Rust warns pub glob export is unused

[orlp]

I tried this code:

#[cfg(feature = "dtype-categorical")]
pub use crate::chunked_array::logical::categorical::*;

I expected to see this happen: no warning, since it is a pub export in a pub module.

Instead, a warning is raised:

warning: unused import: `crate::chunked_array::logical::categorical::*`
  --> crates/polars-core/src/prelude.rs:30:9
   |
30 | pub use crate::chunked_array::logical::categorical::*;

Reproduces in orlp/polars@0626047, with the command

cargo check -p polars-core --features dtype-categorical 2>&1 | rg "pub use" --before-context=3

There are other warnings due to an incomplete set of features, but the specific warning here is invalid regardless of features.

Meta

rustc --version --verbose:

rustc 1.99.0-nightly (7608eb7b0 2026-08-05)
binary: rustc
commit-hash: 7608eb7b07eaf93f16d7cf5bcb2098eca87503df
commit-date: 2026-08-05
host: aarch64-apple-darwin
release: 1.99.0-nightly
LLVM version: 23.1.0

Backtrace

<backtrace>

[orlp]

This is a regression from nightly-2026-04-01, currently bisecting.

[orlp]

I added #[deny(unused_imports) to the relevant line, giving the following bisect:

searched nightlies: from nightly-2026-04-01 to nightly-2026-08-07
regressed nightly: nightly-2026-06-09
searched commit range: f20a92e...cb46fbb
regressed commit: 029c9e1

bisected with cargo-bisect-rustc v0.6.11

Host triple: aarch64-apple-darwin
Reproduce with:

cargo bisect-rustc --start=2026-04-1 -- check -p polars-core --features dtype-categorical 

[orlp]

Based on that bisect, from that commit it is likely #145108. By @LorrensP-2158466, reviewed by @petrochenkov.

I find it a bit concerning that this is the third regression in Polars in a row (#152004, #157420 are the previous two) when trying to bump to the latest nightly, all on import warnings. I understand that Polars is a rather extreme case of glob imports (something we'd like to move away from but haven't had the resources yet to do so), but it does indicate some sort of lack of testing in this area.

[petrochenkov]

I find it a bit concerning that this is the third regression in Polars in a row (#152004, #157420 are the previous two)

The core reason for all of them is the same, polars uses many (potentially) ambiguous glob imports in creative ways + rustc has been trying to fix bugs in this area.

but it does indicate some sort of lack of testing in this area.

Polars is our testing 😅 (And the rest of the crates.io ecosystem in general.)
All the changes were accompanied with tests of course, but with glob imports there are a lot of possibilities for things to go wrong.
It would be good to extend the testing with some fuzzing, but similarly, there's a problem with resources.

[rustbot]

Assigning P-high (discussion on Zulip).