> Which statements can be executed with many XSQLDA's ?
> INSERT\UPDATE\DELETE only ?

Yes.

> Is SELECT allowed ? If so - how
> result sets will be returned ? As UNION ALL or something different ?

SELECTs are not allowed.

> How errors is handled ? Will we return N instances of status-vector
> or we will stop execution after the first error ?

Implementation defined, I would suggest to stop execution after first
error.

However, it would be required to return insert/update/delete count for
each executed statement (if I remember correctly, that are the only
items that can be queried via isc_dsql_sql_info after execution, since
we cannot ask for those values after the call. So, we have to design
new "status vector" for batch updates - an array of status vectors for
single statements: for successful records it should contain
insert/update/delete count, for failed records normal status vector
including the error code.

In worst case, stop after first error and single status vector with
error code would be enough. We will just have to say that information
about statement outcome is not available.

Roman