> >If you have a filesystem that forces NF-D, then I would say its a > >poorly designed filesystem that makes such choices, because its > >way to low level to care about things like that. Filenames should > >be "string of bytes", > > Filenames are _names_. They are names for _human_ use; computers would > be just as happy passing about inode numbers. Humans don't like > dealing with strings of bytes. They like dealing with strings of > characters, and you'll note that almost every filename created by > humans make sense as strings of characters - in fact, even filenames > created by the computer tend to be strings of graphic characters from > the ISO-646-INVARIANT set.
I think we both agree that different file names should look different. The question is not whether normalization should be done, but where. My argument was that it should not be done inside the kernel, filesystem, compiler, linker, etc. But instead it should be dealt with at the Input Method, and user interface level. (Normalized strings would always be generated by user input, and non-normal strings would be displayed as escape sequences.) Do you think that normalization should be built into the filesystem, so you can ask it for the file in any form, and it normalizes it internally before looking it up? -- Linux-UTF8: i18n of Linux on all levels Archive: http://mail.nl.linux.org/linux-utf8/
