Step one is building an index of the file system. This is simply done by walking the filesystem. The resulting index is stored in RAM and a file. On the next app start the index ia loaded from that file, which is much quicker than walking the file system.
Step two is using this in RAM index for searching. This scales really well with the number or CPU cores and on modern systems a normal case insensitive substring search should finish almost instantly with few million files.
The next release will support file system monitoring with inotify and fanotify to keep the index updated. Although this has some drawbacks.
This is the part I'm wondering about. Everything scans the filesystem very fast and there is no way it is just using 'stat' on every file then diving into the directories.
Are you just using stat from C to walk the filesystem or are you doing something else?
I've used sqlite to cache filesystem results and it is also extremely fast once everything is in there, but I think a lot of approaches should work once the file attributes are cached.
> Everything scans the filesystem very fast and there is no way it is just using 'stat' on every file then diving into the directories.
The last time I checked, Everything worked by using the AV calls microsoft provides; anytime a file is written, the name (and other metadata) can be written to a log that Everything can check once every 5 seconds or so.
If I thought there was any money at all to be made from providing an Everything equivalent[1] on Linux, I'd spend the week or so to write it, but as far as I can tell there's just no market for something like this.
[1] By that I mean "similar in performance and query capabilities"; I would obviously need more time than that to hook into the common file-open dialog widgets (Gnome/KDE/etc) so that users could run their queries straight from existing file dialog widgets.
What you are talking about is file change notifications. A huge part of Everything's speed comes from reading the master file table that other people mentioned, so you would need a way to quickly read file table entries on linux.
> What you are talking about is file change notifications. A huge part of Everything's speed comes from reading the master file table that other people mentioned, so you would need a way to quickly read file table entries on linux.
Not a problem. And no, I'm not talking about inotify either, and I'll additionally index the contents of (text) files as well with a negligible additional performance hit. It can be done as fast as, or faster than, `Everything`.
TBH, if I thought I could make even $100 in donations from this, I'd start it tomorrow, but absolutely no one misses ultra-fast searching when they don't have it.
Even on Windows, the number of users who go out and look for something that searches as fast as Everything is a rounding error - statistical noise. Now go and divide that fractional percentage of Everything users on Windows by 100 to get the number of Linux users who might use this.
> Not a problem. And no, I'm not talking about inotify either, and I'll additionally index the contents of (text) files as well with a negligible additional performance hit. It can be done as fast as, or faster than, `Everything`.
Please enlighten us how that would work.
> TBH, if I thought I could make even $100 in donations from this, I'd start it tomorrow, but absolutely no one misses ultra-fast searching when they don't have it.
You can easily make $100 in donations with this. I did it with this piece of software while it was still less performant and powerful and without an official release and by only mentioning it on one or two forums.
If the software delivers what you're saying, I'll guarantee you, that this will lead to more than 100$ per month in donations.
It can be done as fast as, or faster than, `Everything`.
Then how would you do it? That's what I'm asking, how would you get the file attributes off of the disk as fast as everything on linux? Once you get them off the disk any modern computer can burn through them, but getting that data into memory in the first place is the problem.
Yes, it's simply using stat on every file/folder. There's probably some room of improvement there with clever parallelization, but it'll remain a bottleneck.
Everything is parsing a file called the MFT to build its index. This much more efficient but unfortunately this file only present on NTFS volumes, which makes it super useful on Windows systems, but not so much everywhere else.
Another benefit you get on Windows is the USN journal, which allows Everything to keep the index updated much more efficiently.
I've never used fsearch, but I use a CLI tool that replaces locate (https://plocate.sesse.net/). Do you have an idea of how the performance and index format compares with fsearch?
Maybe, but I'm not sure if there's much benefit to that. The most inefficient part of the inotify or fanotify solution is that you have to walk the file system before monitoring can even start, because you first need to know which folders and files are there to begin with. And unfortunately this can't be avoided with eBPF.
Step one is building an index of the file system. This is simply done by walking the filesystem. The resulting index is stored in RAM and a file. On the next app start the index ia loaded from that file, which is much quicker than walking the file system.
Step two is using this in RAM index for searching. This scales really well with the number or CPU cores and on modern systems a normal case insensitive substring search should finish almost instantly with few million files.
The next release will support file system monitoring with inotify and fanotify to keep the index updated. Although this has some drawbacks.