In lib/sqlalchemy/orm/bulk_persistence.py, the default synchronize_session='auto' is automatically converted to 'evaluate' for bulk operations (line 791). However, 'evaluate' is then explicitly rejected when the UPDATE has additional WHERE criteria (lines 1642–1652):

# bulk_persistence.py, lines 1642-1652
if (
    statement._where_criteria
    and update_options._synchronize_session == "evaluate"
):
    raise sa_exc.InvalidRequestError(
        "bulk synchronize of persistent objects not supported "
        "when using bulk update with additional WHERE criteria right now. "
        "add synchronize_session=None execution option ..."
    )

The result is:

1. A user writes session.query(Model).filter(Model.x == 1).update({...}) — a completely ordinary pattern
2. synchronize_session defaults to 'auto', which silently becomes 'evaluate'
3. At execution time, not statement-construction time, an InvalidRequestError is raised

This is a late-stage, opaque failure. The error message says "right now", implying a known limitation, but:

• The public Query.update() and Session.execute() docs don't highlight this constraint under the auto strategy
• The conversion from auto → evaluate happens in one place (line 791) without a guard for WHERE criteria
• Users must read the error message carefully to understand they need synchronize_session=False or None

Questions:

1. Could the auto strategy detect WHERE criteria at conversion time and choose 'fetch' instead of 'evaluate' in that case?
2. If the evaluate + WHERE combination is permanently unsupported, should auto never resolve to evaluate for bulk UPDATEs?
3. Can a more descriptive error message or warning be emitted earlier (at update() call time rather than execution time)?

File: lib/sqlalchemy/orm/bulk_persistence.py, lines 791, 1642–1652.