don't apply passthrough substitution inside stem macros

I just moved processing of the stem substitutions before processing of
the passthrough substitutions. It seems that it works well and the side
effects of this change are acceptable, i.e. they will probably affect
much less users than the current undesirable behaviour of + signs inside
the stem macros.

Fixes asciidoctor#3409

lib/asciidoctor/substitutors.rb

@@ -1053,6 +1053,22 @@ def extract_passthroughs text
       %(#{preceding || ''}#{PASS_START}#{passthru_key}#{PASS_END})
     end if (text.include? '++') || (text.include? '$$') || (text.include? 'ss:')
 
+    text = text.gsub InlineStemMacroRx do
+      # honor the escape
+      next $&.slice 1, $&.length if $&.start_with? RS
+
+      if (type = $1.to_sym) == :stem
+        type = STEM_TYPE_ALIASES[@document.attributes['stem']].to_sym
+      end
+      subs = $2
+      content = normalize_text $3, nil, true
+      # NOTE drop enclosing $ signs around latexmath for backwards compatibility with AsciiDoc Python
+      content = content.slice 1, content.length - 2 if type == :latexmath && (content.start_with? '$') && (content.end_with? '$')
+      subs = subs ? (resolve_pass_subs subs) : ((@document.basebackend? 'html') ? BASIC_SUBS : nil)
+      passthrus[passthru_key = passthrus.size] = { text: content, subs: subs, type: type }
+      %(#{PASS_START}#{passthru_key}#{PASS_END})
+    end if (text.include? ':') && ((text.include? 'stem:') || (text.include? 'math:'))
+
     pass_inline_char1, pass_inline_char2, pass_inline_rx = InlinePassRx[compat_mode]
     text = text.gsub pass_inline_rx do
       preceding = $1
@@ -1102,23 +1118,6 @@ def extract_passthroughs text
       %(#{preceding}#{PASS_START}#{passthru_key}#{PASS_END})
     end if (text.include? pass_inline_char1) || (pass_inline_char2 && (text.include? pass_inline_char2))
 
-    # NOTE we need to do the stem in a subsequent step to allow it to be escaped by the former
-    text = text.gsub InlineStemMacroRx do
-      # honor the escape
-      next $&.slice 1, $&.length if $&.start_with? RS
-
-      if (type = $1.to_sym) == :stem
-        type = STEM_TYPE_ALIASES[@document.attributes['stem']].to_sym
-      end
-      subs = $2
-      content = normalize_text $3, nil, true
-      # NOTE drop enclosing $ signs around latexmath for backwards compatibility with AsciiDoc Python
-      content = content.slice 1, content.length - 2 if type == :latexmath && (content.start_with? '$') && (content.end_with? '$')
-      subs = subs ? (resolve_pass_subs subs) : ((@document.basebackend? 'html') ? BASIC_SUBS : nil)
-      passthrus[passthru_key = passthrus.size] = { text: content, subs: subs, type: type }
-      %(#{PASS_START}#{passthru_key}#{PASS_END})
-    end if (text.include? ':') && ((text.include? 'stem:') || (text.include? 'math:'))
-
     text
   end
 

test/substitutions_test.rb

@@ -1955,15 +1955,7 @@
       end
 
       test 'should passthrough math macro inside another passthrough' do
-        input = 'the text `asciimath:[x = y]` should be passed through as +literal+ text'
-        para = block_from_string input, attributes: { 'compat-mode' => '' }
-        assert_equal 'the text <code>asciimath:[x = y]</code> should be passed through as <code>literal</code> text', para.content
-
-        input = 'the text [x-]`asciimath:[x = y]` should be passed through as `literal` text'
-        para = block_from_string input
-        assert_equal 'the text <code>asciimath:[x = y]</code> should be passed through as <code>literal</code> text', para.content
-
-        input = 'the text `+asciimath:[x = y]+` should be passed through as `literal` text'
+        input = 'the text `++asciimath:[x = y]++` should be passed through as `literal` text'
         para = block_from_string input
         assert_equal 'the text <code>asciimath:[x = y]</code> should be passed through as <code>literal</code> text', para.content
       end