Rollup merge of #120053 - AldanTanneo:specialize-stdinlock-bytes, r=t…

…he8472

Specialize `Bytes` on `StdinLock<'_>`

I noticed recently, while profiling a little project, that I was spending a lot of time reading from stdin (even with locking). I was using the `.bytes()` iterator adaptor; I figured, since `StdinLock` is a `BufReader` internally, it would work just as fast. But this is not the case, as `Bytes` is only specialized for the raw `BufReader`, and not the `StdinLock`/`MutexGuard` wrapper. Performance improved significantly when I wrapped the lock in a new `BufReader`, but I was still a bit sore about the double buffer indirection.

This PR attempts to specialize it, by simply calling the already specialized implementation on `BufReader`.

library/std/src/io/stdio.rs

@@ -8,7 +8,9 @@ use crate::io::prelude::*;
 use crate::cell::{Cell, RefCell};
 use crate::fmt;
 use crate::fs::File;
-use crate::io::{self, BorrowedCursor, BufReader, IoSlice, IoSliceMut, LineWriter, Lines};
+use crate::io::{
+    self, BorrowedCursor, BufReader, IoSlice, IoSliceMut, LineWriter, Lines, SpecReadByte,
+};
 use crate::sync::atomic::{AtomicBool, Ordering};
 use crate::sync::{Arc, Mutex, MutexGuard, OnceLock, ReentrantMutex, ReentrantMutexGuard};
 use crate::sys::stdio;
@@ -483,6 +485,13 @@ impl Read for StdinLock<'_> {
     }
 }
 
+impl SpecReadByte for StdinLock<'_> {
+    #[inline]
+    fn spec_read_byte(&mut self) -> Option<io::Result<u8>> {
+        BufReader::spec_read_byte(&mut *self.inner)
+    }
+}
+
 #[stable(feature = "rust1", since = "1.0.0")]
 impl BufRead for StdinLock<'_> {
     fn fill_buf(&mut self) -> io::Result<&[u8]> {