Re: [I18ndir] I-D on filesystem I18N

[Nico Williams <[e-mail]>]

On Wed, Jul 08, 2020 at 04:38:11PM -0400, John Levine wrote:
> In article <9044C737C36C0787B9EAE190@PSB> you write:
> >Well, Unix, and every Unix-derived system I know of (definitely
> >including Linux, FreeBSD, and NetBSD) are case-sensitive and
> >getting anywhere near their file names with Case Folding or even
> >lower casing will cause rather interesting problems. 
> 
> FWIW MacOS is Unix underneath and its native file system is case
> folding.  I've been surprised how little trouble it's caused when
> I'm moving code back and forth to FreeBSD and even NFS mounting
> my MacOS filesystem on FreeBSD virtual machines.
> 
> The point about ZFS is also a good one, since it's gotten very popular
> on BSD and Solaris due to its flexible disk management. (Less so on
> most linux distros due to poor kernel integration.) It gives you a
> per-filesystem choice of case sensitifity, and also optionally doing
> any of the four Unicode normalizations on filenames.

Yeah.

In ZFS the NF is about form-insensitivity, so there's no observable
difference between picking NFC and NFD, or NFKC and NFKD.

BTW, I found this thread where Git users on OS X using ZFS were bitten
by normalization to NFD on create.  It took some doing to find it!

http://public-inbox.org/git/[e-mail]/

and here Linus blames ZFS (though others correctly blame Apple)

http://public-inbox.org/git/[e-mail]/

saying that

| UFS is a traditional Unix filesystem, and will not mangle your filenames.
| 
| ZFS apparently acts like a case-sensitive HFS+: it still tries to 
| normalize to UTF-8 (and does it badly, at that - picking an Apple-specific 
| almost-NFD form of normalization rather than the more sensible NFC form).
| 
| So ZFS may not corrupt cases, but it still corrupts UTF-8 filenames.

But again, the fault was Apple's: they chose to normalize to NFD on
create.

Normalizing to NFC would have been better because that would have
aligned better with input modes' output.  Input modes don't necessarily
produce NFC though.  This is why form-insensitive and form-preserving
behavior is a better choice.

As it is, important apps with squeaky users get the normalization grease
(rsync, Git, as mentioned).  But if you're using ZFS on anything other
than OS X, no apps need any grease.

Nico
--