Auto merge of rust-lang#91352 - nnethercote:RawVec-reserve_for_push, …

…r=dtolnay

Introduce `RawVec::reserve_for_push`.

If `Vec::push`'s capacity check fails it calls `RawVec::reserve`, which
then also does a capacity check.

This commit introduces `reserve_for_push` which skips the redundant
capacity check, for some slight compile time speed-ups.

I tried lots of minor variations on this, e.g. different inlining
attributes. This was the best one I could find.

r? `@ghost`

library/alloc/src/raw_vec.rs

@@ -289,6 +289,14 @@ impl<T, A: Allocator> RawVec<T, A> {
         }
     }
 
+    /// A specialized version of `reserve()` used only by the hot and
+    /// oft-instantiated `Vec::push()`, which does its own capacity check.
+    #[cfg(not(no_global_oom_handling))]
+    #[inline(never)]
+    pub fn reserve_for_push(&mut self, len: usize) {
+        handle_reserve(self.grow_amortized(len, 1));
+    }
+
     /// The same as `reserve`, but returns on errors instead of panicking or aborting.
     pub fn try_reserve(&mut self, len: usize, additional: usize) -> Result<(), TryReserveError> {
         if self.needs_to_grow(len, additional) {

library/alloc/src/vec/mod.rs

@@ -1726,7 +1726,7 @@ impl<T, A: Allocator> Vec<T, A> {
         // This will panic or abort if we would allocate > isize::MAX bytes
         // or if the length increment would overflow for zero-sized types.
         if self.len == self.buf.capacity() {
-            self.reserve(1);
+            self.buf.reserve_for_push(self.len);
         }
         unsafe {
             let end = self.as_mut_ptr().add(self.len);