Here's an idea I had for handling splits.
Right now, QueryAsembler.dbRelationshipAdded(..) removes "duplicates".
If instead we temporarily "overrode" aliases when resolving a split
expression, it seems as if this might work without any further
changes.
Going back to my standard example:
SELECT * FROM (SELECT [...] FROM
ENG_WORK_MGMT.FEE t0,
ENG_WORK_MGMT.FEE_TYPE t1,
ENG_WORK_MGMT.FEE_CYCLE
ENG_WORK_MGMT.AUTHORIZATION_DOCUMENT t3,
ENG_WORK_MGMT.PERMIT_DOCUMENT t4,
ENG_WORK_MGMT.AUTHORIZATION_DOCUMENT t5
WHERE
t0.FEE_TYPE_ID = t1.FEE_TYPE_ID
AND t0.FEE_ID = t2.FEE_CYCLE_ID
AND t2.FEE_CYCLE_ID = t3.INITIAL_FEE_CYCLE_ID(+)
AND t3.AUTHORIZATION_DOCUMENT_ID = t4.PERMIT_DOCUMENT_ID(+)
AND t2.FEE_CYCLE_ID = t5.RECURRING_FEE_CYCLE_ID(+)
AND ((t1.DESCRIPTION = ?)
AND ((t4.AGENCY_ID = ?)
OR (t4.AGENCY_ID = ?))))
We start off with an alias table
FEE = t0
FEE_TYPE = t1
FEE_CYLE = t2
Now we process an expression with a split
AUTHORIZATION_DOCUMENT = t3
PERMIT_DOCUMENT = t4
the split expression ends, so we revert these two values.
Next expression with a split gives us
AUTHORIZATION_DOCUMENT = t5
PERMIT_DOCUMENT = t6
And these are again reverted in the alias table when the split expression ends.
The reason I say reverted is that there may be regular inner joins
(non-split) against these same tables. My example doesn't really
show that off well.
Assume there's another AUTHORIZATION_DOCUMENT.date = yyyy-mm-dd phrase
that is an inner join. It already exists as an alias of
AUTHORIZATION_DOCUMENT = t2Point5, so we want this to still be back in
the alias table as soon as we are no longer processing the split.
Hope this makes sense. I don't feel I'm explaining it very well.
It's probably easier to implement than to explain, so I think this
will be my next step.
This archive was generated by hypermail 2.0.0 : Sat Aug 19 2006 - 20:03:26 EDT