rusackas opened a new pull request, #7787:
URL: https://github.com/apache/kyuubi/pull/7787
### Why are the changes needed?
`python/pyhive/sqlalchemy_hive.py`'s `HiveDialect` never defines its own
`colspecs`, so it inherits `sqlalchemy.engine.default.DefaultDialect.colspecs`
directly, the same dict object, not a copy:
```python
>>> from sqlalchemy.engine.default import DefaultDialect
>>> from pyhive.sqlalchemy_hive import HiveDialect
>>> HiveDialect.colspecs is DefaultDialect.colspecs
True
```
Every SQLAlchemy built-in dialect (`postgresql`, `mysql`, etc.) defines its
own `colspecs` dict, even when there's nothing to override yet, specifically to
avoid this. Mutating a dialect's `colspecs` in place
(`HiveDialect.colspecs[SomeType] = ...`) is a completely normal way to
customize how a type renders for one dialect. Without its own dict, that
mutation silently corrupts type handling for every other dialect in the same
process that also hasn't defined its own `colspecs`, not just Hive.
This isn't hypothetical: Apache Superset's Databricks engine spec subclasses
`HiveDialect` for Databricks support and patches
`HiveDialect.colspecs[types.String] = ...` to customize string-type handling.
Because of the shared object, that one line broke type compilation for ~38
unrelated third-party dialects loaded in the same process, until Superset
worked around it on its own side (apache/superset#44707).
Filed the identical fix upstream at dropbox/PyHive#490 before noticing this
module had already been donated here. Confirmed the bug is present in this copy
too.
Fixes #7786.
### How was this patch tested?
Gave `HiveDialect` its own `colspecs = {}` and confirmed the isolation
directly, same check as in the linked issue:
```python
>>> HiveDialect.colspecs is DefaultDialect.colspecs
False
>>> HiveDialect.colspecs['marker'] = 'hive-only'
>>> DefaultDialect.colspecs
{}
```
`colspecs` was and remains empty for this dialect; it's just no longer the
same object as `DefaultDialect`'s, so a mutation on one no longer leaks into
the other.
I don't have a live Hive/Kyuubi server available in my environment to run
the full `python/` test suite end to end, but this change touches only a
class-level attribute with no effect on connection or query logic, and verified
it above directly against this module.
### Was this patch assisted by generative AI tooling?
Assisted-by: Claude:claude-sonnet-5
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
