Обсуждение: Re: [HACKERS] [GENERAL] Large databases, performance

> if i'm not mistaken, a char(n)/varchar(n) column is stored as a 32-bit
> integer specifying the length followed by as many characters as the
> length tells.  On 32-bit Intel hardware this structure is aligned on a
> 4-byte boundary.

Yes.

> | opc0     char (3)   no    no       8     4
> | opc1     char (3)   no    no       8     4
> | opc2     char (3)   no    no       8     4

> Hackers, do you think it's possible to hack together a quick and dirty
> patch, so that string length is represented by one byte?  IOW can a
> database be built that doesn't contain any char/varchar/text value
> longer than 255 characters in the catalog?

Since he is only using fixchar how about doing a fixchar implemetation, that
does not store length at all ? It is the same for every row anyways !

Andreas

On Mon, Oct 07, 2002 at 05:42:12PM +0200, Zeugswetter Andreas SB SD wrote:
> > Hackers, do you think it's possible to hack together a quick and dirty
> > patch, so that string length is represented by one byte?  IOW can a
> > database be built that doesn't contain any char/varchar/text value
> > longer than 255 characters in the catalog?
>
> Since he is only using fixchar how about doing a fixchar implemetation, that
> does not store length at all ? It is the same for every row anyways !

Remember that in Unicode, 1 char != 1 byte. In fact, any encoding that's not
Latin will have a problem. I guess you could put a warning on it: not for
use for asian character sets. So what do you do if someone tries to insert
such a string anyway?

Perhaps a better approach is to vary the number of bytes used for the
length. So one byte for lengths < 64, two bytes for lengths < 16384.
Unfortunatly, two bits in the length are already used (IIRC) for other
things making it a bit more tricky.
--
Martijn van Oosterhout   <kleptog@svana.org>   http://svana.org/kleptog/
> There are 10 kinds of people in the world, those that can do binary
> arithmetic and those that can't.