Re: [gnome-db] Data Types translations.

[Bas Driessen <bas driessen xobas com>]

On Fri, 2006-09-01 at 09:15 +0200, Vivien Malerba wrote:
> On 9/1/06, Bas Driessen <bas driessen xobas com> wrote:
> >
> >  On Wed, 2006-08-30 at 17:15 +0200, Vivien Malerba wrote:
> >
> >  On 8/30/06, Bas Driessen <bas driessen xobas com> wrote:
> > >
> > > On Wed, 2006-08-30 at 15:36 +0200, Vivien Malerba wrote:
> > > On 8/30/06, Bas Driessen <bas driessen xobas com> wrote:
> > > [...]
> > >
> > >
> > >
> > > >
> > > > Further to the discussion regarding the creation of the table. There is
> > > the > Path: /FIELDS_A/@COLUMN_TYPE. Here a complete type must be set. For
> > >
> > > postgresql that is varchar(30) for instance. As mentioned before I would >
> > > like to see this part more provider independent. Perhaps we can reach some
> > >
> > > sort of a compromise. First I tell you what I would like to see: > >
> > > /FIELDS_A/@COLUMN_GTYPE > /FIELDS_A/@COLUMN_SIZE > /FIELDS_A/@COLUMN_SCALE
> > >
> > > > example values: > GDA_TYPE_NUMERIC > 10 > 2 > > This then will be
> > > translated by the postgresql provider for instance to > numeric(10,2). > >
> > > To provide both options to the user, we can say that if >
> > > /FIELDS_A/@COLUMN_TYPE (the current situation), this overrules any >
> > > situation. If it is not set, the user should set > >
> > /FIELDS_A/@COLUMN_GTYPE
> > > > /FIELDS_A/@COLUMN_SIZE (if applicable) > /FIELDS_A/@COLUMN_SCALE (if
> > > applicable) > > Then for the various providers we can make this into
> > > something that works > for the data provider. > I agree on having the
> > > /FIELDS_A/@COLUMN_SIZE and /FIELDS_A/@COLUMN_SCALE as this makes things
> > more
> > > provider independant. However I prefer to stay with a
> > /FIELDS_A/@COLUMN_TYPE
> > > because it's closer to what a user can easily fill-in and it's easy for a
> > > programmer (using
> > > gda_server_provider_get_default_dbms_type()) to convert
> > > from a GType to a DBMS type. It also makes it possible to check and take
> > > actions when the provider doesn't support a particular GType before trying
> > > to execute the action. So I propose to have /FIELDS_A/@COLUMN_TYPE
> > > (required) /FIELDS_A/@COLUMN_SIZE (optional) /FIELDS_A/@COLUMN_SCALE
> > > (optional) Is it Ok for you? If so, can you make the modifications in the
> > > xml.in files and in the -ddl.c files for some or all of the providers?
> > > Ok with me, but if you bring back everything to just COLUMN_TYPE (as
> > > opposed to also have COLUMN_GTYPE), how would you deal with the situation
> > if
> > > a user sets the values as follows:
> > >
> > > numeric(10,5) (rather than GDA_TYPE_NUMERIC)
> > > 10
> > > 5
> > >
> > > In this case we should ignore the size and the scale, since
> > > numeric(10,5)(10,5) is rubbish of course. What is the best approach to
> > deal
> > > with this you reckon?
> >
> > This is a potential problem and whatever we end up doing, the user
> > will always have the responsibility to avoid entering stupid things
> > (my point is that we can't make something completely water-proof).
> >
> > So I suppose in this case we should just notice that there are
> > parenthesis and then ignore the /FIELDS_A/@COLUMN_SIZE and
> > /FIELDS_A/@COLUMN_SCALE values, just as we should ignore them when the
> > data type does not support size and scale attributes.
> >
> > >
> > > I am more than happy to make the changes for the postgresql and mysql
> > > providers. Once I have everything stable/working again, I want/have to
> > look
> > > into the Oracle provider as well. I need to bring that to the same level
> > in
> > > libgda as mysql and postgresql (including recordset etc etc).
> >
> > Ok, great!
> >
> > BTW: do you have/want a gmail account so it is possible to do some chat?
> >
> >
> >  OK Vivien will start working on the SIZE/SCALE this weekend.
>
> Ok, I think I'll roll out a beta this WE, and I'll merge tour
> modifications after it.

I have made the modifications for the Postgresql provider. Tested OK and committed to CVS. Please review/test and let me know any issues. Will work on the MySQL provider next.

It is pretty impossible to make this optional SIZE and SCALE water-proof as discussed before. I decided not to check for parenthesis etc at all, so the user will see in their error output what goes wrong. If we start making assumptions that if a user typed a parenthesis we should not add ours etc, we may get burned at some stage with some special type in some special situation.

Thanks,
Bas.