feat(ibis): use the default input dialect

[goldmedal]

Description

Close #930
The root cause of #930 is the trino dialect has a specific limit for the date_add function. We can't use it as a normal scalar function. We should follow the spec of trino doc.

import sqlglot
sql = "SELECT date_add('second', 1, CAST('2024-01-01' AS DATE))"
result = sqlglot.transpile(sql, read="trino", write="bigquery")[0]
print(result)
# SELECT DATE_ADD(CAST('2024-01-01' AS DATE), INTERVAL 1 SECOND)

However, we want a common SQL base for some general purpose. I think we can change the input dialect to the default dialect which has fewer limitations for the function usage.

import sqlglot
sql = "SELECT date_add(CAST('2024-01-01' AS DATE), INTERVAL 1 DAY)"
result = sqlglot.transpile(sql, write="bigquery")[0]
print(result)
# SELECT DATE_ADD(CAST('2024-01-01' AS DATE), INTERVAL (INTERVAL '1' DAY) DAY)

However, the output for the BigQuery is still wrong. INTERVAL (INTERVAL '1' DAY) DAY) isn't a valid SQL for BigQuery. It seems SQLGlot has some specific definition for the date_add function. The SQL should be

import sqlglot
sql = "SELECT date_add(CAST('2024-01-01' AS DATE), 1)"
result = sqlglot.transpile(sql, write="bigquery")[0]
print(result)
# SELECT DATE_ADD(CAST('2024-01-01' AS DATE), INTERVAL 1 DAY)

It's different from the original definition of BigQuery

Additional Changed

• Modified the date_add definition for bigquery
• Disable instr for MySQL
  • Because instr is an alias of strpos in DataFusion, the input instr function will be transformed to strpos, then the sqlglot can't transpile strposto instr when using the default dialect to mysql dialect`. We need to file another issue for it.

[grieve54706]

This method is used for v2 and v3. For v2, the read dialect is trino. How about splitting this method for each class?

[grieve54706]

Or, do we not receive the statements with the trino dialect? I'm not sure. If the input statement of v2 is already not trino dialect, your way is right.

[goldmedal]

hmm.. No, maybe you're right. We're better to split v2 and v3 because the trino dialect also handles the functions converting between trino and others data source ( Currently, the v2 function white list used by AI service is based on the trion function) 🤔

[goldmedal]

The BigQuery CI has issues. I tested BigQuery in my local environment and all pass.

$ just test bigquery
poetry run pytest -m 'bigquery'
============================================================================================================================== test session starts ===============================================================================================================================
platform darwin -- Python 3.11.6, pytest-8.3.3, pluggy-1.5.0
rootdir: /ibis-server
configfile: pyproject.toml
plugins: anyio-4.6.2.post1
collected 205 items / 175 deselected / 30 selected                                                                                                                                                                                                                               

tests/routers/v2/connector/test_bigquery.py .................                                                                                                                                                                                                              [ 56%]
tests/routers/v3/connector/bigquery/test_functions.py ....                                                                                                                                                                                                                 [ 70%]
tests/routers/v3/connector/bigquery/test_query.py .........                                                                                                                                                                                                                [100%]

================================================================================================================= 30 passed, 175 deselected in 283.35s (0:04:43) =================================================================================================================

[grieve54706]

LGTM