chromium/chromium-123-rust-clap_lex.patch

diff -Nur chromium-123.0.6312.46/third_party/rust/chromium_crates_io/vendor/clap_lex-0.7.0/src/ext.rs.me chromium-123.0.6312.46/third_party/rust/chromium_crates_io/vendor/clap_lex-0.7.0/src/ext.rs
--- chromium-123.0.6312.46/third_party/rust/chromium_crates_io/vendor/clap_lex-0.7.0/src/ext.rs.me	2024-03-13 20:36:17.000000000 +0100
+++ chromium-123.0.6312.46/third_party/rust/chromium_crates_io/vendor/clap_lex-0.7.0/src/ext.rs	2024-03-13 00:38:18.000000000 +0100
@@ -2,9 +2,6 @@
 
 pub trait OsStrExt: private::Sealed {
     /// Converts to a string slice.
-    ///
-    /// The Utf8Error is guaranteed to have a valid UTF8 boundary
-    /// in its `valid_up_to()`
     fn try_str(&self) -> Result<&str, std::str::Utf8Error>;
     /// Returns `true` if the given pattern matches a sub-slice of
     /// this string slice.
@@ -183,7 +180,7 @@
 
 impl OsStrExt for OsStr {
     fn try_str(&self) -> Result<&str, std::str::Utf8Error> {
-        let bytes = self.as_encoded_bytes();
+        let bytes = to_bytes(self);
         std::str::from_utf8(bytes)
     }
 
@@ -192,22 +189,22 @@
     }
 
     fn find(&self, needle: &str) -> Option<usize> {
-        let bytes = self.as_encoded_bytes();
+        let bytes = to_bytes(self);
         (0..=self.len().checked_sub(needle.len())?)
             .find(|&x| bytes[x..].starts_with(needle.as_bytes()))
     }
 
     fn strip_prefix(&self, prefix: &str) -> Option<&OsStr> {
-        let bytes = self.as_encoded_bytes();
+        let bytes = to_bytes(self);
         bytes.strip_prefix(prefix.as_bytes()).map(|s| {
             // SAFETY:
-            // - This came from `as_encoded_bytes`
-            // - Since `prefix` is `&str`, any split will be along UTF-8 boundary
-            unsafe { OsStr::from_encoded_bytes_unchecked(s) }
+            // - This came from `to_bytes`
+            // - Since `prefix` is `&str`, any split will be along UTF-8 boundarie
+            unsafe { to_os_str_unchecked(s) }
         })
     }
     fn starts_with(&self, prefix: &str) -> bool {
-        let bytes = self.as_encoded_bytes();
+        let bytes = to_bytes(self);
         bytes.starts_with(prefix.as_bytes())
     }
 
@@ -222,18 +219,13 @@
     fn split_once(&self, needle: &'_ str) -> Option<(&OsStr, &OsStr)> {
         let start = self.find(needle)?;
         let end = start + needle.len();
-        let haystack = self.as_encoded_bytes();
+        let haystack = to_bytes(self);
         let first = &haystack[0..start];
         let second = &haystack[end..];
         // SAFETY:
-        // - This came from `as_encoded_bytes`
-        // - Since `needle` is `&str`, any split will be along UTF-8 boundary
-        unsafe {
-            Some((
-                OsStr::from_encoded_bytes_unchecked(first),
-                OsStr::from_encoded_bytes_unchecked(second),
-            ))
-        }
+        // - This came from `to_bytes`
+        // - Since `needle` is `&str`, any split will be along UTF-8 boundarie
+        unsafe { Some((to_os_str_unchecked(first), to_os_str_unchecked(second))) }
     }
 }
 
@@ -243,6 +235,45 @@
     impl Sealed for std::ffi::OsStr {}
 }
 
+/// Allow access to raw bytes
+///
+/// As the non-UTF8 encoding is not defined, the bytes only make sense when compared with
+/// 7-bit ASCII or `&str`
+///
+/// # Compatibility
+///
+/// There is no guarantee how non-UTF8 bytes will be encoded, even within versions of this crate
+/// (since its dependent on rustc)
+fn to_bytes(s: &OsStr) -> &[u8] {
+    // SAFETY:
+    // - Lifetimes are the same
+    // - Types are compatible (`OsStr` is effectively a transparent wrapper for `[u8]`)
+    // - The primary contract is that the encoding for invalid surrogate code points is not
+    //   guaranteed which isn't a problem here
+    //
+    // There is a proposal to support this natively (https://github.com/rust-lang/rust/pull/95290)
+    // but its in limbo
+    unsafe { std::mem::transmute(s) }
+}
+
+/// Restore raw bytes as `OsStr`
+///
+/// # Safety
+///
+/// - `&[u8]` must either by a `&str` or originated with `to_bytes` within the same binary
+/// - Any splits of the original `&[u8]` must be done along UTF-8 boundaries
+unsafe fn to_os_str_unchecked(s: &[u8]) -> &OsStr {
+    // SAFETY:
+    // - Lifetimes are the same
+    // - Types are compatible (`OsStr` is effectively a transparent wrapper for `[u8]`)
+    // - The primary contract is that the encoding for invalid surrogate code points is not
+    //   guaranteed which isn't a problem here
+    //
+    // There is a proposal to support this natively (https://github.com/rust-lang/rust/pull/95290)
+    // but its in limbo
+    std::mem::transmute(s)
+}
+
 pub struct Split<'s, 'n> {
     haystack: Option<&'s OsStr>,
     needle: &'n str,
@@ -275,10 +306,7 @@
 ///
 /// `index` must be at a valid UTF-8 boundary
 pub(crate) unsafe fn split_at(os: &OsStr, index: usize) -> (&OsStr, &OsStr) {
-    let bytes = os.as_encoded_bytes();
+    let bytes = to_bytes(os);
     let (first, second) = bytes.split_at(index);
-    (
-        OsStr::from_encoded_bytes_unchecked(first),
-        OsStr::from_encoded_bytes_unchecked(second),
-    )
+    (to_os_str_unchecked(first), to_os_str_unchecked(second))
 }