Speed up the lexer for Ruby 3.4+ (#1832)

* Speed up lexing

* Bump msrv to 3.0 (from 2.7)

* Normalize test for ruby-head compat

* Fix bug when parsing negative numbers

.github/workflows/liquid.yml

@@ -11,10 +11,10 @@ jobs:
     strategy:
       matrix:
         entry:
-          - { ruby: 2.7, allowed-failure: false } # minimum supported
+          - { ruby: 3.0, allowed-failure: false } # minimum supported
           - { ruby: 3.2, allowed-failure: false }
           - { ruby: 3.3, allowed-failure: false } # latest
-          - { ruby: ruby-head, allowed-failure: true }
+          - { ruby: ruby-head, allowed-failure: false }
     name: Test Ruby ${{ matrix.entry.ruby }}
     steps:
       - uses: actions/checkout@v3

.gitignore

@@ -6,3 +6,4 @@ pkg
 .rvmrc
 .bundle
 .byebug_history
+Gemfile.lock

.ruby-version

@@ -1 +1 @@
-3.3.0
+3.3.4

Rakefile

@@ -81,6 +81,14 @@ namespace :benchmark do
   task :strict do
     ruby "./performance/benchmark.rb strict"
   end
+
+  desc "Run unit benchmarks"
+  task :unit do
+    Dir["./performance/unit/*_benchmark.rb"].each do |file|
+      puts "🧪 Running #{file}"
+      ruby file
+    end
+  end
 end
 
 namespace :profile do

lib/liquid/lexer.rb

@@ -1,8 +1,9 @@
 # frozen_string_literal: true
 
 require "strscan"
+
 module Liquid
-  class Lexer
+  class Lexer1
     SPECIALS = {
       '|' => :pipe,
       '.' => :dot,
@@ -58,4 +59,157 @@ def tokenize
       @output << [:end_of_string]
     end
   end
+
+  class Lexer2
+    CLOSE_ROUND = [:close_round, ")"].freeze
+    CLOSE_SQUARE = [:close_square, "]"].freeze
+    COLON = [:colon, ":"].freeze
+    COMMA = [:comma, ","].freeze
+    COMPARISION_NOT_EQUAL = [:comparison, "!="].freeze
+    COMPARISON_CONTAINS = [:comparison, "contains"].freeze
+    COMPARISON_EQUAL = [:comparison, "=="].freeze
+    COMPARISON_GREATER_THAN = [:comparison, ">"].freeze
+    COMPARISON_GREATER_THAN_OR_EQUAL = [:comparison, ">="].freeze
+    COMPARISON_LESS_THAN = [:comparison, "<"].freeze
+    COMPARISON_LESS_THAN_OR_EQUAL = [:comparison, "<="].freeze
+    COMPARISON_NOT_EQUAL_ALT = [:comparison, "<>"].freeze
+    CONTAINS = /contains(?=\s)/
+    DASH = [:dash, "-"].freeze
+    DOT = [:dot, "."].freeze
+    DOTDOT = [:dotdot, ".."].freeze
+    DOT_ORD = ".".ord
+    DOUBLE_STRING_LITERAL = /"[^\"]*"/
+    EOS = [:end_of_string].freeze
+    IDENTIFIER            = /[a-zA-Z_][\w-]*\??/
+    NUMBER_LITERAL        = /-?\d+(\.\d+)?/
+    OPEN_ROUND = [:open_round, "("].freeze
+    OPEN_SQUARE = [:open_square, "["].freeze
+    PIPE = [:pipe, "|"].freeze
+    QUESTION = [:question, "?"].freeze
+    RUBY_WHITESPACE = [" ", "\t", "\r", "\n", "\f"].freeze
+    SINGLE_STRING_LITERAL = /'[^\']*'/
+    WHITESPACE_OR_NOTHING = /\s*/
+
+    COMPARISON_JUMP_TABLE = [].tap do |table|
+      table["=".ord] = [].tap do |sub_table|
+        sub_table["=".ord] = COMPARISON_EQUAL
+        sub_table.freeze
+      end
+      table["!".ord] = [].tap do |sub_table|
+        sub_table["=".ord] = COMPARISION_NOT_EQUAL
+        sub_table.freeze
+      end
+      table["<".ord] = [].tap do |sub_table|
+        sub_table["=".ord] = COMPARISON_LESS_THAN_OR_EQUAL
+        sub_table[">".ord] = COMPARISON_NOT_EQUAL_ALT
+        RUBY_WHITESPACE.each { |c| sub_table[c.ord] = COMPARISON_LESS_THAN }
+        sub_table.freeze
+      end
+      table[">".ord] = [].tap do |sub_table|
+        sub_table["=".ord] = COMPARISON_GREATER_THAN_OR_EQUAL
+        RUBY_WHITESPACE.each { |c| sub_table[c.ord] = COMPARISON_GREATER_THAN }
+        sub_table.freeze
+      end
+      table.freeze
+    end
+
+    NEXT_MATCHER_JUMP_TABLE = [].tap do |table|
+      "a".upto("z") do |c|
+        table[c.ord] = [:id, IDENTIFIER].freeze
+        table[c.upcase.ord] = [:id, IDENTIFIER].freeze
+      end
+      table["_".ord] = [:id, IDENTIFIER].freeze
+
+      "0".upto("9") do |c|
+        table[c.ord] = [:number, NUMBER_LITERAL].freeze
+      end
+      table["-".ord] = [:number, NUMBER_LITERAL].freeze
+
+      table["'".ord] = [:string, SINGLE_STRING_LITERAL].freeze
+      table["\"".ord] = [:string, DOUBLE_STRING_LITERAL].freeze
+      table.freeze
+    end
+
+    SPECIAL_TABLE = [].tap do |table|
+      table["|".ord] = PIPE
+      table[".".ord] = DOT
+      table[":".ord] = COLON
+      table[",".ord] = COMMA
+      table["[".ord] = OPEN_SQUARE
+      table["]".ord] = CLOSE_SQUARE
+      table["(".ord] = OPEN_ROUND
+      table[")".ord] = CLOSE_ROUND
+      table["?".ord] = QUESTION
+      table["-".ord] = DASH
+    end
+
+    NUMBER_TABLE = [].tap do |table|
+      "0".upto("9") do |c|
+        table[c.ord] = true
+      end
+      table.freeze
+    end
+
+    def initialize(input)
+      @ss = StringScanner.new(input)
+    end
+
+    # rubocop:disable Metrics/BlockNesting
+    def tokenize
+      @output = []
+
+      until @ss.eos?
+        @ss.skip(WHITESPACE_OR_NOTHING)
+
+        break if @ss.eos?
+
+        peeked = @ss.peek_byte
+
+        if (special = SPECIAL_TABLE[peeked])
+          @ss.scan_byte
+          # Special case for ".."
+          if special == DOT && @ss.peek_byte == DOT_ORD
+            @ss.scan_byte
+            @output << DOTDOT
+          elsif special == DASH
+            # Special case for negative numbers
+            if NUMBER_TABLE[@ss.peek_byte]
+              @ss.pos -= 1
+              @output << [:number, @ss.scan(NUMBER_LITERAL)]
+            else
+              @output << special
+            end
+          else
+            @output << special
+          end
+        elsif (sub_table = COMPARISON_JUMP_TABLE[peeked])
+          @ss.scan_byte
+          if (found = sub_table[@ss.peek_byte])
+            @output << found
+            @ss.scan_byte
+          else
+            raise SyntaxError, "Unexpected character #{peeked.chr}"
+          end
+        else
+          type, pattern = NEXT_MATCHER_JUMP_TABLE[peeked]
+
+          if type && (t = @ss.scan(pattern))
+            # Special case for "contains"
+            @output << if type == :id && t == "contains"
+              COMPARISON_CONTAINS
+            else
+              [type, t]
+            end
+          else
+            raise SyntaxError, "Unexpected character #{peeked.chr}"
+          end
+        end
+      end
+      # rubocop:enable Metrics/BlockNesting
+
+      @output << EOS
+    end
+  end
+
+  Lexer = StringScanner.instance_methods.include?(:scan_byte) ? Lexer2 : Lexer1
 end

lib/liquid/parser.rb

@@ -53,7 +53,7 @@ def expression
         str = consume
         str << variable_lookups
       when :open_square
-        str = consume
+        str = consume.dup
         str << expression
         str << consume(:close_square)
         str << variable_lookups

liquid.gemspec

@@ -17,7 +17,7 @@ Gem::Specification.new do |s|
   s.license     = "MIT"
   # s.description = "A secure, non-evaling end user template engine with aesthetic markup."
 
-  s.required_ruby_version     = ">= 2.7.0"
+  s.required_ruby_version     = ">= 3.0.0"
   s.required_rubygems_version = ">= 1.3.7"
 
   s.metadata['allowed_push_host'] = 'https://rubygems.org'
@@ -28,6 +28,9 @@ Gem::Specification.new do |s|
 
   s.require_path = "lib"
 
+  s.add_dependency("strscan")
+  s.add_dependency("bigdecimal")
+
   s.add_development_dependency('rake', '~> 13.0')
   s.add_development_dependency('minitest')
 end

performance/benchmark.rb

@@ -3,6 +3,7 @@
 require 'benchmark/ips'
 require_relative 'theme_runner'
 
+RubyVM::YJIT.enable if defined?(RubyVM::YJIT)
 Liquid::Template.error_mode = ARGV.first.to_sym if ARGV.first
 profiler = ThemeRunner.new
 

performance/unit/lexer_benchmark.rb

@@ -0,0 +1,62 @@
+# frozen_string_literal: true
+
+require "benchmark/ips"
+
+# benchmark liquid lexing
+
+require 'liquid'
+
+RubyVM::YJIT.enable
+
+EXPRESSIONS = [
+  "foo[1..2].baz",
+  "12.0",
+  "foo.bar.based",
+  "21 - 62",
+  "foo.bar.baz",
+  "foo > 12",
+  "foo < 12",
+  "foo <= 12",
+  "foo >= 12",
+  "foo <> 12",
+  "foo == 12",
+  "foo != 12",
+  "foo contains 12",
+  "foo contains 'bar'",
+  "foo != 'bar'",
+  "'foo' contains 'bar'",
+  '234089',
+  "foo | default: -1",
+]
+
+EXPRESSIONS.each do |expr|
+  lexer_1_result = Liquid::Lexer1.new(expr).tokenize
+  lexer_2_result = Liquid::Lexer2.new(expr).tokenize
+
+  next if lexer_1_result == lexer_2_result
+
+  warn "Lexer1 and Lexer2 results are different for expression: #{expr}"
+  warn "expected: #{lexer_1_result}"
+  warn "got: #{lexer_2_result}"
+  abort
+end
+
+Benchmark.ips do |x|
+  x.config(time: 10, warmup: 5)
+
+  x.report("Liquid::Lexer1#tokenize") do
+    EXPRESSIONS.each do |expr|
+      l = Liquid::Lexer1.new(expr)
+      l.tokenize
+    end
+  end
+
+  x.report("Liquid::Lexer2#tokenize") do
+    EXPRESSIONS.each do |expr|
+      l = Liquid::Lexer2.new(expr)
+      l.tokenize
+    end
+  end
+
+  x.compare!
+end

test/integration/standard_filter_test.rb

@@ -32,7 +32,7 @@ def initialize(value:)
   attr_reader :value
 
   def registers
-    { @value => @context.registers[@value] }
+    "{#{@value.inspect}=>#{@context.registers[@value].inspect}}"
   end
 end