revert upper bound size hints

Signed-off-by: Connor Tsui <[email protected]>

Author: connortsui20

vortex-buffer/src/bit/buf_mut.rs

@@ -511,14 +511,18 @@ impl FromIterator<bool> for BitBufferMut {
     fn from_iter<T: IntoIterator<Item = bool>>(iter: T) -> Self {
         let mut iter = iter.into_iter();
 
-        // Note that these hints might be incorrect.
-        let (lower_bound, upper_bound_opt) = iter.size_hint();
-        let capacity = upper_bound_opt.unwrap_or(lower_bound);
+        // Since we do not know the length of the iterator, we can only guess how much memory we
+        // need to reserve. Note that these hints may be inaccurate.
+        let (lower_bound, _) = iter.size_hint();
 
-        let mut buf = BitBufferMut::new_unset(capacity);
+        // We choose not to use the optional upper bound size hint to match the standard library.
+
+        // Initialize all bits to 0 with the given length. By doing this, we only need to set bits
+        // that are true (and this is faster from benchmarks).
+        let mut buf = BitBufferMut::new_unset(lower_bound);
 
         // Directly write within our known capacity.
-        for i in 0..capacity {
+        for i in 0..buf.len() {
             let Some(v) = iter.next() else {
                 // SAFETY: We are definitely under the capacity and all values are already
                 // initialized from `new_unset`.

vortex-buffer/src/buffer.rs

@@ -193,9 +193,10 @@ impl<T> Buffer<T> {
     /// Create a buffer with values from the TrustedLen iterator.
     /// Should be preferred over `from_iter` when the iterator is known to be `TrustedLen`.
     pub fn from_trusted_len_iter<I: TrustedLen<Item = T>>(iter: I) -> Self {
-        let (_, high) = iter.size_hint();
-        let mut buffer =
-            BufferMut::with_capacity(high.vortex_expect("TrustedLen iterator has no upper bound"));
+        let (_, upper_bound) = iter.size_hint();
+        let mut buffer = BufferMut::with_capacity(
+            upper_bound.vortex_expect("TrustedLen iterator has no upper bound"),
+        );
         buffer.extend_trusted(iter);
         buffer.freeze()
     }

vortex-buffer/src/buffer_mut.rs

@@ -484,14 +484,11 @@ impl<T> BufferMut<T> {
     fn extend_iter(&mut self, mut iter: impl Iterator<Item = T>) {
         // Since we do not know the length of the iterator, we can only guess how much memory we
         // need to reserve. Note that these hints may be inaccurate.
-        let (lower_bound, upper_bound_opt) = iter.size_hint();
-
-        // In the case that the upper bound is adversarial, we put a hard limit on the amount of
-        // memory we reserve (and the OS should handle the rest with zero pages).
-        let reserve_amount = upper_bound_opt
-            .unwrap_or(lower_bound)
-            .min(i32::MAX as usize);
-        self.reserve(reserve_amount);
+        let (lower_bound, _) = iter.size_hint();
+
+        // We choose not to use the optional upper bound size hint to match the standard library.
+
+        self.reserve(lower_bound);
 
         let unwritten = self.capacity() - self.len();
 

vortex-buffer/src/trusted_len.rs

@@ -72,17 +72,17 @@ pub trait TrustedLenExt: Iterator + Sized {
     ///
     /// The caller must guarantee that the iterator does indeed have an exact length.
     unsafe fn trusted_len(self) -> TrustedLenAdapter<Self> {
-        let (lower, maybe_upper) = self.size_hint();
-        if let Some(upper) = maybe_upper {
+        let (lower_bound, upper_bound_opt) = self.size_hint();
+        if let Some(upper_bound) = upper_bound_opt {
             assert_eq!(
-                lower, upper,
+                lower_bound, upper_bound,
                 "TrustedLenExt: iterator size hints must match if upper bound is given"
             );
         }
 
         TrustedLenAdapter {
             inner: self,
-            len: lower,
+            len: lower_bound,
             #[cfg(debug_assertions)]
             count: 0,
         }

vortex-vector/src/bool/from_iter.rs

@@ -28,14 +28,12 @@ impl FromIterator<Option<bool>> for BoolVectorMut {
         let iter = iter.into_iter();
         // Since we do not know the length of the iterator, we can only guess how much memory we
         // need to reserve. Note that these hints may be inaccurate.
-        let (lower_bound, upper_bound_opt) = iter.size_hint();
+        let (lower_bound, _) = iter.size_hint();
 
-        // In the case that the upper bound is adversarial, we put a hard limit on the amount of
-        // memory we reserve (and the OS should handle the rest with zero pages).
-        let reserve_amount = upper_bound_opt.unwrap_or(lower_bound);
+        // We choose not to use the optional upper bound size hint to match the standard library.
 
-        let mut bits = BitBufferMut::with_capacity(reserve_amount);
-        let mut validity = MaskMut::with_capacity(reserve_amount);
+        let mut bits = BitBufferMut::with_capacity(lower_bound);
+        let mut validity = MaskMut::with_capacity(lower_bound);
 
         for opt_val in iter {
             match opt_val {

vortex-vector/src/primitive/from_iter.rs

@@ -29,16 +29,12 @@ impl<T: NativePType> FromIterator<Option<T>> for PVectorMut<T> {
         let iter = iter.into_iter();
         // Since we do not know the length of the iterator, we can only guess how much memory we
         // need to reserve. Note that these hints may be inaccurate.
-        let (lower_bound, upper_bound_opt) = iter.size_hint();
+        let (lower_bound, _) = iter.size_hint();
 
-        // In the case that the upper bound is adversarial, we put a hard limit on the amount of
-        // memory we reserve (and the OS should handle the rest with zero pages).
-        let reserve_amount = upper_bound_opt
-            .unwrap_or(lower_bound)
-            .min(i32::MAX as usize);
+        // We choose not to use the optional upper bound size hint to match the standard library.
 
-        let mut elements = BufferMut::with_capacity(reserve_amount);
-        let mut validity = MaskMut::with_capacity(reserve_amount);
+        let mut elements = BufferMut::with_capacity(lower_bound);
+        let mut validity = MaskMut::with_capacity(lower_bound);
 
         for opt_val in iter {
             match opt_val {