Hello everyone!
Recently I've found out about the specialization and min_specialization features that allow some type eligible for some generic trait implementation to have a different implementation of the trait. The former is considered unsound, the latter has stricter rules but is considered safe. For example, this code compiles with specialization, but not with min_specialization:
pub trait IsBuffered
{
fn is_buffered(&self) -> bool;
}
impl<T> IsBuffered for T where T: std::io::Read
{
default fn is_buffered(&self) -> bool { false }
}
impl<T> IsBuffered for T where T: std::io::BufRead
{
fn is_buffered(&self) -> bool { true }
}
Compiler error is "cannot specialize on trait BufRead".
The same applies to generic types:
pub trait IsBuffered
{
fn is_buffered(&self) -> bool;
}
impl<T> IsBuffered for T where T: std::io::Read
{
default fn is_buffered(&self) -> bool { false }
}
impl<T> IsBuffered for std::io::BufReader<T> where T: std::io::Read
{
fn is_buffered(&self) -> bool { true }
}
The error is "cannot specialize on trait Read". However, this code:
pub trait IsBuffered
{
fn is_buffered(&self) -> bool;
}
impl<T> IsBuffered for T where T: std::io::Read
{
default fn is_buffered(&self) -> bool { false }
}
impl<T> IsBuffered for std::io::BufReader<std::fs::File>
{
fn is_buffered(&self) -> bool { true }
}
is allowed in min_specialization. I wonder, what are exact rules for min_specialization? Do they mean that default fn implementations can only be overridden for concrete types but not generic types?