Improved performance.

src/io/parquet/read/binary/basic.rs

@@ -214,19 +214,23 @@ fn read_plain_optional<O: Offset>(
 
 pub(super) fn read_plain_required<O: Offset>(
     buffer: &[u8],
-    _length: usize,
+    additional: usize,
     offsets: &mut MutableBuffer<O>,
     values: &mut MutableBuffer<u8>,
 ) {
     let mut last_offset = *offsets.as_mut_slice().last().unwrap();
 
     let values_iterator = utils::BinaryIter::new(buffer);
 
+    // each value occupies 4 bytes + len declared in 4 bytes => reserve accordingly.
+    values.reserve(buffer.len() - 4 * additional);
+    let a = values.capacity();
     for value in values_iterator {
         last_offset += O::from_usize(value.len()).unwrap();
         values.extend_from_slice(value);
         offsets.push(last_offset);
     }
+    debug_assert_eq!(a, values.capacity());
 }
 
 pub(super) fn extend_from_page<O: Offset>(

src/io/parquet/read/utils.rs

@@ -1,4 +1,6 @@
-use parquet2::encoding::{get_length, Encoding};
+use std::convert::TryInto;
+
+use parquet2::encoding::Encoding;
 use parquet2::metadata::ColumnDescriptor;
 use parquet2::page::{split_buffer as _split_buffer, DataPage, DataPageHeader};
 
@@ -22,7 +24,7 @@ impl<'a> Iterator for BinaryIter<'a> {
         if self.values.is_empty() {
             return None;
         }
-        let length = get_length(self.values) as usize;
+        let length = u32::from_le_bytes(self.values[0..4].try_into().unwrap()) as usize;
         self.values = &self.values[4..];
         let result = &self.values[..length];
         self.values = &self.values[length..];