Column order/data scrambled when reading back from ipc file (regression?)

[alexander-beedie]

What language/platform are you using?

Python 3.9, macOS 12.4, Polars 0.13.46

Describe your bug.

The columns directive for pl.read_ipc appears to have a regression in one of the recent updates; if a non-default column order is provided, the column names/data can load in an order that is neither the original order, nor the requested order. (A somewhat similar issue was previously fixed by #3591).

What are the steps to reproduce the behavior?

import polars as pl
df = pl.DataFrame(
    data = [
        ['x',123, 4.5, 'misc'],
        ['y',456,10.0,'other'],
        ['z',789,10.0,'value'],
    ],
    columns = ['a','b','c','d'],
)
print( df )
# ┌─────┬─────┬──────┬───────┐
# │ a   ┆ b   ┆ c    ┆ d     │
# │ --- ┆ --- ┆ ---  ┆ ---   │
# │ str ┆ i64 ┆ f64  ┆ str   │
# ╞═════╪═════╪══════╪═══════╡
# │ x   ┆ 123 ┆ 4.5  ┆ misc  │
# │ y   ┆ 456 ┆ 10.0 ┆ other │
# │ z   ┆ 789 ┆ 10.0 ┆ value │
# └─────┴─────┴──────┴───────┘

# save frame data to feather/ipc file
df.write_ipc( 'test.feather' )

# load back in requested (different) column order: data gets scrambled
dx = pl.read_ipc( 'test.feather', columns=['a','c','d','b'] )
print( dx )
# ┌─────┬───────┬─────┬──────┐
# │ a   ┆ c     ┆ d   ┆ b    │  << column *names* are in the requested order,
# │ --- ┆ ---   ┆ --- ┆ ---  │     but the associated column *data* is incorrect
# │ str ┆ str   ┆ i64 ┆ f64  │     
# ╞═════╪═══════╪═════╪══════╡     col 'b' should have i64 data, not f64
# │ x   ┆ misc  ┆ 123 ┆ 4.5  │     col 'c' should have f64 data, not str
# │ y   ┆ other ┆ 456 ┆ 10.0 │     col 'd' should have str data, not i64
# │ z   ┆ value ┆ 789 ┆ 10.0 │
# └─────┴───────┴─────┴──────┘

What is the actual behavior?

Loaded column data is not correct.

What is the expected behavior?

Load the column data in the requested order.

dx = pl.read_ipc( 'test.feather', columns=['a','c','d','b'] )
print( dx )
# ┌─────┬──────┬───────┬─────┐
# │ a   ┆ c    ┆ d     ┆ b   │
# │ --- ┆ ---  ┆ ---   ┆ --- │
# │ str ┆ f64  ┆ str   ┆ i64 │
# ╞═════╪══════╪═══════╪═════╡
# │ x   ┆ 4.5  ┆ misc  ┆ 123 │
# │ y   ┆ 10.0 ┆ other ┆ 456 │
# │ z   ┆ 10.0 ┆ value ┆ 789 │
# └─────┴──────┴───────┴─────┘

[alexander-beedie]

Actually this still seems to be misbehaving (in a slightly different way?)
Using the same DataFrame as above ::

# save frame data to feather/ipc file in column-default order
df.write_ipc( 'test.feather' )

# load back in requested (different) column order
dx = pl.read_ipc( 'test.feather', columns=['a','c','d','b'] )

print( dx )
# ┌─────┬───────┬─────┬──────┐
# │ a   ┆ d     ┆ b   ┆ c    │  <<< columns not in requested order 
# │ --- ┆ ---   ┆ --- ┆ ---  │      (though associated with the correct datatype)
# │ str ┆ str   ┆ i64 ┆ f64  │ 
# ╞═════╪═══════╪═════╪══════╡
# │ x   ┆ misc  ┆ 123 ┆ 4.5  │
# ├╌╌╌╌╌┼╌╌╌╌╌╌╌┼╌╌╌╌╌┼╌╌╌╌╌╌┤
# │ y   ┆ other ┆ 456 ┆ 10.0 │
# ├╌╌╌╌╌┼╌╌╌╌╌╌╌┼╌╌╌╌╌┼╌╌╌╌╌╌┤
# │ z   ┆ value ┆ 789 ┆ 10.0 │
# └─────┴───────┴─────┴──────┘