Add convenience methods for getting a struct from a ruby object.

If we are trying to get the struct from something other than self, then we
should make sure to check the class of the object.  This util functions
make this easier.

.gitignore

@@ -6,3 +6,5 @@ pkg
 .rvmrc
 .ruby-version
 Gemfile.lock
+*.bundle
+/tmp

Rakefile

@@ -1,5 +1,6 @@
 require 'rake'
 require 'rake/testtask'
+require 'rake/extensiontask'
 $LOAD_PATH.unshift File.expand_path("../lib", __FILE__)
 require "liquid/version"
 
@@ -75,3 +76,8 @@ desc "Run example"
 task :example do
   ruby "-w -d -Ilib example/server/server.rb"
 end
+
+Rake::ExtensionTask.new "liquid" do |ext|
+  ext.lib_dir = "lib/liquid"
+end
+Rake::Task[:test].prerequisites << :compile

ext/liquid/block.c

@@ -0,0 +1,168 @@
+#include "liquid_ext.h"
+
+VALUE cLiquidBlock;
+ID intern_assert_missing_delimitation, intern_block_delimiter, intern_is_blank, intern_new,
+   intern_new_with_options, intern_tags, intern_unknown_tag, intern_unterminated_tag,
+   intern_unterminated_variable;
+
+struct liquid_tag
+{
+    char *name, *markup;
+    long name_length, markup_length;
+};
+
+static bool parse_tag(struct liquid_tag *tag, char *token, long token_length)
+{
+    // Strip {{ and }} braces
+    token += 2;
+    token_length -= 4;
+
+    char *end = token + token_length;
+    while (token < end && isspace(*token))
+        token++;
+    tag->name = token;
+
+    char c = *token;
+    while (token < end && (isalnum(c) || c == '_'))
+        c = *(++token);
+    tag->name_length = token - tag->name;
+    if (!tag->name_length) {
+        memset(tag, 0, sizeof(*tag));
+        return false;
+    }
+
+    while (token < end && isspace(*token))
+        token++;
+    tag->markup = token;
+
+    char *last = end - 1;
+    while (token < last && isspace(*last))
+        last--;
+    end = last + 1;
+    tag->markup_length = end - token;
+    return true;
+}
+
+static VALUE rb_parse_body(VALUE self, VALUE tokenizerObj)
+{
+    struct liquid_tokenizer *tokenizer = LIQUID_TOKENIZER_GET_STRUCT(tokenizerObj);
+
+    bool blank = true;
+    VALUE nodelist = rb_iv_get(self, "@nodelist");
+    if (nodelist == Qnil) {
+        nodelist = rb_ary_new();
+        rb_iv_set(self, "@nodelist", nodelist);
+    } else {
+        rb_ary_clear(nodelist);
+    }
+
+    struct token token;
+    while (true) {
+        liquid_tokenizer_next(tokenizer, &token);
+        switch (token.type) {
+        case TOKEN_NONE:
+            /*
+             * Make sure that it's ok to end parsing in the current block.
+             * Effectively this method will throw an exception unless the current block is
+             * of type Document
+             */
+            rb_funcall(self, intern_assert_missing_delimitation, 0);
+            goto done;
+        case TOKEN_INVALID:
+        {
+            VALUE token_obj = rb_str_new(token.str, token.length);
+            if (token.str[1] == '%')
+                rb_funcall(self, intern_unterminated_tag, 1, token_obj);
+            else
+                rb_funcall(self, intern_unterminated_variable, 1, token_obj);
+            break;
+        }
+        case TOKEN_TAG:
+        {
+            struct liquid_tag tag;
+            if (!parse_tag(&tag, token.str, token.length)) {
+                // FIXME: provide more appropriate error message
+                rb_funcall(self, intern_unterminated_tag, 1, rb_str_new(token.str, token.length));
+            } else {
+                if (tag.name_length >= 3 && !memcmp(tag.name, "end", 3)) {
+                    VALUE block_delimiter = rb_funcall(self, intern_block_delimiter, 0);
+                    if (TYPE(block_delimiter) == T_STRING &&
+                        tag.name_length == RSTRING_LEN(block_delimiter) &&
+                        !memcmp(tag.name, RSTRING_PTR(block_delimiter), tag.name_length))
+                    {
+                        goto done;
+                    }
+                }
+
+                VALUE tags = rb_funcall(cLiquidTemplate, intern_tags, 0);
+                Check_Type(tags, T_HASH);
+                VALUE tag_name = rb_str_new(tag.name, tag.name_length);
+                VALUE tag_class = rb_hash_lookup(tags, tag_name);
+                VALUE markup = rb_str_new(tag.markup, tag.markup_length);
+                if (tag_class != Qnil) {
+                    VALUE options = rb_iv_get(self, "@options");
+                    if (options == Qnil)
+                        options = rb_hash_new();
+                    VALUE new_tag = rb_funcall(tag_class, intern_new_with_options, 4,
+                                               tag_name, markup, tokenizerObj, options);
+                    if (blank) {
+                        VALUE blank_block = rb_funcall(new_tag, intern_is_blank, 0);
+                        if (blank_block == Qnil || blank_block == Qfalse)
+                            blank = false;
+                    }
+                    rb_ary_push(nodelist, new_tag);
+                } else {
+                    rb_funcall(self, intern_unknown_tag, 3, tag_name, markup, tokenizerObj);
+                    /*
+                     * multi-block tags may store the nodelist in a block array on unknown_tag
+                     * then replace @nodelist with a new array. We need to use the new array
+                     * for the block following the tag token.
+                     */
+                    nodelist = rb_iv_get(self, "@nodelist");
+                }
+            }
+            break;
+        }
+        case TOKEN_VARIABLE:
+        {
+            VALUE markup = rb_str_new(token.str + 2, token.length - 4);
+            VALUE options = rb_iv_get(self, "@options");
+            VALUE new_var = rb_funcall(cLiquidVariable, intern_new, 2, markup, options);
+            rb_ary_push(nodelist, new_var);
+            blank = false;
+            break;
+        }
+        case TOKEN_STRING:
+            rb_ary_push(nodelist, rb_str_new(token.str, token.length));
+            if (blank) {
+                int i;
+                for (i = 0; i < token.length; i++) {
+                    if (!isspace(token.str[i])) {
+                        blank = false;
+                        break;
+                    }
+                }
+            }
+            break;
+        }
+    }
+done:
+    rb_iv_set(self, "@blank", blank ? Qtrue : Qfalse);
+    return Qnil;
+}
+
+void init_liquid_block()
+{
+    intern_assert_missing_delimitation = rb_intern("assert_missing_delimitation!");
+    intern_block_delimiter = rb_intern("block_delimiter");
+    intern_is_blank = rb_intern("blank?");
+    intern_new = rb_intern("new");
+    intern_new_with_options = rb_intern("new_with_options");
+    intern_tags = rb_intern("tags");
+    intern_unknown_tag = rb_intern("unknown_tag");
+    intern_unterminated_tag = rb_intern("unterminated_tag");
+    intern_unterminated_variable = rb_intern("unterminated_variable");
+
+    cLiquidBlock = rb_define_class_under(mLiquid, "Block", cLiquidTag);
+    rb_define_method(cLiquidBlock, "parse_body", rb_parse_body, 1);
+}

ext/liquid/block.h

@@ -0,0 +1,8 @@
+#ifndef LIQUID_BLOCK_H
+#define LIQUID_BLOCK_H
+
+void init_liquid_block();
+
+extern VALUE cLiquidBlock;
+
+#endif

ext/liquid/extconf.rb

@@ -0,0 +1,3 @@
+require 'mkmf'
+$CFLAGS << ' -Wall'
+create_makefile("liquid/liquid")

ext/liquid/liquid_ext.c

@@ -0,0 +1,15 @@
+#include "liquid_ext.h"
+
+VALUE mLiquid;
+VALUE cLiquidTemplate, cLiquidTag, cLiquidVariable;
+
+void Init_liquid(void)
+{
+    mLiquid = rb_define_module("Liquid");
+    cLiquidTemplate = rb_define_class_under(mLiquid, "Template", rb_cObject);
+    cLiquidTag = rb_define_class_under(mLiquid, "Tag", rb_cObject);
+    cLiquidVariable = rb_define_class_under(mLiquid, "Variable", rb_cObject);
+
+    init_liquid_tokenizer();
+    init_liquid_block();
+}

ext/liquid/liquid_ext.h

@@ -0,0 +1,15 @@
+#ifndef LIQUID_EXT_H
+#define LIQUID_EXT_H
+
+#include <stdbool.h>
+#include <ctype.h>
+#include <ruby.h>
+
+#include "tokenizer.h"
+#include "block.h"
+#include "utils.h"
+
+extern VALUE mLiquid;
+extern VALUE cLiquidTemplate, cLiquidTag, cLiquidVariable;
+
+#endif

ext/liquid/tokenizer.c

@@ -0,0 +1,113 @@
+#include "liquid_ext.h"
+
+VALUE cLiquidTokenizer;
+
+static void free_tokenizer(void *ptr)
+{
+    struct liquid_tokenizer *tokenizer = ptr;
+    xfree(tokenizer);
+}
+
+static VALUE rb_allocate(VALUE klass)
+{
+    VALUE obj;
+    struct liquid_tokenizer *tokenizer;
+
+    obj = Data_Make_Struct(klass, struct liquid_tokenizer, NULL, free_tokenizer, tokenizer);
+    return obj;
+}
+
+static VALUE rb_initialize(VALUE self, VALUE source)
+{
+    struct liquid_tokenizer *tokenizer;
+
+    Check_Type(source, T_STRING);
+    Data_Get_Struct(self, struct liquid_tokenizer, tokenizer);
+    tokenizer->cursor = RSTRING_PTR(source);
+    tokenizer->length = RSTRING_LEN(source);
+    return Qnil;
+}
+
+void liquid_tokenizer_next(struct liquid_tokenizer *tokenizer, struct token *token)
+{
+    if (tokenizer->length <= 0) {
+        memset(token, 0, sizeof(*token));
+        return;
+    }
+    token->type = TOKEN_STRING;
+
+    char *cursor = tokenizer->cursor;
+    char *last = tokenizer->cursor + tokenizer->length - 1;
+
+    while (cursor < last) {
+        if (*cursor++ != '{')
+            continue;
+
+        char c = *cursor++;
+        if (c != '%' && c != '{')
+            continue;
+        if (cursor - tokenizer->cursor > 2) {
+            token->type = TOKEN_STRING;
+            cursor -= 2;
+            goto found;
+        }
+        char *incomplete_end = cursor;
+        token->type = TOKEN_INVALID;
+        if (c == '%') {
+            while (cursor < last) {
+                if (*cursor++ != '%')
+                    continue;
+                c = *cursor++;
+                while (c == '%' && cursor <= last)
+                    c = *cursor++;
+                if (c != '}')
+                    continue;
+                token->type = TOKEN_TAG;
+                goto found;
+            }
+            cursor = incomplete_end;
+            goto found;
+        } else {
+            while (cursor < last) {
+                if (*cursor++ != '}')
+                    continue;
+                if (*cursor++ != '}') {
+                    incomplete_end = cursor - 1;
+                    continue;
+                }
+                token->type = TOKEN_VARIABLE;
+                goto found;
+            }
+            cursor = incomplete_end;
+            goto found;
+        }
+    }
+    cursor = last + 1;
+found:
+    token->str = tokenizer->cursor;
+    token->length = cursor - tokenizer->cursor;
+    tokenizer->cursor += token->length;
+    tokenizer->length -= token->length;
+}
+
+static VALUE rb_next(VALUE self)
+{
+    struct liquid_tokenizer *tokenizer;
+    Data_Get_Struct(self, struct liquid_tokenizer, tokenizer);
+
+    struct token token;
+    liquid_tokenizer_next(tokenizer, &token);
+    if (token.type == TOKEN_NONE)
+        return Qnil;
+
+    return rb_str_new(token.str, token.length);
+}
+
+void init_liquid_tokenizer()
+{
+    cLiquidTokenizer = rb_define_class_under(mLiquid, "Tokenizer", rb_cObject);
+    rb_define_alloc_func(cLiquidTokenizer, rb_allocate);
+    rb_define_method(cLiquidTokenizer, "initialize", rb_initialize, 1);
+    rb_define_method(cLiquidTokenizer, "next", rb_next, 0);
+    rb_define_alias(cLiquidTokenizer, "shift", "next");
+}

ext/liquid/tokenizer.h

@@ -0,0 +1,30 @@
+#ifndef LIQUID_TOKENIZER_H
+#define LIQUID_TOKENIZER_H
+
+extern VALUE cLiquidTokenizer;
+
+enum token_type {
+    TOKEN_NONE,
+    TOKEN_INVALID,
+    TOKEN_STRING,
+    TOKEN_TAG,
+    TOKEN_VARIABLE
+};
+
+struct token {
+    enum token_type type;
+    char *str;
+    int length;
+};
+
+struct liquid_tokenizer {
+    char *cursor;
+    int length;
+};
+
+void init_liquid_tokenizer();
+void liquid_tokenizer_next(struct liquid_tokenizer *tokenizer, struct token *token);
+
+#define LIQUID_TOKENIZER_GET_STRUCT(obj) ((struct liquid_tokenizer *)obj_get_data_ptr(obj, cLiquidTokenizer))
+
+#endif

ext/liquid/utils.c

@@ -0,0 +1,21 @@
+#include <ruby.h>
+
+void raise_type_error(VALUE expected, VALUE got)
+{
+    rb_raise(rb_eTypeError, "wrong argument type %s (expected %s)",
+                             rb_class2name(got), rb_class2name(expected));
+}
+
+void check_class(VALUE obj, int type, VALUE klass)
+{
+    Check_Type(obj, type);
+    VALUE obj_klass = RBASIC_CLASS(obj);
+    if (obj_klass != klass)
+        raise_type_error(klass, obj_klass);
+}
+
+void *obj_get_data_ptr(VALUE obj, VALUE klass)
+{
+    check_class(obj, T_DATA, klass);
+    return DATA_PTR(obj);
+}

ext/liquid/utils.h

@@ -0,0 +1,8 @@
+#ifndef LIQUID_UTILS_H
+#define LIQUID_UTILS_H
+
+void raise_type_error(VALUE expected, VALUE got);
+void check_class(VALUE klass);
+void *obj_get_data_ptr(VALUE obj, VALUE klass);
+
+#endif

lib/liquid.rb

@@ -30,16 +30,13 @@ module Liquid
   VariableSegment             = /[\w\-]/
   VariableStart               = /\{\{/
   VariableEnd                 = /\}\}/
-  VariableIncompleteEnd       = /\}\}?/
   QuotedString                = /"[^"]*"|'[^']*'/
   QuotedFragment              = /#{QuotedString}|(?:[^\s,\|'"]|#{QuotedString})+/o
   TagAttributes               = /(\w+)\s*\:\s*(#{QuotedFragment})/o
-  AnyStartingTag              = /\{\{|\{\%/
-  PartialTemplateParser       = /#{TagStart}.*?#{TagEnd}|#{VariableStart}.*?#{VariableIncompleteEnd}/om
-  TemplateParser              = /(#{PartialTemplateParser}|#{AnyStartingTag})/om
   VariableParser              = /\[[^\]]+\]|#{VariableSegment}+\??/o
 end
 
+require 'liquid/liquid'
 require "liquid/version"
 require 'liquid/lexer'
 require 'liquid/parser'

lib/liquid/block.rb

@@ -1,82 +1,26 @@
 module Liquid
   class Block < Tag
-    IsTag             = /\A#{TagStart}/o
-    IsVariable        = /\A#{VariableStart}/o
-    FullToken         = /\A#{TagStart}\s*(\w+)\s*(.*)?#{TagEnd}\z/om
-    ContentOfVariable = /\A#{VariableStart}(.*)#{VariableEnd}\z/om
+    def initialize(tag_name, markup, tokens)
+      super
+      parse_body(tokens)
+    end
 
     def blank?
       @blank || false
     end
 
-    def parse(tokens)
-      @blank = true
-      @nodelist ||= []
-      @nodelist.clear
-
-      # All child tags of the current block.
-      @children = []
-
-      while token = tokens.shift
-        case token
-        when IsTag
-          if token =~ FullToken
-
-            # if we found the proper block delimiter just end parsing here and let the outer block
-            # proceed
-            if block_delimiter == $1
-              end_tag
-              return
-            end
-
-            # fetch the tag from registered blocks
-            if tag = Template.tags[$1]
-              new_tag = tag.parse($1, $2, tokens, @options)
-              @blank &&= new_tag.blank?
-              @nodelist << new_tag
-              @children << new_tag
-            else
-              # this tag is not registered with the system
-              # pass it to the current block for special handling or error reporting
-              unknown_tag($1, $2, tokens)
-            end
-          else
-            raise SyntaxError.new(options[:locale].t("errors.syntax.tag_termination".freeze, :token => token, :tag_end => TagEnd.inspect))
-          end
-        when IsVariable
-          new_var = create_variable(token)
-          @nodelist << new_var
-          @children << new_var
-          @blank = false
-        when ''.freeze
-          # pass
-        else
-          @nodelist << token
-          @blank &&= (token =~ /\A\s*\z/)
-        end
-      end
-
-      # Make sure that it's ok to end parsing in the current block.
-      # Effectively this method will throw an exception unless the current block is
-      # of type Document
-      assert_missing_delimitation!
-    end
-
     # warnings of this block and all sub-tags
     def warnings
       all_warnings = []
       all_warnings.concat(@warnings) if @warnings
 
-      (@children || []).each do |node|
-        all_warnings.concat(node.warnings || [])
+      (nodelist || []).each do |node|
+        all_warnings.concat(node.warnings || []) if node.respond_to?(:warnings)
       end
 
       all_warnings
     end
 
-    def end_tag
-    end
-
     def unknown_tag(tag, params, tokens)
       case tag
       when 'else'.freeze
@@ -112,6 +56,14 @@ module Liquid
 
     protected
 
+    def unterminated_variable(token)
+      raise SyntaxError.new(options[:locale].t("errors.syntax.variable_termination".freeze, :token => token, :tag_end => VariableEnd.inspect))
+    end
+
+    def unterminated_tag(token)
+      raise SyntaxError.new(options[:locale].t("errors.syntax.tag_termination".freeze, :token => token, :tag_end => TagEnd.inspect))
+    end
+
     def assert_missing_delimitation!
       raise SyntaxError.new(options[:locale].t("errors.syntax.tag_never_closed".freeze, :block_name => block_name))
     end

lib/liquid/document.rb

@@ -7,7 +7,7 @@ module Liquid
 
     # There isn't a real delimiter
     def block_delimiter
-      []
+      nil
     end
 
     # Document blocks don't need to be terminated since they are not actually opened

lib/liquid/tag.rb

@@ -19,9 +19,6 @@ module Liquid
       @options    = options
     end
 
-    def parse(tokens)
-    end
-
     def name
       self.class.name.downcase
     end

lib/liquid/tags/ifchanged.rb

@@ -4,7 +4,7 @@ module Liquid
     def render(context)
       context.stack do
 
-        output = render_all(@nodelist, context)
+        output = super
 
         if output != context.registers[:ifchanged]
           context.registers[:ifchanged] = output

lib/liquid/tags/include.rb

@@ -35,9 +35,6 @@ module Liquid
       end
     end
 
-    def parse(tokens)
-    end
-
     def blank?
       false
     end

lib/liquid/tags/raw.rb

@@ -2,16 +2,13 @@ module Liquid
   class Raw < Block
     FullTokenPossiblyInvalid = /\A(.*)#{TagStart}\s*(\w+)\s*(.*)?#{TagEnd}\z/om
 
-    def parse(tokens)
+    def parse_body(tokens)
       @nodelist ||= []
       @nodelist.clear
       while token = tokens.shift
         if token =~ FullTokenPossiblyInvalid
           @nodelist << $1 if $1 != "".freeze
-          if block_delimiter == $2
-            end_tag
-            return
-          end
+          return if block_delimiter == $2
         end
         @nodelist << token if not token.empty?
       end

lib/liquid/tags/table_row.rb

@@ -54,7 +54,7 @@ module Liquid
 
           col += 1
 
-          result << "<td class=\"col#{col}\">" << render_all(@nodelist, context) << '</td>'
+          result << "<td class=\"col#{col}\">" << super << '</td>'
 
           if col == cols and (index != length - 1)
             col  = 0

lib/liquid/template.rb

@@ -162,16 +162,9 @@ module Liquid
 
     private
 
-    # Uses the <tt>Liquid::TemplateParser</tt> regexp to tokenize the passed source
     def tokenize(source)
       source = source.source if source.respond_to?(:source)
-      return [] if source.to_s.empty?
-      tokens = source.split(TemplateParser)
-
-      # removes the rogue empty element at the beginning of the array
-      tokens.shift if tokens[0] and tokens[0].empty?
-
-      tokens
+      Tokenizer.new(source.to_s)
     end
 
   end

liquid.gemspec

@@ -18,13 +18,17 @@ Gem::Specification.new do |s|
   s.required_rubygems_version = ">= 1.3.7"
 
   s.test_files  = Dir.glob("{test}/**/*")
-  s.files       = Dir.glob("{lib}/**/*") + %w(MIT-LICENSE README.md)
+  s.files       = Dir.glob("{lib,ext}/**/*") + %w(MIT-LICENSE README.md)
+  s.extensions  = ['ext/liquid/extconf.rb']
 
   s.extra_rdoc_files  = ["History.md", "README.md"]
 
   s.require_path = "lib"
 
-  s.add_development_dependency 'stackprof' if Gem::Version.new(RUBY_VERSION) >= Gem::Version.new("2.1.0")
   s.add_development_dependency 'rake'
   s.add_development_dependency 'activesupport'
+  if RUBY_ENGINE == 'ruby'
+    s.add_development_dependency 'rake-compiler'
+    s.add_development_dependency 'stackprof' if Gem::Version.new(RUBY_VERSION) >= Gem::Version.new("2.1.0")
+  end
 end

test/liquid/template_test.rb

@@ -25,26 +25,6 @@ end
 class TemplateTest < Test::Unit::TestCase
   include Liquid
 
-  def test_tokenize_strings
-    assert_equal [' '], Template.new.send(:tokenize, ' ')
-    assert_equal ['hello world'], Template.new.send(:tokenize, 'hello world')
-  end
-
-  def test_tokenize_variables
-    assert_equal ['{{funk}}'], Template.new.send(:tokenize, '{{funk}}')
-    assert_equal [' ', '{{funk}}', ' '], Template.new.send(:tokenize, ' {{funk}} ')
-    assert_equal [' ', '{{funk}}', ' ', '{{so}}', ' ', '{{brother}}', ' '], Template.new.send(:tokenize, ' {{funk}} {{so}} {{brother}} ')
-    assert_equal [' ', '{{  funk  }}', ' '], Template.new.send(:tokenize, ' {{  funk  }} ')
-  end
-
-  def test_tokenize_blocks
-    assert_equal ['{%comment%}'], Template.new.send(:tokenize, '{%comment%}')
-    assert_equal [' ', '{%comment%}', ' '], Template.new.send(:tokenize, ' {%comment%} ')
-
-    assert_equal [' ', '{%comment%}', ' ', '{%endcomment%}', ' '], Template.new.send(:tokenize, ' {%comment%} {%endcomment%} ')
-    assert_equal ['  ', '{% comment %}', ' ', '{% endcomment %}', ' '], Template.new.send(:tokenize, "  {% comment %} {% endcomment %} ")
-  end
-
   def test_instance_assigns_persist_on_same_template_object_between_parses
     t = Template.new
     assert_equal 'from instance assigns', t.parse("{% assign foo = 'from instance assigns' %}{{ foo }}").render!

test/liquid/tokenizer_test.rb

@@ -0,0 +1,64 @@
+require 'test_helper'
+
+class TokenizerTest < Test::Unit::TestCase
+  def test_tokenize_strings
+    assert_equal [' '], tokenize(' ')
+    assert_equal ['hello world'], tokenize('hello world')
+  end
+
+  def test_tokenize_variables
+    assert_equal ['{{funk}}'], tokenize('{{funk}}')
+    assert_equal [' ', '{{funk}}', ' '], tokenize(' {{funk}} ')
+    assert_equal [' ', '{{funk}}', ' ', '{{so}}', ' ', '{{brother}}', ' '], tokenize(' {{funk}} {{so}} {{brother}} ')
+    assert_equal [' ', '{{  funk  }}', ' '], tokenize(' {{  funk  }} ')
+  end
+
+  def test_tokenize_blocks
+    assert_equal ['{%comment%}'], tokenize('{%comment%}')
+    assert_equal [' ', '{%comment%}', ' '], tokenize(' {%comment%} ')
+
+    assert_equal [' ', '{%comment%}', ' ', '{%endcomment%}', ' '], tokenize(' {%comment%} {%endcomment%} ')
+    assert_equal ['  ', '{% comment %}', ' ', '{% endcomment %}', ' '], tokenize("  {% comment %} {% endcomment %} ")
+  end
+
+  def test_tokenize_incomplete_end
+    assert_tokens 'before{{ incomplete }after', ['before', '{{ incomplete }', 'after']
+    assert_tokens 'before{% incomplete %after', ['before', '{%', ' incomplete %after']
+  end
+
+  def test_tokenize_no_end
+    assert_tokens 'before{{ unterminated ', ['before', '{{', ' unterminated ']
+    assert_tokens 'before{% unterminated ', ['before', '{%', ' unterminated ']
+  end
+
+  private
+
+  def assert_tokens(source, expected)
+    assert_equal expected, tokenize(source)
+    assert_equal expected, old_tokenize(source)
+  end
+
+  def tokenize(source)
+    tokenizer = Liquid::Tokenizer.new(source)
+    tokens = []
+    while token = tokenizer.next
+      tokens << token
+    end
+    tokens
+  end
+
+  AnyStartingTag        = /\{\{|\{\%/
+  VariableIncompleteEnd = /\}\}?/
+  PartialTemplateParser = /#{Liquid::TagStart}.*?#{Liquid::TagEnd}|#{Liquid::VariableStart}.*?#{VariableIncompleteEnd}/o
+  TemplateParser        = /(#{PartialTemplateParser}|#{AnyStartingTag})/o
+
+  def old_tokenize(source)
+    return [] if source.to_s.empty?
+    tokens = source.split(TemplateParser)
+
+    # removes the rogue empty element at the beginning of the array
+    tokens.shift if tokens[0] and tokens[0].empty?
+
+    tokens
+  end
+end