Adding a forced UTF-8 conversion to prevent encoding errors

Changelog: fixed

config/feature_flags/development/convert_diff_to_utf8_with_replacement_symbol.yml

+---
+name: convert_diff_to_utf8_with_replacement_symbol
+introduced_by_url:
+rollout_issue_url:
+milestone: '14.9'
+type: development
+group: group::code review
+default_enabled: false

lib/gitlab/encoding_helper.rb

@@ -15,6 +15,8 @@ module Gitlab
     # https://gitlab.com/gitlab-org/gitlab_git/merge_requests/77#note_4754193
     ENCODING_CONFIDENCE_THRESHOLD = 50
 
+    INVALID_UTF_CHARACTER_PLACEHOLDER = "☃"
+
     def encode!(message)
       message = force_encode_utf8(message)
       return message if message.valid_encoding?
@@ -48,6 +50,14 @@ module Gitlab
       detect && detect[:type] == :binary && detect[:confidence] == 100
     end
 
+    def detect_invalid_utf8?(data)
+      data.include?(INVALID_UTF_CHARACTER_PLACEHOLDER)
+    end
+
+    def fix_invalid_utf8(data)
+      encode_utf8(data, replace: INVALID_UTF_CHARACTER_PLACEHOLDER)
+    end
+
     # EncodingDetector checks the first 1024 * 1024 bytes for NUL byte, libgit2 checks
     # only the first 8000 (https://github.com/libgit2/libgit2/blob/2ed855a9e8f9af211e7274021c2264e600c0f86b/src/filter.h#L15),
     # which is what we use below to keep a consistent behavior.

lib/gitlab/git/diff_collection.rb

@@ -157,6 +157,12 @@ module Gitlab
         @iterator.size == 1 || !@enforce_limits || @expanded
       end
 
+      def fix_invalid_diff!(diff)
+        converted_diff = Gitlab::EncodingHelper.fix_invalid_utf8(diff.diff)
+
+        diff.diff = converted_diff if Gitlab::EncodingHelper.detect_invalid_utf8?(converted_diff)
+      end
+
       def each_gitaly_patch
         i = @array.length
 
@@ -187,6 +193,10 @@ module Gitlab
 
           diff = Gitlab::Git::Diff.new(raw, expanded: expand_diff?)
 
+          if Feature.enabled?(:convert_diff_to_utf8_with_replacement_symbol, default_enabled: :yaml)
+            fix_invalid_diff!(diff)
+          end
+
           if !expand_diff? && over_safe_limits?(i) && diff.line_count > 0
             diff.collapse!
           end

spec/lib/gitlab/git/diff_collection_spec.rb

@@ -775,6 +775,19 @@ RSpec.describe Gitlab::Git::DiffCollection, :seed_helper do
         end
       end
     end
+
+    context 'when diff contains invalid characters' do
+      let(:bad_string) { [0xae].pack("C*") }
+      let(:bad_string_two) { [0x89].pack("C*") }
+
+      let(:collection) do
+        Gitlab::Git::DiffCollection.new([{ diff: bad_string }, { diff: bad_string_two }])
+      end
+
+      it 'will not error out' do
+        expect { Oj.dump(collection) }.not_to raise_error(EncodingError)
+      end
+    end
   end
 
   def fake_diff(line_length, line_count)