Motivation
This benchmark seems to suggest that building a string-based IAlternateEqualityComparer for ReadOnlySpan<byte> keys can have substantial performance and usability benefits over manually converting the key to an intermediate ReadOnlySpan<char>.
We should enhance StringComparer with an IAlternateEqualityComparer<ReadOnlySpan<byte>, string> implementation. I see a couple of potential approaches we could follow:
Approach 1: Factory method in System.Text.Encoding or UTF8Encoding
public partial class Encoding
{
public virtual EncodedStringComparer GetEncodedStringComparer(StringComparer stringComparer);
}
// Essentially just an intersection type for the two interfaces
public abstract class EncodedStringComparer : IEqualityComparer<string>,
IAlternateEqualityComparer<ReadOnlySpan<byte>, string>
{
// Interface implementations
}Which can then be used as follows:
EncodedStringComparer comparer = Encoding.UTF8.GetEncodedStringComparer(StringComparer.OrdinalIgnoreCase);
Dictionary<string, int> dictionary = new(comparer);
dictionary.GetAlternateLookup<ReadOnlySpan<byte>>(); // Success
Approach 2: hardcoding UTF-8 equality comparison into StringComparer
Title says it all, we could just make the assumption that UTF-8 is the encoding most people will end up using for ROS<byte> so we just bake it into StringComparer directly:
Dictionary<string, int> dictionary = new(StringComparer.Ordinal);
dictionary.GetAlternateLookup<ReadOnlySpan<byte>>(); // UTF-8 semantics whether you like it or not
cc @stephentoub @davidfowl
Motivation
This benchmark seems to suggest that building a string-based IAlternateEqualityComparer for
ReadOnlySpan<byte>keys can have substantial performance and usability benefits over manually converting the key to an intermediateReadOnlySpan<char>.We should enhance
StringComparerwith anIAlternateEqualityComparer<ReadOnlySpan<byte>, string>implementation. I see a couple of potential approaches we could follow:Approach 1: Factory method in
System.Text.EncodingorUTF8EncodingWhich can then be used as follows:
Approach 2: hardcoding UTF-8 equality comparison into
StringComparerTitle says it all, we could just make the assumption that UTF-8 is the encoding most people will end up using for
ROS<byte>so we just bake it intoStringComparerdirectly:cc @stephentoub @davidfowl