Re: VisAD and SQL databases


Bill Hibbard wrote:

Hi Ugo,

How are you going to pass the information about which table is numerical
and which is textual?

I know almost nothing about SQL. Can the interface tell from
the table itself (via JDBC calls, presumably) which domains
are numerical and which are text?
Well, I think this is possible. Something like

rsmd.getColumnType( col_number );

(rsmd is ResultSetMetaData. The method above returns an int, which is a code - e.g. Decimal, Date, String. Anyway, I'd say it's possible, although I might be missing the right method. Any better suggestions out there?)

I used to do, for example, rs.getFloat( "my_column" );, where rs is the ResultSet.

For our app, we had

( x -> y) (FlatField, which represents a time series, i.e. a simple 2D plot)

whereas you're suggesting

( record_index -> (x,y))

which when mapped
x -> XAxis
y -> YAxis

is depicted as 2d scatter plot. (Am I right?)

So I'm still with my approach (though I might take back my "it is
impossible to write an adapter which will work with every JDBC-capable
database" without adding at least "which will give you the data
depiction you want" ;-)

That's right. The application still needs to seperate
independent and dependent variables as in:

If domains split into independent and dependent variables,
then the application would need to rearrange the MathType
to get something like:

((domain1, domain2) -> (domain3, ..., domainN))

with an appropriate domain Set. This is like what Donna has
done, constructing an Irregular2DSet.

Unless there's some way to find out from JDBC which domains are
independent variables and which are dependent.
Don't think so...but I'm definitely not the SQL (or DB)  expert.

But probably no
way to find out whether the set of independent variables form
a Gridded or Irregular topology (this highlights a difference
between the VisAD and SQL data models).


Yes, that's why I think any SQL adapter will be (too much of) a trade-off.