Re: extending the mdb schema format

Lukas Smith wrote: > 1) auto increment > > Currently you can only define sequences as follows: > > <sequence> > <name>foo</name> > <start>23</start> > <on> > <table>foo</table> > <field>bar</field> > </on> > </sequence> > > The <start> and <on> tags are optional. The <on> tag basically lets you > sync the sequence against a certain field in a table. Now I was > pondering extending the <on> tag to also allow for an optional > "<autoincrement>1</autoincrement>". This format is in the sprit of the > rest of the tags in that it doesnt use atributes. > > Now the problem is that we may need a separate syntax for "I only want > autoincrement!" and "Use autoincrement if available." as MDB2 supports a > nice mode where either auto increment or sequences are used. I dont > really like something like "<autoincrement>force</autoincrement>" for > this case. > > I could also add something like this > > <autoincrement> > <table>foo</table> > <field>bar</field> > <start>23</start> > <force>1</force> > </autoincrement> > > This is also not very nice and seriously screws with how the current > code works and the current DTD is structured. Why not go with the dual approach? An autoincrement is not an entity in the RDBMS, like sequences are. It is a behaviour of a field, so define it in the field: <table> <name>foo</name> <declaration> <field> <name>bar</name> <type>integer</type> <default> <autoincrement> <start>1</start> <increment>1</increment> </autoincrement> </default> </field> </declaration> </table> This way, if a RDBMS does not have autoincrement fields, it may behave like Postgresql: create a sequence behind the scenes, and assign sequence::next_val to the field's default value.