gclient_utils: buffer output as bytestrings in Annotated

In Python 3 byestrings and normal strings can't be concatenated.
To fix this we buffer as bytestrings in the Annotated wrapper.
We can't decode to a string because the output might come byte-by-byte, so it doesn't work with Unicode characters like ✔.

Also had to update gclient_test.py, where double-wrapping stdout with Annotated caused made output not work and include_zero=True working caused other unintended side-effects.

Example error from "fetch chromium":
Traceback (most recent call last):
  File "C:\Google\depot_tools\gclient_scm.py", line 1045, in _Clone
    self._Run(clone_cmd, options, cwd=self._root_dir, retry=True,
  File "C:\Google\depot_tools\gclient_scm.py", line 1370, in _Run
    gclient_utils.CheckCallAndFilter(cmd, env=env, **kwargs)
  File "C:\Google\depot_tools\gclient_utils.py", line 583, in CheckCallAndFilter
    show_header_if_necessary(needs_header, attempt)
  File "C:\Google\depot_tools\gclient_utils.py", line 533, in show_header_if_necessary
    stdout_write(header.encode())
  File "C:\Google\depot_tools\gclient_utils.py", line 391, in write
    obj[0] += out
TypeError: can only concatenate str (not "bytes") to str

Bug: 984182
Change-Id: If7037d30e9faf524f2405258281f6e6cd0bcdcae
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/tools/depot_tools/+/1778745
Commit-Queue: Edward Lesmes <ehmaldonado@chromium.org>
Reviewed-by: Edward Lesmes <ehmaldonado@chromium.org>
Auto-Submit: Raul Tambre <raul@tambre.ee>
diff --git a/gclient_utils.py b/gclient_utils.py
index 3a6bc1f..2c38a3a 100644
--- a/gclient_utils.py
+++ b/gclient_utils.py
@@ -40,8 +40,10 @@
 
 if sys.version_info.major == 2:
   from cStringIO import StringIO
+  string_type = basestring
 else:
   from io import StringIO
+  string_type = str
 
 
 RETRY_MAX = 3
@@ -371,6 +373,10 @@
   def write(self, out):
     index = getattr(threading.currentThread(), 'index', 0)
     if not index and not self.__include_zero:
+      # Store as bytes to ensure Unicode characters get output correctly.
+      if isinstance(out, bytes):
+        out = out.decode('utf-8')
+
       # Unindexed threads aren't buffered.
       return self._wrapped.write(out)
 
@@ -380,28 +386,32 @@
       # Strings are immutable, requiring to keep a lock for the whole dictionary
       # otherwise. Using an array is faster than using a dummy object.
       if not index in self.__output_buffers:
-        obj = self.__output_buffers[index] = ['']
+        obj = self.__output_buffers[index] = [b'']
       else:
         obj = self.__output_buffers[index]
     finally:
       self.lock.release()
 
+    # Store as bytes to ensure Unicode characters get output correctly.
+    if isinstance(out, string_type):
+      out = out.encode('utf-8')
+
     # Continue lockless.
     obj[0] += out
     while True:
       # TODO(agable): find both of these with a single pass.
-      cr_loc = obj[0].find('\r')
-      lf_loc = obj[0].find('\n')
+      cr_loc = obj[0].find(b'\r')
+      lf_loc = obj[0].find(b'\n')
       if cr_loc == lf_loc == -1:
         break
       elif cr_loc == -1 or (lf_loc >= 0 and lf_loc < cr_loc):
-        line, remaining = obj[0].split('\n', 1)
+        line, remaining = obj[0].split(b'\n', 1)
         if line:
-          self._wrapped.write('%d>%s\n' % (index, line))
+          self._wrapped.write('%d>%s\n' % (index, line.decode('utf-8')))
       elif lf_loc == -1 or (cr_loc >= 0 and cr_loc < lf_loc):
-        line, remaining = obj[0].split('\r', 1)
+        line, remaining = obj[0].split(b'\r', 1)
         if line:
-          self._wrapped.write('%d>%s\r' % (index, line))
+          self._wrapped.write('%d>%s\r' % (index, line.decode('utf-8')))
       obj[0] = remaining
 
   def flush(self):
@@ -423,7 +433,7 @@
     # Don't keep the lock while writting. Will append \n when it shouldn't.
     for orphan in orphans:
       if orphan[1]:
-        self._wrapped.write('%d>%s\n' % (orphan[0], orphan[1]))
+        self._wrapped.write('%d>%s\n' % (orphan[0], orphan[1].decode('utf-8')))
     return self._wrapped.flush()