Make `AVG_PART_LENGTH` a power of 2

I seem to recall that in general, it's best to request an allocation
with a size that's a power of 2. The low estimate of 5 was probably a
little too low as well.

diff --git a/src/librustdoc/html/url_parts_builder.rs b/src/librustdoc/html/url_parts_builder.rs
index 26cebe8c72beaa78e289b17ffc6e7877ea517b8e..66c879b83925c7d11b3627e391f6252b9debbaa0 100644 (file)
--- a/src/librustdoc/html/url_parts_builder.rs
+++ b/src/librustdoc/html/url_parts_builder.rs
@@ -105,10 +105,17 @@ fn with_capacity_bytes(count: usize) -> Self {
 
 /// This is just a guess at the average length of a URL part,
 /// used for [`String::with_capacity`] calls in the [`FromIterator`]
-/// and [`Extend`] impls.
+/// and [`Extend`] impls, and for [estimating item path lengths].
 ///
-/// This is intentionally on the lower end to avoid overallocating.
-const AVG_PART_LENGTH: usize = 5;
+/// The value `8` was chosen for two main reasons:
+///
+/// * It seems like a good guess for the average part length.
+/// * jemalloc's size classes are all multiples of eight,
+///   which means that the amount of memory it allocates will often match
+///   the amount requested, avoiding wasted bytes.
+///
+/// [estimating item path lengths]: estimate_item_path_byte_length
+const AVG_PART_LENGTH: usize = 8;
 
 /// Estimate the number of bytes in an item's path, based on how many segments it has.
 ///