gh-132962: _pyrepl: Prevent crash on Windows when stdout is redirected

[shuimu5418]

The interactive interpreter (-i) on Windows no longer crashes when standard output is redirected. The check for CAN_USE_PYREPL now also checks if stdout is a TTY on Windows, falling back to the basic REPL if it is not.

• Issue: windows console: -i parameter and redirected output gives error in 3.13 #132962

---

Main branch:

PS F:\github\cpython001> .\python.bat -i .\build\testpys\print_helloworld.py > .\build\file.txt
Traceback (most recent call last):
  File "F:\github\cpython001\Lib\_pyrepl\readline.py", line 394, in multiline_input
    return reader.readline()
           ~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\reader.py", line 742, in readline
    self.prepare()
    ~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\historical_reader.py", line 306, in prepare
    super().prepare()
    ~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\reader.py", line 585, in prepare
    self.console.prepare()
    ~~~~~~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\windows_console.py", line 358, in prepare
    self.height, self.width = self.getheightwidth()
                              ~~~~~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\windows_console.py", line 409, in getheightwidth
    raise WinError(GetLastError())
OSError: [WinError 0] The operation completed successfully.
Traceback (most recent call last):
  File "F:\github\cpython001\Lib\_pyrepl\readline.py", line 394, in multiline_input
    return reader.readline()
           ~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\reader.py", line 742, in readline
    self.prepare()
    ~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\historical_reader.py", line 306, in prepare
    super().prepare()
    ~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\reader.py", line 585, in prepare
    self.console.prepare()
    ~~~~~~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\windows_console.py", line 358, in prepare
    self.height, self.width = self.getheightwidth()
                              ~~~~~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\windows_console.py", line 409, in getheightwidth
    raise WinError(GetLastError())
OSError: [WinError 123] The filename, directory name, or volume label syntax is incorrect.
Traceback (most recent call last):
  File "F:\github\cpython001\Lib\_pyrepl\readline.py", line 394, in multiline_input
    return reader.readline()
           ~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\reader.py", line 742, in readline
    self.prepare()
    ~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\historical_reader.py", line 306, in prepare
    super().prepare()
    ~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\reader.py", line 585, in prepare
    self.console.prepare()
    ~~~~~~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\windows_console.py", line 358, in prepare
    self.height, self.width = self.getheightwidth()
                              ~~~~~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\windows_console.py", line 409, in getheightwidth
    raise WinError(GetLastError())
OSError: [WinError 123] The filename, directory name, or volume label syntax is incorrect.
Traceback (most recent call last):
...... Infinite Tracebacks......

With fix:

PS F:\github\cpython001> .\python.bat -i .\build\testpys\print_helloworld.py > .\build\file.txt
warning: can't use pyrepl: [Errno 25] tty required: 'stdout'
>>> 

[ZeroIntensity]

Please add a test case. You can refer to the devguide or tag me if you're unsure what that entails.

[shuimu5418]

@ZeroIntensity Thanks for your guidance on adding a test.

I've drafted a test case below that correctly fails on the main branch and passes with my fix. I haven't committed and pushed this yet, because I wanted to get your feedback first.

I feel the approach is very very ugly, especially the need to use time.sleep and then kill the process. Is this an acceptable way to test this kind of hanging/crashing bug, or do you have any suggestions for a cleaner approach?

My rationale for this test:
The bug only manifests in a specific I/O environment on Windows (stdin is a TTY, but stdout is redirected to a file). This test launches a subprocess in that state, waits briefly to see if it produces the looping crash traceback, then terminates it and inspects stderr.

diff --git a/Lib/test/test_pyrepl/test_windows_console.py b/Lib/test/test_pyrepl/test_windows_console.py
index f9607e02c60..963f4db26ba 100644
--- a/Lib/test/test_pyrepl/test_windows_console.py
+++ b/Lib/test/test_pyrepl/test_windows_console.py
@@ -576,6 +576,46 @@ def test_up_vt(self):
                          Event(evt='key', data='up', raw=bytearray(b'\x1b[A')))
         self.assertEqual(self.mock.call_count, 3)
 
+import subprocess
+from tempfile import TemporaryDirectory
+import os
+import time
+
+class WindowsCommandLineTests(unittest.TestCase):
+    def test_for_crash_traceback_with_redirected_stdout(self):
+        script_command = "print('script has run')"
+        stderr_output = ""  # Define in an outer scope
+
+        with TemporaryDirectory() as temp_dir:
+            stdout_path = os.path.join(temp_dir, "WindowsCommandLineTests_stdout.txt")
+
+            with open(stdout_path, "w", encoding="utf-8") as stdout_file:
+                with subprocess.Popen(
+                    [sys.executable, '-i', '-c', script_command],
+                    stdin=None,
+                    stdout=stdout_file,
+                    stderr=subprocess.PIPE,
+                    text=True, encoding='utf-8', errors='replace'
+                ) as process:
+                    time.sleep(3)
+                    if process.poll() is None:
+                        process.kill()
+                    
+                    stderr_output = process.stderr.read()
+
+            with open(stdout_path, "r", encoding="utf-8") as f:
+                stdout_from_file = f.read()
+
+        has_crash_traceback = (
+            "OSError" in stderr_output
+            and "getheightwidth" in stderr_output
+            and len(stderr_output) > 500  # brilliant check for looping traceback
+        )
+
+        if has_crash_traceback:
+            self.fail(
+                "Detected the endless OSError traceback.\n"
+            )
 
 if __name__ == "__main__":
     unittest.main()

main branch:

PS F:\github\cpython001> ./python.bat -m test test_pyrepl
Running Debug|x64 interpreter...
Using random seed: 2091289539
0:00:00 Run 1 test sequentially in a single process
0:00:00 [1/1] test_pyrepl
test test_pyrepl failed -- Traceback (most recent call last):
  File "F:\github\cpython001\Lib\test\test_pyrepl\test_windows_console.py", line 616, in test_for_crash_traceback_with_redirected_stdout
    self.fail(
    ~~~~~~~~~^
        "Detected the endless OSError traceback.\n"
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
AssertionError: Detected the endless OSError traceback.


0:00:20 load avg: 0.32 [1/1/1] test_pyrepl failed (1 failure)

== Tests result: FAILURE ==

1 test failed:
    test_pyrepl

Total duration: 20.1 sec
Total tests: run=236 failures=1 skipped=46
Total test files: run=1/1 failed=1
Result: FAILURE

this branch:

PS F:\github\cpython001> ./python.bat -m test -j2 test_pyrepl test_pyrepl
Running Debug|x64 interpreter...
Using random seed: 1651374118
0:00:00 Run 2 tests in parallel using 2 worker processes
0:00:19 load avg: 1.43 [1/2] test_pyrepl passed
0:00:19 load avg: 1.43 [2/2] test_pyrepl passed

== Tests result: SUCCESS ==

All 2 tests OK.

Total duration: 19.5 sec
Total tests: run=472 skipped=94
Total test files: run=2/2
Result: SUCCESS

[ZeroIntensity]

I think using subprocess makes sense, but I suspect that time.sleep(3) won't be very portable, because different systems might just take more time to start processes. It's a race condition.

It would be best to design a test that just exercises the correct behavior; there's no need for the extra checking of the infinite traceback. If the test hangs on the current main, that's OK--our test suite would time it out and fail. Hanging is close enough to a test failure.

[shuimu5418]

Hi @ZeroIntensity ,

Thank you again for your valuable feedback on the pyrepl test case. I tried to implement your suggestion of testing for correct behavior (a fast exit), but I've run into some very specific constraints and would be grateful for your advice on a better approach.

My goal is to test a bug where the Python interpreter, in interactive mode (-i) with stdout redirected to a file, enters an infinite traceback loop. This test is meant to simulate a real-world scenario.

Through testing, I've found several strict environmental requirements to trigger the bug:

stdout must be a file handle: Redirecting to subprocess.PIPE does not trigger the bug.
stdin must not be a pipe: For pyrepl to activate its advanced mode, stdin cannot be a simple pipe. If I set stdin=subprocess.PIPE (e.g., to send an exit() command), pyrepl detects it, falls back to a basic mode, and the buggy code path is never reached. stdin must therefore be inherited (stdin=None).
The biggest challenge I discovered is a critical timing issue:

With the -i flag, the buggy REPL initialization (which calls getheightwidth) happens after the script from the -c argument has finished.
This means my attempt to use -c "import os; ...; os._exit(0)" failed as a test. The process terminated immediately before the buggy code could be reached, leading to a false-positive pass on the buggy branch.
This leaves me in a difficult position. To trigger the bug, the process must not exit. The bug itself is the process hanging. The only way I've been able to reliably detect this is with a time.sleep pattern, which I know is not ideal due to the race condition you pointed out.

My main question is: could you suggest a more robust, non-racy pattern to solve this?

How can I reliably test for a process that is expected to hang (due to this specific bug), given that I cannot programmatically cause it to exit and must adhere to the strict stdin/stdout constraints? Any guidance or alternative strategy you could offer would be incredibly helpful.

Thanks again for your time and expertise.

---

edit to attach:

Now I am trying to use timeout.

import subprocess
from tempfile import TemporaryDirectory
import os
class WindowsCommandLineTests(unittest.TestCase):
    def test_for_crash_traceback_with_redirected_stdout(self):
        """python.bat -i -c "print('hlwd')" > file.txt"""
        script_command = "print('script has run')"
        
        with TemporaryDirectory() as tmp_dir:
            stdout_path = os.path.join(tmp_dir, "WinCMDLineTests.txt")

            with open(stdout_path, "w") as stdout_file, \
                subprocess.Popen(
                    [sys.executable, '-i', '-c', script_command],
                    stdin=None,
                    stdout=stdout_file,
                    stderr=subprocess.PIPE,
                    text=True, errors='replace'
                ) as process:
                
                stderr_output = ""

                try:
                    process.wait(timeout=3)
                    self.fail("Process exited unexpectedly within the timeout.")
                except subprocess.TimeoutExpired:
                    process.kill()
                    _, stderr_output = process.communicate()

                has_crash_traceback = (
                    "OSError" in stderr_output and
                    len(stderr_output) > 1200
                )
            
                if has_crash_traceback:
                    self.fail("Detected endless OSError traceback."
                          f"\n--- stderr ---\n{stderr_output[:1200]}")

main branch:

PS F:\github\cpython001> ./python.bat -m test test_pyrepl
Running Debug|x64 interpreter...
Using random seed: 2225918861
0:00:00 Run 1 test sequentially in a single process
0:00:00 [1/1] test_pyrepl
test test_pyrepl failed -- Traceback (most recent call last):
  File "F:\github\cpython001\Lib\test\test_pyrepl\test_windows_console.py", line 615, in test_for_crash_traceback_with_redirected_stdout
    self.fail("Detected endless OSError traceback."
    ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
          f"\n--- stderr ---\n{stderr_output[:1200]}")
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: Detected endless OSError traceback.
--- stderr ---
Traceback (most recent call last):
  File "F:\github\cpython001\Lib\_pyrepl\readline.py", line 394, in multiline_input
    return reader.readline()
           ~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\reader.py", line 742, in readline
    self.prepare()
    ~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\historical_reader.py", line 306, in prepare
    super().prepare()
    ~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\reader.py", line 585, in prepare
    self.console.prepare()
    ~~~~~~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\windows_console.py", line 358, in prepare
    self.height, self.width = self.getheightwidth()
                              ~~~~~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\windows_console.py", line 409, in getheightwidth       
    raise WinError(GetLastError())
OSError: [WinError 6] The handle is invalid.
Traceback (most recent call last):
  File "F:\github\cpython001\Lib\_pyrepl\readline.py", line 394, in multiline_input
    return reader.readline()
           ~~~~~~~~~~~~~~~^^
  File "F:\github\cpython001\Lib\_pyrepl\reader.py", line 742, in readline
    self.prepare()
    ~~~~~~~~~~~~^^
  File "F:\github\cpyth

0:00:19 load avg: 0.18 [1/1/1] test_pyrepl failed (1 failure)

== Tests result: FAILURE ==

1 test failed:
    test_pyrepl

Total duration: 20.0 sec
Total tests: run=236 failures=1 skipped=46
Total test files: run=1/1 failed=1
Result: FAILURE

this branch

PS F:\github\cpython001> ./python.bat -m test -j2 test_pyrepl test_pyrepl
Running Debug|x64 interpreter...
Using random seed: 1927743178
0:00:00 Run 2 tests in parallel using 2 worker processes
0:00:19 load avg: 0.02 [1/2] test_pyrepl passed
0:00:19 load avg: 0.02 [2/2] test_pyrepl passed

== Tests result: SUCCESS ==

All 2 tests OK.

Total duration: 19.7 sec
Total tests: run=472 skipped=94
Total test files: run=2/2
Result: SUCCESS