read_sql function eats informative error messages.

[chrisgrimm]

Consider the following code.

import pandas as pd
import psycopg2 

bad_query = 'select bad_column from my_table;'
conn = psycopg2.connect(...)
pd.read_sql(bad_query, conn)

Running the above in my environment produces a DatabaseError exception with the following message:
"Execution failed on sql: select bad_column from my_table;"

However, when I run the below code to query the database:

import psycopg2

bad_query = 'select bad_column from my_table;'
conn = psycopg2.connect(...)
conn.cursor().execute(bad_query)

The following ProgrammingError message is returned:
"column "bad_column" does not exist in my_table"

This information is helpful for debugging, especially as queries become more complicated, and I believe it should be propagated through to the user.

[jorisvandenbossche]

Thanks for the report, and I agree it would be helpful to propagate this.

Could you also try it with a sqlalchemy engine instead of a raw psycopg2 connection? (to see of the error message is also suppressed in that mode)

BTW, I also recommend to use an sqlalchemy engine instead of a raw connection, as postgresql is not really supported without sqlalchemy (for reading a query, it will work (but not guaranteed), but for reading tables or writing to the database, it will not work anymore)

[chrisgrimm]

I just checked. The informative error message is successfully propagated when an sqlalchemy engine is used as opposed to the psycopg2 connection object.

[jorisvandenbossche]

Thanks for checking! Then it is only an issue with the fallback mode