Similar to #161350, but with zip instead of take.
I tried this code:
use std::fmt::Debug;
fn main() {
eprintln!("does_print:");
consume(does_print());
eprintln!("does_not_print:");
consume(does_not_print());
}
fn does_not_print() -> impl ExactSizeIterator<Item: Debug> + DoubleEndedIterator {
[1, 2, 3].iter()
}
fn does_print() -> impl ExactSizeIterator<Item: Debug> + DoubleEndedIterator {
[1, 2, 3].iter().inspect(|_| {}) // use any noop adapter without `TrustedRandomAccessNoCoerce` here
}
// all of the commented out methods have the same inconsistency
fn consume(iter: impl ExactSizeIterator<Item: Debug> + DoubleEndedIterator) {
iter.map(|it| dbg!(it))
.zip(([] as [i32; 0]).iter()) // zip with an empty random access iterator *on the right*
.next();
// .next_back();
// .fold((), |(), _| ());
// .nth(0);
}The TrustedLen + !TrustedRandomAccessNoCoerce specialization for fold is also affected. Use std::iter::empty as the empty rhs iterator and replace inspect with filter(true) to see this.
I expected to see this happen:
Either both iterations print, or neither does, since the only difference is that one iterator is being adapted with inspect(|| {}), which is a noop.
Instead, this happened:
does_print:
[src/main.rs:19:19] it = 1
does_not_print:
@rustbot label A-iterators T-libs
Similar to #161350, but with
zipinstead oftake.I tried this code:
The
TrustedLen + !TrustedRandomAccessNoCoercespecialization forfoldis also affected. Usestd::iter::emptyas the empty rhs iterator and replaceinspectwithfilter(true)to see this.I expected to see this happen:
Either both iterations print, or neither does, since the only difference is that one iterator is being adapted with
inspect(|| {}), which is a noop.Instead, this happened:
@rustbot label A-iterators T-libs