Avoid ascii/utf encoding errors on some systems

Author: kastiglione

Utilities/build-script-helper.py

@@ -9,9 +9,6 @@
 import sys
 import errno
 
-sys.stdout.reconfigure(encoding='utf-8')
-sys.stderr.reconfigure(encoding='utf-8')
-
 if platform.system() == 'Darwin':
  shared_lib_ext = '.dylib'
 else:
@@ -564,19 +561,24 @@ def cmake_build(args, swiftc_exec, cmake_args, swift_flags, source_path,
 
  if args.verbose:
  print(' '.join(ninja_cmd))
+ # Note: encoding is explicitly set to None to indicate that the output must
+ # be bytes, not strings. This is to work around per-system differences in
+ # default encoding. Some systems have a default encoding of 'ascii', but that
+ # conflicts with this output, which can contain UTF encoded characters. The
+ # bytes are then written, instead of printed, to bypass issues with encoding.
  ninjaProcess = subprocess.Popen(ninja_cmd, cwd=build_dir,
  stdout=subprocess.PIPE,
  stderr=subprocess.PIPE,
- env = os.environ,
- encoding='utf-8')
+ env=os.environ,
+ encoding=None)
  stdout, stderr = ninjaProcess.communicate()
  if ninjaProcess.returncode != 0:
- print(stdout)
+ sys.stdout.buffer.write(stdout)
  print('Ninja invocation failed: ')
- print(stderr)
+ sys.stderr.buffer.write(stderr)
  sys.exit(ninjaProcess.returncode)
  if args.verbose:
- print(stdout)
+ sys.stdout.buffer.write(stdout)
 
 def get_build_target(swiftc_path, args, cross_compile=False):
  """Returns the target-triple of the current machine."""