Issue #7151: regrtest would generate a JSON failure if there was output

to stderr during the test run and it happened to get emitted after the
worker thread emitted the result JSON.  Now we capture stdout and stderr
separately, which avoids that problem.  It also means that _all_ stderr
output is after all stdout output when we print the test results, but
that seems acceptable, since output ordering is not guaranteed anyway.
The patch also moves the emit of the test name into the output block
generated after the test completes.  Otherwise test names and test
output/errors were mixed in the terminal display, making it difficult
to determine which test generated the output.

1 files changed, 16 insertions, 13 deletions

diff --git a/Lib/test/regrtest.py b/Lib/test/regrtest.py
index be88716..d8456df 100755
--- a/Lib/test/regrtest.py
+++ b/Lib/test/regrtest.py
@@ -426,35 +426,38 @@ def main(tests=None, testdir=None, verbose=0, quiet=False,
                     try:
                         test, args_tuple = pending.popleft()
                     except IndexError:
-                        output.put((None, None, None))
+                        output.put((None, None, None, None))
                         return
-                    if not quiet:
-                        print test
-                        sys.stdout.flush()
                     # -E is needed by some tests, e.g. test_import
                     popen = Popen([sys.executable, '-E', '-m', 'test.regrtest',
                                    '--slaveargs', json.dumps(args_tuple)],
-                                   stdout=PIPE, stderr=STDOUT,
+                                   stdout=PIPE, stderr=PIPE,
                                    universal_newlines=True, close_fds=True)
-                    out = popen.communicate()[0].strip()
-                    out = debug_output_pat.sub("", out)
-                    out, _, result = out.strip().rpartition("\n")
+                    stdout, stderr = popen.communicate()
+                    # Strip last refcount output line if it exists, since it
+                    # comes from the shutdown of the interpreter in the subcommand.
+                    stderr = debug_output_pat.sub("", stderr)
+                    stdout, _, result = stdout.strip().rpartition("\n")
                     result = json.loads(result)
-                    output.put((test, out.strip(), result))
+                    if not quiet:
+                        stdout = test+'\n'+stdout
+                    output.put((test, stdout.rstrip(), stderr.rstrip(), result))
             except BaseException:
-                output.put((None, None, None))
+                output.put((None, None, None, None))
                 raise
         workers = [Thread(target=work) for i in range(use_mp)]
         for worker in workers:
             worker.start()
         finished = 0
         while finished < use_mp:
-            test, out, result = output.get()
+            test, stdout, stderr, result = output.get()
             if test is None:
                 finished += 1
                 continue
-            if out:
-                print out
+            if stdout:
+                print stdout
+            if stderr:
+                print >>sys.stderr, stderr
             if result[0] == -4:
                 assert result[1] == 'KeyboardInterrupt'
                 pending.clear()