Conversation
…files rb_file_load_ok is called for every $LOAD_PATH candidate that exists during `require`. It did GetFileAttributesW (cheap) followed by CreateFileW + CloseHandle (~50us, two extra syscalls) to verify ACL readability. For 99% of candidates (regular .rb / .so files on a normal NTFS volume) the attribute check is enough — if access really is denied the subsequent load step fails with a precise error. The CreateFileW fallback is preserved for reparse points (FILE_ATTRIBUTE_REPARSE_POINT) so dangling symlinks / AF_UNIX sockets keep reporting "not loadable" exactly as before. Result on this VM (LoadLibrary-heavy and pure-Ruby cases): require "gtk3": ~1755 ms -> ~1721 ms (-34 ms / -1.9%) require "tzinfo": ~145 ms -> ~139 ms (-6 ms / -4.1%) require "sqlite3": unchanged
rb_file_load_ok is called ~2000 times during a heavy `require`. Each call did a malloc / free pair via mbstr_to_wstr for the UTF-8 -> UTF-16 path conversion. For typical-length paths (under MAX_PATH + slack) a stack buffer is sufficient and avoids the heap traffic entirely. For longer paths the existing mbstr_to_wstr fallback kicks in, so behavior on long paths is preserved. Measured wall-clock impact on this VM is within noise — modern Windows low-fragmentation heap makes small allocations essentially free. Kept anyway because the pattern is correct and removes ~2000 unnecessary heap allocations per gtk3 require (visible in the profile counter: mbstr_to_wstr drops from 7611 to 5545).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
rb_file_load_okis called for every$LOAD_PATHcandidate that exists duringrequire— roughly 2000 times in a heavy one. Two changes:Skip the redundant
CreateFile. The function didGetFileAttributesW(cheap) and thenCreateFileW+CloseHandle(~50 µs, two extra syscalls) just to verify ACL readability. For regular.rb/.sofiles on a normal NTFS volume the attribute check is enough; if access really is denied, the subsequent load step fails with a precise error anyway. TheCreateFileWfallback is kept for reparse points, so dangling symlinks and AF_UNIX sockets still report "not loadable" exactly as before.Stack-buffer the path conversion. Each call did a
malloc/freepair viambstr_to_wstr; typical-length paths fit a stack buffer, and longer ones still take the existing fallback.Measured on this VM:
require "gtk3"~1755 → ~1721 ms (−1.9%),require "tzinfo"~145 → ~139 ms (−4.1%). The allocation change is within noise on wall-clock — the Windows low-fragmentation heap makes small allocations nearly free — but it removes ~2000 allocations per gtk3 require (mbstr_to_wstrcalls drop 7611 → 5545).Windows-only. Not built or tested on Windows CI — measured in a local MSYS2/UCRT64 VM.