From 0cbd210c3a3a9f2ed603f704d915ffc8dd51bc7f Mon Sep 17 00:00:00 2001
From: Earlopain <14981592+Earlopain@users.noreply.github.com>
Date: Mon, 20 Apr 2026 19:16:56 +0200
Subject: [PATCH] [ruby/prism] Optimize ripper bounds

Basically a port of https://github.com/ruby/ruby/commit/c45f781771314a71856c9b348c640ba532f54349 into ruby

It's quite effective at ~97% hit rate for me.
Speeds it up from ~6.77x slower to only 4.07x slower.

For the lexer `on_sp` it also gives a bit of an improvement:
1.04x slower to 1.10x faster

I guess the class may be universally useful but for now I just made it nodoc.

https://github.com/ruby/prism/commit/3ad9db38fe
---
 lib/prism/lex_compat.rb         | 11 +++++-
 lib/prism/parse_result.rb       |  4 +-
 lib/prism/translation/ripper.rb | 69 ++++++++++++++++++++++++++++++---
 3 files changed, 74 insertions(+), 10 deletions(-)

diff --git a/lib/prism/lex_compat.rb b/lib/prism/lex_compat.rb
index e1b04fc6ce..7aacec037d 100644
--- a/lib/prism/lex_compat.rb
+++ b/lib/prism/lex_compat.rb
@@ -23,6 +23,12 @@ module Prism
   #            def self.[]: (Integer value) -> State
   #          end
   #        end
+  #
+  #        class LineAndColumnCache
+  #          def initialize: (Source source) -> void
+  #
+  #          def line_and_column: (Integer byte_offset) -> [Integer, Integer]
+  #        end
   #      end
   #    end
 
@@ -837,6 +843,8 @@ module Prism
       prev_token_state = Translation::Ripper::Lexer::State[Translation::Ripper::EXPR_BEG]
       prev_token_end = bom ? 3 : 0
 
+      cache = Translation::Ripper::LineAndColumnCache.new(source)
+
       tokens.each do |token|
         # Skip missing heredoc ends.
         next if token[1] == :on_heredoc_end && token[2] == ""
@@ -851,8 +859,7 @@ module Prism
 
         if start_offset > prev_token_end
           sp_value = source.slice(prev_token_end, start_offset - prev_token_end)
-          sp_line = source.line(prev_token_end)
-          sp_column = source.column(prev_token_end)
+          sp_line, sp_column = cache.line_and_column(prev_token_end)
           # Ripper reports columns on line 1 without counting the BOM
           sp_column -= 3 if sp_line == 1 && bom
           continuation_index = sp_value.byteindex("\\")
diff --git a/lib/prism/parse_result.rb b/lib/prism/parse_result.rb
index 7cf6630f44..93d3c006b7 100644
--- a/lib/prism/parse_result.rb
+++ b/lib/prism/parse_result.rb
@@ -225,9 +225,7 @@ module Prism
       freeze
     end
 
-    private
-
-    # Binary search through the offsets to find the line number for the given
+    # Binary search through the offsets to find the index for the given
     # byte offset.
     #--
     #: (Integer byte_offset) -> Integer
diff --git a/lib/prism/translation/ripper.rb b/lib/prism/translation/ripper.rb
index 77ee2c337d..cbbc0b5db8 100644
--- a/lib/prism/translation/ripper.rb
+++ b/lib/prism/translation/ripper.rb
@@ -445,6 +445,64 @@ module Prism
       autoload :SexpBuilder, "prism/translation/ripper/sexp"
       autoload :SexpBuilderPP, "prism/translation/ripper/sexp"
 
+      # Provides optimized access to line and column information.
+      # Ripper bounds are mostly accessed in a linear fashion, so
+      # we can try a linear scan first and fall back to binary search.
+      class LineAndColumnCache # :nodoc:
+        # How many should it look ahead/behind before falling back to binary searching.
+        WINDOW = 8
+        private_constant :WINDOW
+
+        #: (Source source) -> void
+        def initialize(source)
+          @source = source
+          @offsets = source.offsets
+          @hint = 0
+        end
+
+        #: (Integer byte_offset) -> [Integer, Integer]
+        def line_and_column(byte_offset)
+          @hint = new_hint(byte_offset) || @source.find_line(byte_offset)
+          return [@hint + @source.start_line, byte_offset - @offsets[@hint]]
+        end
+
+        private
+
+        def new_hint(byte_offset)
+          if @offsets[@hint] <= byte_offset
+            # Same line?
+            if (@hint + 1 >= @offsets.size || @offsets[@hint + 1] > byte_offset)
+              return @hint
+            end
+
+            # Scan forwards
+            limit = [@hint + WINDOW + 1, @offsets.size].min
+            idx = @hint + 1
+            while idx < limit
+              if @offsets[idx] > byte_offset
+                return idx - 1
+              end
+              if @offsets[idx] == byte_offset
+                return idx
+              end
+              idx += 1
+            end
+          else
+            # Scan backwards
+            limit = @hint > WINDOW ? @hint - WINDOW : 0
+            idx = @hint
+            while idx >= limit + 1
+              if @offsets[idx - 1] <= byte_offset
+                return idx - 1
+              end
+              idx -= 1
+            end
+          end
+
+          nil
+        end
+      end
+
       # :stopdoc:
       # This is not part of the public API but used by some gems.
 
@@ -488,6 +546,7 @@ module Prism
         @lineno = lineno
         @column = 0
         @result = nil
+        @line_and_column_cache = nil
       end
 
       ##########################################################################
@@ -4033,6 +4092,10 @@ module Prism
         @result ||= Prism.parse(source, partial_script: true, version: "current")
       end
 
+      def line_and_column_cache
+        @line_and_column_cache ||= LineAndColumnCache.new(result.source)
+      end
+
       ##########################################################################
       # Helpers
       ##########################################################################
@@ -4133,12 +4196,8 @@ module Prism
 
       # This method is responsible for updating lineno and column information
       # to reflect the current node.
-      #
-      # This method could be drastically improved with some caching on the start
-      # of every line, but for now it's good enough.
       def bounds(location)
-        @lineno = location.start_line
-        @column = location.start_column
+        @lineno, @column = line_and_column_cache.line_and_column(location.start_offset)
       end
 
       # :startdoc: