Use each_line. Avoid comparison of partial. Add UTF-8 spec.

lib/gitlab/ci/trace/stream.rb

@@ -125,11 +125,10 @@ module Gitlab
             stream.seek(-pos, IO::SEEK_END)
             stream.read(BUFFER_SIZE).tap do |buf|
               buf = buf + debris
-              lines = buf.split("\n")
-              lines.reverse.each do |line|
+              debris, *lines = buf.each_line.to_a
+              lines.reverse_each do |line|
                 yield(line)
               end
-              debris = lines.count > 0 ? lines[0] : ''
             end
             pos += BUFFER_SIZE
           end
@@ -139,8 +138,7 @@ module Gitlab
           last = (max > BUFFER_SIZE) ? (max % BUFFER_SIZE) : max
           stream.read(last).tap do |buf|
             buf = buf + debris
-            lines = buf.split("\n")
-            lines.reverse.each do |line|
+            buf.each_line.reverse_each do |line|
               yield(line)
             end
           end

spec/lib/gitlab/ci/trace/stream_spec.rb

@@ -265,6 +265,17 @@ describe Gitlab::Ci::Trace::Stream do
       it { is_expected.to eq("98.29") }
     end
 
+    context 'when regex is multi-byte char' do
+      let(:data) { 'ゴッドファット\n' }
+      let(:regex) { 'ゴッドファット' }
+
+      before do
+        stub_const('Gitlab::Ci::Trace::Stream::BUFFER_SIZE', 5)
+      end
+
+      it { is_expected.to be_nil }
+    end
+
     context 'when BUFFER_SIZE is equal to stream.size' do
       let(:data) { 'Coverage 1033 / 1051 LOC (98.29%) covered\n' }
       let(:regex) { '\(\d+.\d+\%\) covered' }