Add dtype keyword to read_sql_query to control per column dtypes. · Issue #10285 · pandas-dev/pandas (original) (raw)

When reading from SQL queries - especially when reading chunk-wise - issues with type conversion can occur.

It would be ideal if one could specify the types of the columns, so that type conversions go right.

A very common case is reading pure float results chunk-wise from a large table. The nuisance comes in when and chunk contains NULL values only in a column. In such cases these NULL values are stored as None rather than float("nan"). I see that this is an issue with inconsistent NULL types in pandas and missing type information in read_sql_query.
Another case could be when text columns happen to contain numbers only in a chunk.

Specifying types for a query manually could resolve these issues.