Auto merge of #121428 - okaneco:ipaddr_parse, r=cuviper

net: Don't use checked arithmetic when parsing numbers with known max digits

Add a branch to `Parser::read_number` that determines whether checked or regular arithmetic is used.

- If `max_digits.is_some()`, then we know we are parsing a `u8` or `u16` because `read_number` is only called with `Some(3)` or `Some(4)`. Both types fit within a `u32` without risk of overflow. Thus, we can use plain arithmetic to avoid extra instructions from `checked_mul` and `checked_add`.

Add benches for `IpAddr`, `Ipv4Addr`, `Ipv6Addr`, `SocketAddr`, `SocketAddrV4`, and `SocketAddrV6` parsing

library/core/benches/lib.rs

@@ -16,6 +16,7 @@
 mod fmt;
 mod hash;
 mod iter;
+mod net;
 mod num;
 mod ops;
 mod pattern;

library/core/benches/net/addr_parser.rs

@@ -0,0 +1,78 @@
+use core::net::{IpAddr, Ipv4Addr, Ipv6Addr, SocketAddr, SocketAddrV4, SocketAddrV6};
+use core::str::FromStr;
+
+use test::{black_box, Bencher};
+
+const IPV4_STR: &str = "192.168.0.1";
+const IPV4_STR_PORT: &str = "192.168.0.1:8080";
+
+const IPV6_STR_FULL: &str = "2001:db8:0:0:0:0:c0a8:1";
+const IPV6_STR_COMPRESS: &str = "2001:db8::c0a8:1";
+const IPV6_STR_V4: &str = "2001:db8::192.168.0.1";
+const IPV6_STR_PORT: &str = "[2001:db8::c0a8:1]:8080";
+const IPV6_STR_PORT_SCOPE_ID: &str = "[2001:db8::c0a8:1%1337]:8080";
+
+#[bench]
+fn bench_parse_ipv4(b: &mut Bencher) {
+    b.iter(|| Ipv4Addr::from_str(black_box(IPV4_STR)));
+}
+
+#[bench]
+fn bench_parse_ipv6_full(b: &mut Bencher) {
+    b.iter(|| Ipv6Addr::from_str(black_box(IPV6_STR_FULL)));
+}
+
+#[bench]
+fn bench_parse_ipv6_compress(b: &mut Bencher) {
+    b.iter(|| Ipv6Addr::from_str(black_box(IPV6_STR_COMPRESS)));
+}
+
+#[bench]
+fn bench_parse_ipv6_v4(b: &mut Bencher) {
+    b.iter(|| Ipv6Addr::from_str(black_box(IPV6_STR_V4)));
+}
+
+#[bench]
+fn bench_parse_ipaddr_v4(b: &mut Bencher) {
+    b.iter(|| IpAddr::from_str(black_box(IPV4_STR)));
+}
+
+#[bench]
+fn bench_parse_ipaddr_v6_full(b: &mut Bencher) {
+    b.iter(|| IpAddr::from_str(black_box(IPV6_STR_FULL)));
+}
+
+#[bench]
+fn bench_parse_ipaddr_v6_compress(b: &mut Bencher) {
+    b.iter(|| IpAddr::from_str(black_box(IPV6_STR_COMPRESS)));
+}
+
+#[bench]
+fn bench_parse_ipaddr_v6_v4(b: &mut Bencher) {
+    b.iter(|| IpAddr::from_str(black_box(IPV6_STR_V4)));
+}
+
+#[bench]
+fn bench_parse_socket_v4(b: &mut Bencher) {
+    b.iter(|| SocketAddrV4::from_str(black_box(IPV4_STR_PORT)));
+}
+
+#[bench]
+fn bench_parse_socket_v6(b: &mut Bencher) {
+    b.iter(|| SocketAddrV6::from_str(black_box(IPV6_STR_PORT)));
+}
+
+#[bench]
+fn bench_parse_socket_v6_scope_id(b: &mut Bencher) {
+    b.iter(|| SocketAddrV6::from_str(black_box(IPV6_STR_PORT_SCOPE_ID)));
+}
+
+#[bench]
+fn bench_parse_socketaddr_v4(b: &mut Bencher) {
+    b.iter(|| SocketAddr::from_str(black_box(IPV4_STR_PORT)));
+}
+
+#[bench]
+fn bench_parse_socketaddr_v6(b: &mut Bencher) {
+    b.iter(|| SocketAddr::from_str(black_box(IPV6_STR_PORT)));
+}

library/core/benches/net/mod.rs

@@ -0,0 +1 @@
+mod addr_parser;

library/core/src/net/parser.rs

@@ -3,7 +3,7 @@
 //! This module is "publicly exported" through the `FromStr` implementations
 //! below.
 
-use crate::convert::TryInto;
+use crate::convert::{TryFrom, TryInto};
 use crate::error::Error;
 use crate::fmt;
 use crate::net::{IpAddr, Ipv4Addr, Ipv6Addr, SocketAddr, SocketAddrV4, SocketAddrV6};
@@ -104,12 +104,46 @@ fn read_separator<T, F>(&mut self, sep: char, index: usize, inner: F) -> Option<
     // Read a number off the front of the input in the given radix, stopping
     // at the first non-digit character or eof. Fails if the number has more
     // digits than max_digits or if there is no number.
-    fn read_number<T: ReadNumberHelper>(
+    //
+    // INVARIANT: `max_digits` must be less than the number of digits that `u32`
+    // can represent.
+    fn read_number<T: ReadNumberHelper + TryFrom<u32>>(
         &mut self,
         radix: u32,
         max_digits: Option<usize>,
         allow_zero_prefix: bool,
     ) -> Option<T> {
+        // If max_digits.is_some(), then we are parsing a `u8` or `u16` and
+        // don't need to use checked arithmetic since it fits within a `u32`.
+        if let Some(max_digits) = max_digits {
+            // u32::MAX = 4_294_967_295u32, which is 10 digits long.
+            // `max_digits` must be less than 10 to not overflow a `u32`.
+            debug_assert!(max_digits < 10);
+
+            self.read_atomically(move |p| {
+                let mut result = 0_u32;
+                let mut digit_count = 0;
+                let has_leading_zero = p.peek_char() == Some('0');
+
+                while let Some(digit) = p.read_atomically(|p| p.read_char()?.to_digit(radix)) {
+                    result *= radix;
+                    result += digit;
+                    digit_count += 1;
+
+                    if digit_count > max_digits {
+                        return None;
+                    }
+                }
+
+                if digit_count == 0 {
+                    None
+                } else if !allow_zero_prefix && has_leading_zero && digit_count > 1 {
+                    None
+                } else {
+                    result.try_into().ok()
+                }
+            })
+        } else {
             self.read_atomically(move |p| {
                 let mut result = T::ZERO;
                 let mut digit_count = 0;
@@ -119,11 +153,6 @@ fn read_number<T: ReadNumberHelper>(
                     result = result.checked_mul(radix)?;
                     result = result.checked_add(digit)?;
                     digit_count += 1;
-                if let Some(max_digits) = max_digits {
-                    if digit_count > max_digits {
-                        return None;
-                    }
-                }
                 }
 
                 if digit_count == 0 {
@@ -135,6 +164,7 @@ fn read_number<T: ReadNumberHelper>(
                 }
             })
         }
+    }
 
     /// Read an IPv4 address.
     fn read_ipv4_addr(&mut self) -> Option<Ipv4Addr> {