From 6ade73138cad3fc2a158205241708773da4732fc Mon Sep 17 00:00:00 2001
From: Wilfred Hughes <me@wilfred.me.uk>
Date: Wed, 15 Jun 2022 01:07:25 +0100
Subject: [PATCH] Skip printed lines when computing hunk matched lines

This function is hot for textual diffs, such as the sample files in
issue #293. For those files, this produces a 21% reduction in
instruction count.
---
 src/display/hunks.rs        | 16 ++++++++++++----
 src/display/side_by_side.rs | 13 +++++++++++--
 2 files changed, 23 insertions(+), 6 deletions(-)

diff --git a/src/display/hunks.rs b/src/display/hunks.rs
index 1e5d1e578a..6312035754 100644
--- a/src/display/hunks.rs
+++ b/src/display/hunks.rs
@@ -595,10 +595,10 @@ fn either_side_equal(
     false
 }
 
-pub fn matched_lines_for_hunk<'a>(
-    matched_lines: &'a [(Option<LineNumber>, Option<LineNumber>)],
+pub fn matched_lines_indexes_for_hunk(
+    matched_lines: &[(Option<LineNumber>, Option<LineNumber>)],
     hunk: &Hunk,
-) -> &'a [(Option<LineNumber>, Option<LineNumber>)] {
+) -> (usize, usize) {
     let mut hunk_lhs_novel = hunk.novel_lhs.iter().copied().collect::<Vec<_>>();
     hunk_lhs_novel.sort();
 
@@ -645,7 +645,7 @@ pub fn matched_lines_for_hunk<'a>(
         end_i = matched_lines.len();
     }
 
-    &matched_lines[start_i..end_i]
+    (start_i, end_i)
 }
 
 #[cfg(test)]
@@ -758,6 +758,14 @@ mod tests {
         assert_eq!(hunks[0].lines, vec![(Some(0.into()), Some(0.into()))]);
     }
 
+    fn matched_lines_for_hunk<'a>(
+        matched_lines: &'a [(Option<LineNumber>, Option<LineNumber>)],
+        hunk: &Hunk,
+    ) -> &'a [(Option<LineNumber>, Option<LineNumber>)] {
+        let (start_i, end_i) = matched_lines_indexes_for_hunk(matched_lines, hunk);
+        &matched_lines[start_i..end_i]
+    }
+
     #[test]
     fn test_matched_lines_for_hunk() {
         let matched_lines = &[
diff --git a/src/display/side_by_side.rs b/src/display/side_by_side.rs
index eac5c212bd..0368f517a7 100644
--- a/src/display/side_by_side.rs
+++ b/src/display/side_by_side.rs
@@ -7,7 +7,7 @@ use std::{cmp::max, collections::HashSet};
 use crate::{
     constants::Side,
     display::context::all_matched_lines_filled,
-    display::hunks::{matched_lines_for_hunk, Hunk},
+    display::hunks::{matched_lines_indexes_for_hunk, Hunk},
     display::style::{
         self, apply_colors, color_positions, novel_style, split_and_apply, BackgroundColor,
     },
@@ -385,6 +385,7 @@ pub fn print(
     let mut prev_rhs_line_num = None;
 
     let matched_lines = all_matched_lines_filled(lhs_mps, rhs_mps, &lhs_lines, &rhs_lines);
+    let mut matched_lines_to_print = &matched_lines[..];
 
     for (i, hunk) in hunks.iter().enumerate() {
         println!(
@@ -399,7 +400,15 @@ pub fn print(
             )
         );
 
-        let aligned_lines = matched_lines_for_hunk(&matched_lines, hunk);
+        let (start_i, end_i) = matched_lines_indexes_for_hunk(matched_lines_to_print, hunk);
+        let aligned_lines = &matched_lines_to_print[start_i..end_i];
+        // We iterate through hunks in order, so we know the next hunk
+        // must appear after start_i. This makes
+        // `matched_lines_indexes_for_hunk` faster on later
+        // iterations, and this function is hot on large textual
+        // diffs.
+        matched_lines_to_print = &matched_lines_to_print[start_i..];
+
         let no_lhs_changes = hunk.novel_lhs.is_empty();
         let no_rhs_changes = hunk.novel_rhs.is_empty();
         let same_lines = aligned_lines.iter().all(|(l, r)| l == r);