Optimize Nokogiri search for post-processing pipeline

Contributes to https://gitlab.com/gitlab-org/gitlab/-/issues/271242

**Problem**

doc.search(...) operation has a bad performance for complex html
documents with many elements.

**Solution**

Avoid unnecessary doc.search calls.

1. Combine multiple search operations
2. Share calculated result between several filters

config/feature_flags/development/optimize_linkable_attributes.yml

+---
+name: optimize_linkable_attributes
+introduced_by_url: https://gitlab.com/gitlab-org/gitlab/-/merge_requests/59983
+rollout_issue_url: https://gitlab.com/gitlab-org/gitlab/-/issues/328696
+milestone: '13.12'
+type: development
+group: group::source code
+default_enabled: false

lib/banzai/filter/base_relative_link_filter.rb

@@ -10,19 +10,16 @@ module Banzai
       protected
 
       def linkable_attributes
-        strong_memoize(:linkable_attributes) do
-          attrs = []
-
-          attrs += doc.search('a:not(.gfm)').map do |el|
-            el.attribute('href')
-          end
-
-          attrs += doc.search('img:not(.gfm), video:not(.gfm), audio:not(.gfm)').flat_map do |el|
-            [el.attribute('src'), el.attribute('data-src')]
-          end
-
-          attrs.reject do |attr|
-            attr.blank? || attr.value.start_with?('//')
+        if Feature.enabled?(:optimize_linkable_attributes, project, default_enabled: :yaml)
+          # Nokorigi Nodeset#search performs badly for documents with many nodes
+          #
+          # Here we store fetched attributes in the shared variable "result"
+          # This variable is passed through the chain of filters and can be
+          # accessed by them
+          result[:linkable_attributes] ||= fetch_linkable_attributes
+        else
+          strong_memoize(:linkable_attributes) do
+            fetch_linkable_attributes
           end
         end
       end
@@ -40,6 +37,16 @@ module Banzai
       def unescape_and_scrub_uri(uri)
         Addressable::URI.unescape(uri).scrub.delete("\0")
       end
+
+      def fetch_linkable_attributes
+        attrs = []
+
+        attrs += doc.search('a:not(.gfm), img:not(.gfm), video:not(.gfm), audio:not(.gfm)').flat_map do |el|
+          [el.attribute('href'), el.attribute('src'), el.attribute('data-src')]
+        end
+
+        attrs.reject { |attr| attr.blank? || attr.value.start_with?('//') }
+      end
     end
   end
 end

lib/banzai/filter/upload_link_filter.rb

@@ -15,8 +15,16 @@ module Banzai
       def call
         return doc if context[:system_note]
 
-        linkable_attributes.each do |attr|
-          process_link_to_upload_attr(attr)
+        if Feature.enabled?(:optimize_linkable_attributes, project, default_enabled: :yaml)
+          # We exclude processed upload links from the linkable attributes to
+          # prevent further modifications by RepositoryLinkFilter
+          linkable_attributes.reject! do |attr|
+            process_link_to_upload_attr(attr)
+          end
+        else
+          linkable_attributes.each do |attr|
+            process_link_to_upload_attr(attr)
+          end
         end
 
         doc

spec/lib/banzai/pipeline/post_process_pipeline_spec.rb

@@ -3,24 +3,56 @@
 require 'spec_helper'
 
 RSpec.describe Banzai::Pipeline::PostProcessPipeline do
-  context 'when a document only has upload links' do
-    it 'does not make any Gitaly calls', :request_store do
-      markdown = <<-MARKDOWN.strip_heredoc
-        [Relative Upload Link](/uploads/e90decf88d8f96fe9e1389afc2e4a91f/test.jpg)
+  subject { described_class.call(doc, context) }
+
+  let_it_be(:project) { create(:project, :public, :repository) }
 
-        ![Relative Upload Image](/uploads/e90decf88d8f96fe9e1389afc2e4a91f/test.jpg)
-      MARKDOWN
+  let(:context) { { project: project, ref: 'master' } }
 
-      context = {
-        project: create(:project, :public, :repository),
-        ref: 'master'
-      }
+  context 'when a document only has upload links' do
+    let(:doc) do
+      <<-HTML.strip_heredoc
+        <a href="/uploads/e90decf88d8f96fe9e1389afc2e4a91f/test.jpg">Relative Upload Link</a>
+        <img src="/uploads/e90decf88d8f96fe9e1389afc2e4a91f/test.jpg">
+      HTML
+    end
 
+    it 'does not make any Gitaly calls', :request_store do
       Gitlab::GitalyClient.reset_counts
 
-      described_class.call(markdown, context)
+      subject
 
       expect(Gitlab::GitalyClient.get_request_count).to eq(0)
     end
   end
+
+  context 'when both upload and repository links are present' do
+    let(:html) do
+      <<-HTML.strip_heredoc
+        <a href="/uploads/e90decf88d8f96fe9e1389afc2e4a91f/test.jpg">Relative Upload Link</a>
+        <img src="/uploads/e90decf88d8f96fe9e1389afc2e4a91f/test.jpg">
+        <a href="/test.jpg">Just a link</a>
+      HTML
+    end
+
+    let(:doc) { HTML::Pipeline.parse(html) }
+
+    it 'searches for attributes only once' do
+      expect(doc).to receive(:search).once.and_call_original
+
+      subject
+    end
+
+    context 'when "optimize_linkable_attributes" is disabled' do
+      before do
+        stub_feature_flags(optimize_linkable_attributes: false)
+      end
+
+      it 'searches for attributes twice' do
+        expect(doc).to receive(:search).twice.and_call_original
+
+        subject
+      end
+    end
+  end
 end