polars-u64-idx: frame_equal fails when number of rows reaches 2^32

[cbilot]

polars-u64-idx 0.13.38
python 3.10.4
Linux Mint 20.3

Describe your bug.

The frame_equals method fails when the number of rows exceeds 2^32 - 1.

What are the steps to reproduce the behavior?

Despite the number of rows, I designed this MWE to work using polars-u64-idx on any computer with reasonable RAM.

import polars as pl

_nbr_rows = (2**32)
df = pl.select(pl.repeat(False, n=_nbr_rows, eager=True, name="col1"))
df

df.frame_equal(df)

>>> df
shape: (4294967296, 1)
┌───────┐
│ col1  │
│ ---   │
│ bool  │
╞═══════╡
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ ...   │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
└───────┘

>>> df.frame_equal(df)
False

However, if we reduce _nbr_rows to 2^32 - 1

shape: (4294967295, 1)
┌───────┐
│ col1  │
│ ---   │
│ bool  │
╞═══════╡
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ ...   │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
├╌╌╌╌╌╌╌┤
│ false │
└───────┘
>>> df.frame_equal(df)
True

Other Notes:

I've checked that I'm running polars-u64-idx

>>> pl.select(pl.repeat(False, n=(2**32) + 100,
...           eager=True, name="col1")
...           ).with_row_count()
shape: (4294967396, 2)
┌────────────┬───────┐
│ row_nr     ┆ col1  │
│ ---        ┆ ---   │
│ u64        ┆ bool  │
╞════════════╪═══════╡
│ 0          ┆ false │
├╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┤
│ 1          ┆ false │
├╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┤
│ 2          ┆ false │
├╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┤
│ 3          ┆ false │
├╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┤
│ ...        ┆ ...   │
├╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┤
│ 4294967392 ┆ false │
├╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┤
│ 4294967393 ┆ false │
├╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┤
│ 4294967394 ┆ false │
├╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌┤
│ 4294967395 ┆ false │
└────────────┴───────┘

The row_nr datatype is u64.

This came about as I tried to figure out why large datasets I wrote to parquet and IPC formats seemed corrupted. I'm guessing it's not about file corruption or file format issues ... but rather a row-indexing issue with frame_equal (which I was using to test the datasets read in from the files).