Fix subprocess encoding for Python 3.3/3.4 tests

The issue was that Python 3.3/3.4 don't have subprocess.run() so they
use the legacy Popen path, but unlike Python 2.7, they need bytes for
stdin, not strings.

Fixed by checking Python version separately from subprocess.run feature
detection and encoding strings to bytes for Python 3.x versions.

Author: dflook

test/subprocess_compat.py

@@ -1,5 +1,6 @@
 """Subprocess compatibility utilities for Python 2.7/3.x."""
 import subprocess
+import sys
 
 
 def run_subprocess(cmd, timeout=None, input_data=None, env=None):
@@ -10,11 +11,13 @@ def run_subprocess(cmd, timeout=None, input_data=None, env=None):
         return subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, 
                             input=input_bytes, timeout=timeout, env=env)
     else:
-        # Python 2.7, 3.3, 3.4 - use str as-is, no timeout support
+        # Python 2.7, 3.3, 3.4 - no subprocess.run, no timeout support
         popen = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, 
                                stdin=subprocess.PIPE if input_data else None, env=env)
         # For Python 3.3/3.4, communicate() doesn't support timeout
-        # We'll ignore timeout for older versions - tests should be fast enough
+        # Also, Python 3.x needs bytes for stdin, Python 2.x needs str
+        if input_data and sys.version_info[0] >= 3 and isinstance(input_data, str):
+            input_data = input_data.encode('utf-8')
         stdout, stderr = popen.communicate(input_data)
         # Create a simple result object similar to subprocess.CompletedProcess
         class Result: