[smallvec] Optionally use SmallVec for some internal scratch vectors

We have some temporary vectors used during processing that will
usually only hold a few entries. By using SmallVec here, we can
avoid extra heap allocations in common cases.

(In Gecko's usage, I'm seeing around 4% improvement on the bidi-
resolution subtest of perf_reftest_singletons with this change.)

Cargo.toml

@@ -35,6 +35,7 @@ name = "unicode_bidi"
 flame = { version = "0.2", optional = true }
 flamer = { version = "0.4", optional = true }
 serde = { version = ">=0.8, <2.0", default-features = false, optional = true, features = ["derive"] }
+smallvec = { version = ">=1.13", optional = true, features = ["union"] }
 
 [dev-dependencies]
 serde_test = ">=0.8, <2.0"

src/implicit.rs

@@ -11,6 +11,8 @@
 
 use alloc::vec::Vec;
 use core::cmp::max;
+#[cfg(feature = "smallvec")]
+use smallvec::SmallVec;
 
 use super::char_data::BidiClass::{self, *};
 use super::level::Level;
@@ -39,7 +41,13 @@ pub fn resolve_weak<'a, T: TextSource<'a> + ?Sized>(
     // The previous class for the purposes of rule W1, not tracking changes from any other rules.
     let mut prev_class_before_w1 = sequence.sos;
     let mut last_strong_is_al = false;
+    #[cfg(feature = "smallvec")]
+    let mut et_run_indices = SmallVec::<[usize; 8]>::new(); // for W5
+    #[cfg(not(feature = "smallvec"))]
     let mut et_run_indices = Vec::new(); // for W5
+    #[cfg(feature = "smallvec")]
+    let mut bn_run_indices = SmallVec::<[usize; 8]>::new(); // for W5 +  <https://www.unicode.org/reports/tr9/#Retaining_Explicit_Formatting_Characters>
+    #[cfg(not(feature = "smallvec"))]
     let mut bn_run_indices = Vec::new(); // for W5 +  <https://www.unicode.org/reports/tr9/#Retaining_Explicit_Formatting_Characters>
 
     for (run_index, level_run) in sequence.runs.iter().enumerate() {
@@ -177,7 +185,7 @@ pub fn resolve_weak<'a, T: TextSource<'a> + ?Sized>(
                         _ => {
                             // <https://www.unicode.org/reports/tr9/#Retaining_Explicit_Formatting_Characters>
                             // If there was a BN run before this, that's now a part of this ET run.
-                            et_run_indices.extend(&bn_run_indices);
+                            et_run_indices.extend(bn_run_indices.clone());
 
                             // In case this is followed by an EN.
                             et_run_indices.push(i);
@@ -411,6 +419,9 @@ pub fn resolve_neutral<'a, D: BidiDataSource, T: TextSource<'a> + ?Sized>(
     let mut prev_class = sequence.sos;
     while let Some(mut i) = indices.next() {
         // Process sequences of NI characters.
+        #[cfg(feature = "smallvec")]
+        let mut ni_run = SmallVec::<[usize; 8]>::new();
+        #[cfg(not(feature = "smallvec"))]
         let mut ni_run = Vec::new();
         // The BN is for <https://www.unicode.org/reports/tr9/#Retaining_Explicit_Formatting_Characters>
         if is_NI(processing_classes[i]) || processing_classes[i] == BN {
@@ -484,7 +495,10 @@ fn identify_bracket_pairs<'a, T: TextSource<'a> + ?Sized, D: BidiDataSource>(
     original_classes: &[BidiClass],
 ) -> Vec<BracketPair> {
     let mut ret = vec![];
-    let mut stack = vec![];
+    #[cfg(feature = "smallvec")]
+    let mut stack = SmallVec::<[(char, usize, usize); 8]>::new();
+    #[cfg(not(feature = "smallvec"))]
+    let mut stack = Vec::new();
 
     for (run_index, level_run) in run_sequence.runs.iter().enumerate() {
         for (i, ch) in text.subrange(level_run.clone()).char_indices() {

src/lib.rs

@@ -71,6 +71,8 @@
 extern crate std;
 #[macro_use]
 extern crate alloc;
+#[cfg(feature = "smallvec")]
+extern crate smallvec;
 
 pub mod data_source;
 pub mod deprecated;
@@ -99,6 +101,8 @@ use core::cmp;
 use core::iter::repeat;
 use core::ops::Range;
 use core::str::CharIndices;
+#[cfg(feature = "smallvec")]
+use smallvec::SmallVec;
 
 use crate::format_chars as chars;
 use crate::BidiClass::*;
@@ -300,6 +304,9 @@ fn compute_initial_info<'a, D: BidiDataSource, T: TextSource<'a> + ?Sized>(
     let mut original_classes = Vec::with_capacity(text.len());
 
     // The stack contains the starting code unit index for each nested isolate we're inside.
+    #[cfg(feature = "smallvec")]
+    let mut isolate_stack = SmallVec::<[usize; 8]>::new();
+    #[cfg(not(feature = "smallvec"))]
     let mut isolate_stack = Vec::new();
 
     debug_assert!(