Ah sheesh, too late to correct my own thoughts. Responses came in. Anyway,
notes to self here.
- Foreign keys must point to UNIQUE or PRIMARY indexes
I don't see a point in having a relation where the referencing entity has a record that points
ambiguously to more than 1 record in the referenced entity.
- Foreign keys must point to *full* indexes, not partial
Consider how a ON DELETE CASCADE can cause a child record to be erroneously deleted. Why? The
conditions for cascading the delete is not specific enough.
That brings me down to only 2 wishlist items.
- Have a "ON DELETE/UPDATE" attribute for <relation>s (optional)
- Build an application-level foreign key constraint enforcement in Entity
Engine.
For eg, PartyContactMechPurpose could have a type "one-app" relation to PartyContactMech's
[partyId, contactMechId, fromDate]. The "one-app" will mean a one-to-one foreign key constraint
enforced by the application.
No real foreign key will be created in database.
What do you think?
And now, to read the many responses...
Jonathon
Jonathon -- Improov wrote:
More thoughts aloud.
> Of course, that would mean we cannot easily change the "fromDate"
field in
> PartyContactMech (unless we do EECA).
Correction. With foreign key constraint in place, it is not possible to
use EECA to do a pseudo "ON UPDATE CASCADE". I can't find any "turn off
foreign key checks" switches in OFBiz, nor any standard way to do so in
various RDBMSs.
I was wondering why we need to have foreign keys *exactly match*
referenced primary keys. In MySQL, it is possible to have a foreign key
reference part of an index (top part, not tail end). This way, we can
have a foreign key "[partyId, contactMechId]" from
PartyContactMechPurpose to PartyContactMech. We can do application-level
(EECA) checks for fromDate. Currently, OFBiz entities don't have foreign
keys at all when it encounters fromDate entities!
Also, why don't we allow foreign keys to reference indexes on referenced
entities? Restricting foreign keys to referencing primary keys only
seems a bit inflexible.
Speaking of application-level (EECA) checks for fromDate, is it a good
idea to have some automatic checks at the Entity Engine level? Much like
how fields "lastUpdatedStamp", "lastUpdatedTxStamp", etc, are handled. I
know it's slower than using "ON UPDATE CASCADE". But not all RDBMSs have
that, I think. We can always have a switch to toggle on/off the
application-level checks for fromDate fields.
Speaking of "ON UPDATE CASCADE", should we allow an attribute in
<relation> for this? Do we also want to auto-detect the RDBMS being
used, and scream a warning if we encounter one that doesn't support "ON
UPDATE CASCADE"?
Am I missing a whole chunk of the Entity Engine? Or is this a commonly
wished wishlist?
Jonathon
Jonathon -- Improov wrote:
Ah, wait. Note one more thing.
In entity PartyContactMechPurpose, there is supposed to be a type
"one" relation to PartyContactMech. Why is it missing? We could have a
field like "partyContactMechFromDate", so it doesn't clash with
"fromDate". Same for "thruDate". Of course, that would mean we cannot
easily change the "fromDate" field in PartyContactMech (unless we do
EECA).
As it is now, it is possible to actually get PartyContactMechPurpose
to point to a non-existent PartyContactMech. Just mix up the partyId
and contactMechId such that no such combination exists.
Advice? Or should we live without foreign key checks in such cases?
Jonathon
Jonathon -- Improov wrote:
I found this from David Jones:
> Foreign keys are done for type "one" relationships, not type many.
A type
> "many" relationship is usually just the reverse direction of a
type "one"
> relationship so the FK covers both.
>
> What would it mean to have a foreign key on a type "many"
relationship?"
Then a corresponding "one" relation will supply the foreign key
constraint?
What about a many-to-many relation? Look at entity PartyAttribute and
PartyTypeAttr for example. Is that a many-to-many? Looks odd, though.
A many-to-many usually requires a separate "match-make" table.
Is it correct to say that OFBiz does not do foreign key constraints
with type "many" relations?
Do we have to insert an additional type "one" relation on field
"attrName" in order to get the foreign key constraints checks? But
can a "one" relation be created without specifying the full primary key?
Looks like the type "many" relation is merely a convenient means to
do a query like "where attrName = whatever", so we can do a simple
getRelated().
Jonathon
