PATH should be a system call | Hacker News Reader

PATH should be a system call

simonsafar.com

18 pointstodsacerdoti15 comments

Open articleView on HN

Make it happen? Doesn't sound very hard to code up. Then have bash call execsomewhere() and benchmark the results.

Challenge accepted!

https://news.ycombinator.com/item?id=43808428

Like this?

https://learn.microsoft.com/en-us/windows/win32/api/shlwapi/...

That's not a single system call, it's just a userspace library function built on GetEnvironmentVariable() followed by a GetFileAttributes() loop, alongside a few other special cases for compatibility.

For what it is worth, I did actually try to tackle the problem after reading this. So thank you!

https://news.ycombinator.com/item?id=43808428

https://www.man7.org/linux/man-pages/man3/io_uring_prep_stat... is there when you need it.

> the latter [Windows] is somewhat known of its less than excellent performance when it comes to small files.

Not at all a Windows guy here, but my understanding was that NTFS kept the contents of small files in the FILE record itself. How does this cause low performance?

Looking up and opening a file is infamously slow on Windows, due to the stack of drivers, compatibility layers, etc. it must go through. So performant Windows programs typically use fewer larger files instead of the many smaller files seen in Unix-like programming.

This is just for the bash case, but I was under the impression that shells cache and hash the binaries they find in the path to avoid doing these lookups. Surprised by the need for strace showing lookups.

The caching is per session so starting up a new session will have an empty cache. To verify:

    strace bash -c 'ls;ls'

You should not see the lookups for the second ls invocation.

The hash is probably not very relevant in the age of plentiful RAM & NVMe. Mine is disabled.

Maybe filesystems should be object stores with metadata in a database.

https://en.m.wikipedia.org/wiki/WinFS

    > Windows Vista, then called by its codename "Longhorn", given to developers at the Microsoft Professional Developers Conference in 2003, included WinFS, but it suffered from significant performance issues

Odd, I did tried out a few Longhorn builds. Not knowing if they were shipped with WinFS, but in my case the "performance issues" of the system was caused by increased RAM demand (I only had 512MB RAM at the time).

The referenced sourced by Paul Thurrott from winsupersite.com mentioned:

    > (https://web.archive.org/web/20070702131752/http://www.winsupersite.com/showcase/winfs_preview.asp)
    >
    > it was pretty clear that WinFS wasn't ready for prime time. As one might expect, WinFS suffered from huge performance issues and simply bogged down the regular builds. And while WinFS was indeed included with the WinHEC 2004 Longhorn build that Microsoft shipped in May 2004, Microsoft was surprisingly quiet about WinFS at that time. A few months later, we found out why.

One old discussion on the subject suggested:

    > (https://ask.metafilter.com/129685/Why-did-WinFS-fail)
    >
    > posted by @troy at 3:03 PM on August 9, 2009:
    > I read that as "slow as a wee lassie on anything less than 16GB and quad dual-cores". They're waiting for PCs to be fast enough.
    >
    > posted by Ptrin at 7:13 PM on August 10, 2009:
    > Because Longhorn was cancelled. The WinFS project was a part of Longhorn, and when Longhorn died, WinFS did as well

But the exact cause for the issues remain undisclosed. Don't you just like these close sourced hypes? LOL

I think WinFS failed for the same reason the Cairo Object filesystem before it did. Microsoft required WinFS to use their SQL server rather than implementing the limited structures directly in the filesystem.

Reply on news.ycombinator.com