Re: setting metadata (emblems) from console ?

[Maciej Katafiasz <mnews22 wp pl>]

Dnia 07-10-2004, czw o godzinie 18:54 +0200, Danny Milosavljevic
napisał:
> > > > gnome_vfs_get_file_info() -> containing metadata as name->value
> > > > hashtable ? Or something more sophisticated ? 
> > > 
> > > The problem is not to just expose the current API. We want to completely
> > > redo it so it doesn't depend on nautilus, gets good performance, is
> > > race-free, can be backed by EA filesystem support etc.
> > > 
> > > Basically, it requires a lot of thinking and designing.
> > 
> > The EA support is a bugbear.  Not every filesystem a user interacts with
> > supports EAs: SMB, NFS, FAT, etc....  Data like icons and emblems are
> > naturally EAs.  
> 
> How about if the filesystem doesnt support them, we don't either? I
> don't see any serious disadvantage of that, but it sure gives incentive
> to add EA support to said filesystems (which is the right thing to do
> imo, rather than emulating). 
> [or, if impossible, move the EA emulation into the kernel; I'm sure for
> that comment I will get added to the horrible-death-list of some kernel
> people now ;)]
> 
> On a side note, actually conceptually I like the reiserfs approach even
> better than EA api, since its more logical and (ideally) doesn't require
> any api extensions: 

Just to disagree here - yes it would require API extensions, since
someone would need to know that metadata is in fact held in subdir, how
to distinguish real subdir from metadata one, how to distinguish proper
directories from files (since both can now have subdirs), etc. That's
why abstractions that try too hard to be neat & simple usually end up
being more or less broken - instead of simplifying things by enabling
legacy clients to do new tricks without changes, effect is legacy
clients dying horrible death due to subtle changes in semantics, and new
clients going through hoops to do new thing since API were artificially
twisted in order to "support" old clients. 

(Note this isn't specifically about reiserv4, or meant to disrespect it,
I haven't seen it in action, but from what I heard about new
possibilities of it, and particularly metadata as subdirs, this seems at
least partially like case of "neatolitis" described above)

> Metadata goes to a "subdirectory" of the file (not
> to a subdirectory of the directory of the file, but literally to a
> subdirectory of the file), like that,
> 
> $ cat /home/dannym/rule-the-world.txt/emblem.important
> 1
> 
> And it would have mtime and permissions per attribute entry too, without
> any extra hassle.
> 
> (obviously for attributes of directories that wouldnt work quite the
> same ;))

Yeah, that's what I mean. ;P

Cheers,
Maciej

-- 
"Tautologizm to coś tautologicznego"
   Maciej Katafiasz <mnews2 wp pl>
         http://mathrick.org