target=id 67,837 67,532 -0.45% [-1.69%, +0.79%] 0.47
target=real 81,951 80,991 -1.17% [-5.91%, +3.57%] 0.62
Main has no mount names, so `target=name` has no baseline.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
0 commit comments