Re: [Gnu-arch-users] Some issues

[Florian Weimer]

* Andrew Suffield:

>> No, it's not.  Have a look at that mutt patch that sorts Maildir
>> access by inode number.  It's a huge win.
>
> All of which is completely irrelevant. tla almost never reads all the
> files in a directory in a project tree, and certainly it is not
> impeded by seek time when it does so; that operation is constrained by
> disk bandwidth.

In this context, a call to one of the stat() function is sufficient to
trigger reading the inode.  And this call is necessary even with inode
optimizations.

>> > since your proposal was to read a file randomly.
>> 
>> Oh, was it?  I don't think so.
>
> Yes, you were advocating reimplementing the filesystem within tla, in
> order to reduce inode consumption and "improve performance".

No, I wasn't talking about file systems.  For example, for something
as simple as the file/ID mapping, you just need two files, one with
the actual records and one listing the deleted records.  This way,
it's possible to efficiently add records, and delete a limited number
of them.  You also don't have to write the file to disk at all when
constructing intermediate revisions.

-- 
Current mail filters: many dial-up/DSL/cable modem hosts, and the
following domains: bigpond.com, di-ve.com, fuorissimo.com, hotmail.com,
jumpy.it, libero.it, netscape.net, postino.it, simplesnet.pt, spymac.com,
tiscali.co.uk, tiscali.cz, tiscali.it, voila.fr, yahoo.com.