On Monday 14 July 2003 13:46, Christoph Haller wrote:
> > Am currently writing (or trying to at least) a postgresql driver for
>
> the
>
> > koffice app kexi.
> >
> > The driver requires that i subclass some kexi classes, one being a
>
> class that
>
> > represents a field.
> >
> > I know the SQL statement used to generate the result and for each
>
> field inthe
>
> > result i create an object. However, one property of the object is the
>
> table
>
> > name for the field. Other properties require that i know the table
>
> name, eg
>
> > given the table and field names i can query the system catalogs to get
>
> things
>
> > like contraints.
> >
> > In the my_sql c library, the result (tuble) structure contains a filed
> >
> > structure that contains the table name that the particular field
>
> originated
>
> > in, but libpq doesnt seem to have any similar kind of functionality
>
> (and i
>
> > have looked a lot i think)
> >
> > Has anyone any suggestions on how i can achieve getting the table name
>
> for a
>
> > filed in a result without having to parse the sql statement (which i
>
> dont
>
> > think wold be easy).
>
> I very much doubt there is such thing, because the result may come from
> a complex join or a function. What does MySQL do when it cames to this?
> I have no idea, but am pretty sure that's why there is no similar field
> in libpq. And even parsing the query would win you nothing, because of
> the reason above.
> Regards, Christoph
>
And this is not the only case when table name has no sense.
What if result column is "calculated", for example:table1.field1 * 2
not to mention join:table1.field1 * table2.field1
However, I think that parser has all this information, and the only way to do
this right would be to get this information from parser, and fill the libpq
result structure with needed data. Column names that have no sense could be
marked as "unknown".
Hackers would know better, but I think that such change is not trivial.
If I remember well, Interbase has such feature.
Regards !