Parquet row group filter pushdown not working as expected

[fpixl]

Checks

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

con = duckdb.connect(database=":memory:")
con.sql(
    """
    SELECT * FROM read_parquet('1BilRowData.parquet')
    WHERE attribute = 'gender';
    """
)

df = pl.scan_parquet("1BilRowData.parquet")
df = df.filter(pl.col("attribute") == "gender").collect()

Log output

No response

Issue description

I have a large parquet file (2 billion rows) in entity-attribute-value format, where data expands as rows instead of columns. The reason we do that is because we can have hundreds of thousands of attributes per entity.
Ie. Instead of being formatted as:

row_id	name	gender	age
1	Alice	f	21
2	Bob	m	22
3	Joe	m	23

It's formatted as:

row_id	attribute	value
1	name	Alice
2	name	Bob
3	name	Joe
1	gender	f
2	gender	m
3	gender	m
1	age	21
2	age	22
3	age	23

The data is produced as a single parquet file with row groups by attribute. The read patterns are by attribute, we never scan the whole file, so the expectation is that polars can filter the data by row group and return the values without reading the whole file in memory (similar to what DuckDB does well).

Expected behavior

When filtering a parquet file by a column used as row group, it uses row group push down to efficiently read the data.

Installed versions

--------Version info---------
Polars:               0.20.16
Index type:           UInt32
Platform:             macOS-13.6-x86_64-i386-64bit
Python:               3.11.5 (main, Sep 22 2023, 10:10:52) [Clang 14.0.0 (clang-1400.0.29.202)]

----Optional dependencies----
adbc_driver_manager:  <not installed>
cloudpickle:          <not installed>
connectorx:           <not installed>
deltalake:            <not installed>
fastexcel:            <not installed>
fsspec:               2023.12.2
gevent:               <not installed>
hvplot:               <not installed>
matplotlib:           <not installed>
numpy:                1.26.4
openpyxl:             <not installed>
pandas:               2.2.1
pyarrow:              15.0.2
pydantic:             2.6.4
pyiceberg:            0.6.0
pyxlsb:               <not installed>
sqlalchemy:           <not installed>
xlsx2csv:             <not installed>
xlsxwriter:           <not installed>

[ritchie46]

When filtering a parquet file by a column used as row group, it uses row group push down to efficiently read the data.

It does. Can you share a MRE that shows/confirms a bug?

[ritchie46]

We cannot do anything with the bug report as it is now.