df.explode() + df.unnest() causes PanicException: called Result::unwrap() on an Err value

[TimovNiedek]

Polars version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of Polars.

Issue description

I'm running into an unexpected error when I have a dataframe with a column of a list of structs that I first explode, and then unnest. Whenever I try to select any of the unnested fields, I get the following exception:

pyo3_runtime.PanicException: called `Result::unwrap()` on an `Err` value: NotFound(Owned("users"))

Reproducible example

import polars as pl

df = pl.DataFrame(
    [
        {"collection": 0, "users": [{"email": "test_0@test.com"}]},
        {"collection": 1, "users": [{"email": "test_1@test.com"}]},
    ]
).lazy()

df = df.explode("users")
df = df.unnest("users")

df.select("email").collect(
    projection_pushdown=True,
)

thread '<unnamed>' panicked at 'called `Result::unwrap()` on an `Err` value: NotFound(Owned("users"))', /Users/runner/work/polars/polars/polars/polars-lazy/polars-plan/src/utils.rs:343:68
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Traceback (most recent call last):
  File "/.../polars_bug_example.py", line 13, in <module>
    df.select("email").collect(
  File "/.../lib/python3.9/site-packages/polars/utils.py", line 337, in wrapper
    return fn(*args, **kwargs)
  File "/.../lib/python3.9/site-packages/polars/internals/lazyframe/frame.py", line 1154, in collect
    return pli.wrap_df(ldf.collect())
pyo3_runtime.PanicException: called `Result::unwrap()` on an `Err` value: NotFound(Owned("users"))

Expected behavior

I expect to be able to select the email column from the users list of structs, after it has been exploded and unnested.

I found that there are two workarounds, but I'm not sure about the implications on performance. Hopefully this helps in finding the root cause:

1. Turning off projection_pushdown optimization
2. Creating a new column with an alias for the exploded column like:

   df = df.explode("users").with_column(pl.col("users").alias("users_flat"))
   df = df.unnest("users_flat")
   

Installed versions

---Version info---
Polars: 0.15.11
Index type: UInt32
Platform: macOS-12.6-arm64-arm-64bit
Python: 3.9.13 (main, Sep 29 2022, 15:54:06) 
[Clang 14.0.0 (clang-1400.0.29.102)]
---Optional dependencies---
pyarrow: 10.0.1
pandas: 1.5.2
numpy: 1.24.1
fsspec: <not installed>
connectorx: <not installed>
xlsx2csv: <not installed>
matplotlib: <not installed>

[martsec]

Hi, I don't know if this fix is already in new polars versions, but I'm facing a similar issue in 1.76.0

The error given was Polars Failed: NoData(ErrString("empty container given"))

let raw: LazyFrame = JsonReader::new(reader: &mut cursor) // JsonReader<'_, &mut Cursor<…>>
    .finish() // Result<DataFrame, PolarsError>
    .unwrap() //DataFrame
    .lazy() //LazyFrame
    .explode(columns: ["aircraft"]) //LazyFrame
    .select(exprs: [col("now"), col("aircraft")]) //  I have to force this in order to work. doing col("*") did not work
    .unnest(cols: ["aircraft"]);

println!("{}", &raw.clone().collect()?);

I have a complex struct to unnest (58 fields), I don't know if this can affect a lot the projection.