> > Compression should be left to user space apps, which can be written
> > to deal with the problem in a more sane manner.
>
> Except then that every app needs to understand compressed files, and
> currently, they don't. I can't do a "grep string HOWTO/*" currently,
> and as soon as I write a script to do that, I'll run into another
> command that does't understand compressed files.
zgrep, libz.
>
> Yes, some things in the kernel will be difficult to do on compressed
> files. For mmap(), etc, it would be ok to just completely decompress
> the file first, but this involves communication between the fs and
> the kernel. For seek(), gack. Perhaps just have the calls fail.
> Howz NT do this?
It works with the current patches..
>
> I don't think there's a clean solution either in the kernel, fs, or
> user space, without doing serious overhaul.
Untrue.
>
> <elsewhere>
> > There are e2fs compression patches (which I would like to see in official
> > linux kernel some time in the future) which work relatively well.
>
> The problem with this, is that it's a full disk compression, not a per
> file, or per directory compression.
Wrong, it uses the ext2 +C attribute.. It is per file and dir.
How about maintaining a buffer of space and uncompressing files in use..