From: William Pietri <william_at_pota.to>
Date: 27 Apr 2003 18:13:01 -0700

On Sun, 2003-04-27 at 13:11, Björn Stenberg wrote:
> The problem is that these types of database building operations are
> rather heavy on both disk and, especially, memory. While it could
> probably be done in Rockbox, chances are it would take a lot of time
> and half your battery to run.
>
> An indexer can be written in simple enough C that it can easily be
> compiled for all platforms that can talk to the archos harddisks.
>
> My main argument, however, is that you already have the disk connected
> to a computer when you are changing the contents of the drive, and
> thus need to reindex it. So why not run the indexer at the same time?

Oh, I'd agree that would be optimal. I just don't believe people will do
it enough. I'd bet you can get a lot of people to do it for the initial
bulk load, but I would bet that it would end up out of sync for a lot of
people. So I think taking that portable C and, as Michael O'Quinn
suggested, making it a module, would be a good step.

But the thing that I was imagining writing to run on the Archos was
something that was pretty opportunistic. For example, it looks like
Rockbox gets the ID3 tags whenever it starts playing a new track; as
long as you have the info in hand, it should be pretty cheap to make
sure it's in the cache, too.

Of course, actual optimal points would depend on the costs of seeks,
directory reads, tag fetching, and whatnot. And, of course, the cost of
complicating the code. So until I actually get into the code and do
various timings and power draw measurements, I'm just doing a lot of
hand-waving. :-)

William

-- 
William Pietri <william_at_pota.to>

Received on 2003-04-28