Re: Problem following Zinf

On Tue, 2003-09-23 at 16:33, David Vrabel wrote:
> Hi,
> 
> > Boost.
> 
> boost::format is icky -- incorrect use of operator overloading. I 
> suggest compose instead:
> 
> http://www.cs.auc.dk/~olau/compose/
> 
> I'd also suggest neither where appropriate (e.g., debugging messages, 
> assertions).

It's useful for some user messages as well and allows for
internationalization.  Format specifiers can be different order
than the actual arguments.  The operator overloading doesn't really
bother me, and it's no worse than using << for output..

I'll take a look at compose.  It's not too late to switch.

> > mk4str.h. I don't know in what package this include file is but I will 
> > find out eventually.
> 
> Metakit database. http://www.equi4.com/metakit.html
> 
> It's not entirely clear to me why zinf even needs a database for 
> metadata.  Considering that the data is inherently flat and there isn't 
> a huge amount of it.  A trivial text or XML file would suffice.

There are a couple of reasons:

The first is that I wanted to support some new types of queries in the
musicbrowser and the metakit was an easy way to do it (See the query
routine in mdb/mkdatabase.cpp)

The second was startup time.   The gdbm database was taking along time
(>30 s for 18Gb of songs, 1MB work of metadata)  to startup everytime.  
The metakit database is instantaneous on startup.  Rythmbox had a lot
of trouble with this.. They use XML as a database format (it's an
exchange format) and startup was in the hours for me..  They've fixed
it recently (I think with a database).

In the latest metadata catalog you can still use gdbm which
loads the entire database into memory on startup.  In fact I
suggest using gdbm for small music collections or really fast/big
machines and metakit elsewhere.  I haven't tested it much, but
right now everything is moving pretty fast.


> Let's say 200 byte[1] of metadata per file, 10000 files (about 60 Gbyte 
> worth) which amounts to 2 Mbyte of metadata in total.  A not 
> unreasonable amount to keep in core.
> 
> David Vrabel
> 
> [1] A sample of 600 or so oggs I have results in approx 75 bytes of 
> metadata per file.
> 
> 

-- 
email:[email protected] office:(805)893-4276 http://www.cs.ucsb.edu/~kris



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf