Hard coded character encoding/transliteration taking place, ignoring II_CHARSETxx (etc.)

[clach04]

For example see:

sqlalchemy-ingres/lib/sqlalchemy_ingres/base.py

Line 637 in 89119af

return name.decode('latin1')

sqlalchemy-ingres/lib/sqlalchemy_ingres/base.py

Line 644 in 89119af

return name.lower().encode('latin1')

[clach04]

@hab6 additional notes, that maybe overkill but hopefully avoid confusion.

Initial idea / recommendations:

1. remove current code
2. then test with BOTH Python 2.7 and Python 3.x with pyodbc/pypyodbc (ignoring the other connectors)
3. If needed add appropriate encode()/decode() (possibly with if_py3 style code) - again focusing on the ODBC connector case

[hab6]

@clach04 Thanks for the suggestions.

To start off, I ran the SQLAlchemy test suite with the Ingres dialect methods normalize_name and denormalize_name both enabled (using Ingres dialect version 0.0.7.dev1) versus disabled (by changing those methods to simply return unaltered what was passed in).

Environment

(Both client and DBMS exist on the same machine)

Microsoft Windows [Version 10.0.19045.3930] 
Python 3.10.7
DBMS: Ingres 11.2.0 15807

mock              5.1.0
pyodbc            5.1.0
pypyodbc          1.3.6
pytest            8.1.1
setuptools        63.2.0
SQLAlchemy        2.0.28.dev0
sqlalchemy-ingres 0.0.7.dev1
tox               4.14.1

Command used to execute the SQLAlchemy test suite

pytest --db ingres_odbc --maxfail=15000 --ignore=test\dialect --ignore=test\orm\test_versioning.py --tb=no

Differences between the methods enabled vs disabled

Test	Failed	Passed	Skipped	Warnings	Errors	Run Time
Enabled	1914	12381	2037	7	2238	22m 39s
Disabled	1915	12438	2037	7	2161	24m 47s
Difference	+1	+57	same	same	-77	2m 8s

Summary: By disabling the dialect methods normalize_name and denormalize_name, the SQLAlchemy test suite ended with 57 more passing tests and 77 fewer errors.

I will run a few more tests with different environments and post the results.

[hab6]

More test results, this time using Ubuntu systems for the client and server. The pytest command to execute the test suite was the same one used in the prior test on the Windows system.

Unexpectedly, these results are way different (much worse) than running the test on Windows. For example, the tests on Ubuntu with VW encountered around 5K more errors than the results on the Windows system. This probably will merit opening a separate issue to investigate the huge disparity of results.

Client: Ubuntu 22.04.2 LTS
DBMS: VectorWise 6.3.0 14608 on Ubuntu 20.04.6 LTS

Python 3.10.12

mock              5.1.0
pluggy            1.4.0
pyodbc            5.1.0
pypyodbc          1.3.6
pytest            8.1.1
setuptools        69.2.0
SQLAlchemy        2.0.28.dev0
sqlalchemy-ingres 0.0.7.dev1
tox               4.14.1

Results from running the SQLAlchemy test suite with the Ingres dialect methods normalize_name and denormalize_name enabled versus disabled.

Test	Failed	Passed	Skipped	Warnings	Errors	Run Time
Enabled	1026	9284	1737	8	7002	28m 32s
Disabled	996	9253	1738	9	7068	29m 8s
Difference	-30	-31	+1	+1	+66	+36s

[clach04]

Change looks good, sounds like next decision is whether to retain or delete these functions now.