AI: address review on PR #461

backend/app/services/post_tag_search.rb

@@ -3,8 +3,8 @@ class PostTagSearch
   def self.literals(locale, raw)
     tokens = raw.split.reject { _1.downcase == 'type:or' }
     names = tokens.map { _1.sub(/\Anot:/i, '') }
-    registered = TagName.where(language_code: locale.language_code, name: names)
-        .pluck(:name, :tag_id).to_h
+    registered = matching_names(TagName.where(language_code: locale.language_code),
+        names, 'tag_names.name', :tag_id).to_h
     primary = TagName.where(language_code: locale.language_code,
         tag_id: registered.values, primary_flg: true).pluck(:tag_id, :name).to_h
     tokens.zip(names).map do |token, name|
@@ -19,12 +19,25 @@ class PostTagSearch
 
   def self.references_for(locale, literals)
     names = literals.map { _1[:name] }.uniq
-    internal = TagName.where(language_code: locale.language_code, name: names)
-        .pluck(:name, :tag_id).group_by(&:first)
-    external = names.empty? ? {} : ExternalTag.where(
-        "CONCAT(platform, ':', name) IN (?)", names)
-        .pluck(Arel.sql("CONCAT(platform, ':', name)"), :id).group_by(&:first)
+    internal = matching_names(TagName.where(language_code: locale.language_code),
+        names, 'tag_names.name', :tag_id).group_by(&:first)
+    external = matching_names(ExternalTag.all, names,
+        "CONCAT(external_tags.platform, ':', external_tags.name)", :id).group_by(&:first)
     names.to_h { |name| [name, [internal.fetch(name, []).map(&:last).uniq,
                               external.fetch(name, []).map(&:last).uniq]] }
   end
+
+  def self.matching_names(scope, names, expression, identifier)
+    return [] if names.empty?
+
+    queries = names.uniq.map do |name|
+      scope.klass.sanitize_sql_array(['SELECT ? AS name', name])
+    end
+    # Return the input spelling so Ruby does not recheck database equivalence.
+    scope.joins(<<~SQL).pluck(Arel.sql('search_names.name'), identifier)
+      INNER JOIN (#{ queries.join(' UNION ALL ') }) AS search_names
+        ON #{ expression } = search_names.name COLLATE utf8mb4_0900_ai_ci
+    SQL
+  end
+  private_class_method :matching_names
 end

backend/spec/requests/posts_spec.rb

@@ -206,6 +206,25 @@ RSpec.describe 'Posts API', type: :request do
     end
 
     context 'when q is provided' do
+      ['SPEC_TAG', 'SpEc_TaG', 'spec_tág', 'MANKO', 'MaNkO', 'mánko'].each do |name|
+        it "matches #{ name } using database collation" do
+          get '/posts', params: { locale: 'ja', tags: name }
+
+          expect(response).to have_http_status(:ok)
+          expect(json.fetch('count')).to eq(2)
+          expect(json.fetch('posts').map { _1.fetch('id') })
+              .to contain_exactly(post_record.id, hit_post.id)
+        end
+
+        it "excludes #{ name } using database collation" do
+          get '/posts', params: { locale: 'ja', tags: "not:#{ name }" }
+
+          expect(response).to have_http_status(:ok)
+          expect(json.fetch('count')).to eq(1)
+          expect(json.fetch('posts').map { _1.fetch('id') }).to eq([miss_post.id])
+        end
+      end
+
       it 'filters posts by q (hit case)' do
         get '/posts', params: { tags: 'spec_tag' }
 
@@ -266,6 +285,26 @@ RSpec.describe 'Posts API', type: :request do
         PostExternalTag.create!(post: miss_post, external_tag:)
       end
 
+      ['NICO:SEARCH_EXTERNAL', 'NiCo:SeArCh_ExTeRnAl', 'nico:séarch_external'].each do |name|
+        it "matches #{ name } using database collation" do
+          get '/posts', params: { locale: 'ja', tags: name }
+
+          expect(response).to have_http_status(:ok)
+          expect(json.fetch('count')).to eq(2)
+          expect(json.fetch('posts').map { _1.fetch('id') })
+              .to contain_exactly(miss_post.id, both_post.id)
+        end
+
+        it "excludes #{ name } using database collation" do
+          get '/posts', params: { locale: 'ja', tags: "not:#{ name }" }
+
+          expect(response).to have_http_status(:ok)
+          expect(json.fetch('count')).to eq(2)
+          expect(json.fetch('posts').map { _1.fetch('id') })
+              .to contain_exactly(post_record.id, hit_post.id)
+        end
+      end
+
       it 'keeps internal name searches independent of colliding external ids' do
         get '/posts', params: { tags: primary_tag_name_for(tag, 'ja').name }