Removing pid of R-R and R-L relationship tables in RDFGraphs as a visible property to users (hashtag2781)

[vikramsubramanian]

The internalID type pid property of the R-R and R-L relationship tables is a system-level optimization. When users query them, they are unable to query them because of binding errors that are confusing or they get confusing outputs. For example:

MATCH (a)-[p:UniKG_rt]->(o) 
WHERE a.iri = " 
RETURN p.pid;

Output:
--------------------------
| p.pid                  |
--------------------------
| 18446744073709551615:1 |
--------------------------
| 18446744073709551615:9 |
--------------------------

Above, 18446744073709551615 looks like a NULL placeholder value. Or:

MATCH (a)-[p:UniKG_rt]->(o) 
WHERE a.iri = " AND p.iri = "  
SET p.pid=7 RETURN a.iri, p.iri, o.iri;
Error: Binder exception: Expression 7 has data type INT64 but expected INTERNAL_ID. Implicit cast is not supported.

Instead we should explicitly say that pids should not be queried and modified.

Instead, we have a feature that you can call an iri property on these relationships. In reality these relationship tables do not have iri properties but we internally don't trigger binding errors because we would like users to query and obtain string IRIs. So we can do this:

MATCH (a)-[p:UniKG_rt]->(o) 
RETURN p.iri

Output:
---------------------------------------------------
| p.iri                                           |
---------------------------------------------------
|                    |
---------------------------------------------------
|                     |
---------------------------------------------------
...

Or:

kuzu> MATCH (a)-[p:UniKG_lt]->(o)  
WHERE p.iri = " 
RETURN a.iri, p.iri, o.val;
------------------------------------------------------------------
| a.iri                      | p.iri                     | o.val |
------------------------------------------------------------------
|  |  | 30    |
------------------------------------------------------------------

But when a user looks at these tables, they see this:

kuzu> call table_info('UniKG_lt') return *;
------------------------------------
| property id | name | type        |
------------------------------------
| 1           | pid  | INTERNAL_ID |
------------------------------------

This behavior is confusing. I think we should just hid pid completely from the user. That is the following should be our behavior:

• Do not show pid in table schema in table_info('UniKG_lt') function calls. Instead show an iri property with STRING . You can also say "(virtual)" next to string.
• Any reference to pid should trigger a binding error.
• When relationship objects from these are returned, we can keep the pid property if this is quite a bit of work but ideally we can remove this feature from the objects as well. )

[mayil-ai[bot]]

Summary: The pid property in the R-R and R-L relationship tables in RDFGraphs should be hidden from users and not be queryable or modifiable.

Possible Solution

Based on the provided information and code snippets, the issue seems to be related to the handling of pid and iri properties in the context of the Kùzu graph database. The error message indicates that there is a type mismatch where an INT64 type is being used where an INTERNAL_ID type is expected, and implicit casting is not supported.

To resolve the issue:

• Ensure that the pid property is not exposed to the user. It should be used internally by the system only.
• Modify the table_info function to hide the pid property and instead expose the iri property as a virtual property with a STRING type.
• Update the MATCH queries to use the iri property instead of pid for user-facing queries.
• In the bind_graph_pattern.cpp file, where the queryRel object is being constructed, make sure that the pid property is not added to the list of properties exposed to the user. Instead, add the iri property as a virtual property.
• If the pid property is being used in any MATCH, RETURN, WHERE, or SET clauses in user-facing queries, replace it with the iri property.
• Review the code to ensure that any attempt to reference pid directly in user-facing queries results in a binding error, as indicated by the error message.
• If necessary, adjust the RelExpression and PropertyExpression classes to handle the iri property correctly as a virtual property.

By making these changes, you should be able to resolve the type mismatch error and ensure that the pid property is handled correctly as an internal identifier, while the iri property is exposed to users as a virtual, user-friendly identifier.

Code snippets to check

• src/binder/bind/bind_graph_pattern.cpp

  This snippet contains logic for binding properties to relationship patterns, which is relevant to hiding the pid property and ensuring it triggers a binding error when referenced.

• test/ddl/e2e_ddl_test.cpp

  This snippet demonstrates how to drop and rename properties in a table, which could be relevant for removing the pid property from the user's view.