On Tue, Aug 10, 1999 at 12:58:08PM +0300, Jaromir Dolecek wrote:
> Ignatios Souvatzis wrote:
> > You can't do this (case-insensitiveness) file system independent.
> 
> Yeah, I know. I'm not talking about making the search FS-I,
> just thinking about how to provide a way for fs's code
> to do case-independant search in the cache.
> 
> The problem is, that the hash value is counted before the fs code is
> able to touch the name to be searched for. If the name passed
> to lookup() is not the same as the the canonified one, we end up
> looking into wrong hash bucket in cache_lookup() and it's very
> likely the entry won't be found.
> 
> If we would leave the counting of cnp->cn_hash in lookup() (as it's
> currently), we have to provide a hook for fs code
> to supply "canonified" name for the name we are currently doing
> the lookup for or the fs code would be forced to recount
> the hash value before calling cache_lookup(). I like neither of
> those two possibilities.
> 
> If the hash value would be counted in cache_lookup() directly,
> fs code can do whatever it wants to canonify the filename
> and the hash value would be counted for the right name. Furthermore,
> the hash will be counted only when it would actually be used -- not
> such a bit optimization, but every bit counts :)

I am not familiar with the code you're proposing to change...

you sure what you want to do works for filesystems that are

- case insensitive but
- case preserving?

(that is, you need to store the original data on the disk, and provide them
 to the reader, but use the canonicalized for searching operations).

(ADOSfs and  OS-9 come to my mind.)

Regards,
	-is
-- 
 * Progress (n.): The process through which Usenet has evolved from
   smart people in front of dumb terminals to dumb people in front of
   smart terminals.  -- obs@burnout.demon.co.uk (obscurity)