Error when a parquet partition key contains numbers and alphabetics characters about duckdb HOT 5 CLOSED

Thanks for reporting this. I can indeed reproduce this on Linux but it does not occur on macOS, so this is a platform-specific bug.

from duckdb.

Forgot to say that a work around is is passing hive_types :

D SELECT * FROM 'ds_pqt/**/*.parquet';
Error: Invalid Input Error: Unable to cast 'a' (from hive partition column 'KEY') to: 'BIGINT'
D 
D SELECT * FROM read_parquet('ds_pqt/**/*.parquet',  hive_types={'key': string});
┌─────────┬─────────┐
│   key   │  name   │
│ varchar │ varchar │
├─────────┼─────────┤
│ 1       │ Hannes  │
│ a       │ Mark    │
└─────────┴─────────┘

from duckdb.

This is sortof expected behaviour:

The default behaviour for read_parquet is to automatically detect hive partitioned paths. The hive partition schema is automatically infered, but this can pick the wrong schema.

The solution is either:

• disabling hive_partitioning (note that duckdb also write hive partition columns to the files)
• provide the schema manually as you did
• disable casting of hive partitions through hive_types_autocast=0 (https://duckdb.org/docs/data/partitioning/hive_partitioning#hive-types)

I think we should look into improving the type detection here though, currently we check only one file which is a bit naive perhaps

from duckdb.

This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 30 days.

from duckdb.

Should be fixed now by #11676

from duckdb.