Return only user defined indexes from IngresDialect::get_indexes

[hab6]

Overview

A number of tests in area ComponentReflectionTest::get_multi_indexes of the SQLAlchemy Dialect Compliance Suite fail with an AssertionError caused by mismatched actual vs expected results.

Stack dumps from running with pytest reveal that the test suite expects only user defined indexes to be returned by the get_multi_index tests, whereas the Ingres dialect returns all indexes, including internal indexes defined automatically by the DBMS.

Here is a simple generic example, showing creation of a user index versus a system generated index:

=> CREATE TABLE staff (id INTEGER NOT NULL, ssn INTEGER NOT NULL, PRIMARY KEY (id), UNIQUE (id, ssn))
=> CREATE INDEX staffssn ON staff (ssn)

=> SELECT index_name, system_use FROM iiindexes WHERE base_name = 'staff'
    index_name                  system_use
    $staff_u000001820000        G
    staffssn                    U

Ingres documentation excerpt on iiindexes catalog :

Column Name	Data Type	Description
system_use	char(1)	S if the object is a system object U if user object G if generated by the system for the user Blank if unknown Used by utilities to determine which tables need reloading

Internal issue II-14927

Code change

The SQL statement used in IngresDialect::get_indexes is modified to also filter on the system_use column and retrieve only user defined indexes (ignoring system generated indexes).

Testing - SQLALchemy Dialect Compliance Suite

Command

pytest --db ingres_odbc -vv --maxfail=10000 .\test\dialect\test_suite.py --tb=no

(test.cfg contents: ingres_odbc=ingres:///sqatestdb)

Results - DBMS: Ingres 12.0

Tests	Passed	Failed	Errors	Skipped	Total
Before Fix	532	110	33	859	1534
With Fix	548	94	33	859	1534

Test results with DBMS targets: Vector 6.3 and Data Platform warehouse 620.0.18 (GCS)

This code change has NO impact when the backend is Vector or Data Platform, since the 84 tests of the dialect compliance suite under area ComponentReflectionTest::get_multi_indexes all involve DDL that tries to create a table with a self-referential constraint, which is not allowed on X100 tables.

When attempting to run the get_multi_indexes tests against an X100 backend, each of the tests fail with this error:

sqlalchemy.exc.DBAPIError: (pyodbc.Error) ('50000', "[50000] [Actian][Actian ODBC Driver
[INGRES]CREATE/ALTER TABLE: You cannot create a self referencing
referential constraint on table 'users' in schema 'actian'
because it is an X100 table. (328949) (SQLExecDirectW)")
[SQL:
CREATE TABLE users (
    user_id INTEGER GENERATED BY DEFAULT AS IDENTITY NOT NULL,
    test1 CHAR(5) NOT NULL,
    test2 FLOAT NOT NULL,
    parent_user_id INTEGER,
    PRIMARY KEY (user_id),
    CONSTRAINT zz_test2_gt_zero CHECK (test2 > 0),
    CHECK (test2 <= 1000),
    CONSTRAINT user_id_fk FOREIGN KEY(parent_user_id) REFERENCES users (user_id)
)

Environment

Microsoft Windows [Version 10.0.19045.4529]

Python 3.10.7

mock              5.1.0
packaging         24.1
pip               24.1.2
platformdirs      4.2.2
pluggy            1.5.0
pyodbc            5.1.0
pyproject-api     1.7.1
pypyodbc          1.3.6
pytest            8.2.2
setuptools        63.2.0
SQLAlchemy        2.0.30.dev0
sqlalchemy-ingres 0.0.8.dev0
tox               4.16.0
typing_extensions 4.12.2
virtualenv        20.26.3

DBMS:  II 12.0.0

[clach04]

Code looks good, commit message really should have had the PR text in the commit message for explanation.

I understand this fixes the test suite, I think we should offer an option to return system index information (off by default so test suite just runs). Either connection string option (preferred) or environment variable control..

Thoughts?

[clach04]

Code looks good, I think this should be conditional under user control.

[hab6]

Thanks for the review and suggestion.

Added logic allowing inspection of system generated indexes by setting connection execution option inspect_indexes="ALL". The default is to return only user-defined indexes.

Example program to check additional code: inspectIndexes.py.txt

Output:

C:\test\> python inspectIndexes.py MYURL1
Inspect1: [{'name': 'address_idx', 'column_names': ['address_id'], 'unique': False}]
Inspect2: [{'name': 'address_idx', 'column_names': ['address_id'], 'unique': False}, {'name': '$house_u000001eb00000000', 'column_names': ['house_id', 'address_id'], 'unique': True}]
Inspect3: [{'name': 'address_idx', 'column_names': ['address_id'], 'unique': False}]

[clach04]

Looks good! Thanks for the new feature :-)