test_pyrepl.test_no_newline() failure on main

Bug report

Bug description:

test_pyrepl started failing for me a day or two ago. According to git bisect, this is the problematic commit:

8a2deea1fc725f8147254f87c6042fcf75a1d03b is the first bad commit
commit 8a2deea1fc725f8147254f87c6042fcf75a1d03b
Author: Jan-Eric Nitschke <47750513+JanEricNitschke@users.noreply.github.com>
Date:   Fri Jan 2 14:04:37 2026 +0100

    gh-128067: Fix pyrepl overriding printed output without newlines (#138732)
    
    Co-authored-by: Łukasz Langa <lukasz@langa.pl>

 Lib/_pyrepl/unix_console.py                                        |  7 +++---
 Lib/_pyrepl/windows_console.py                                     |  7 +++---
 Lib/test/test_pyrepl/test_pyrepl.py                                | 38 +++++++++++++++++++++++++++++++
 Lib/test/test_pyrepl/test_unix_console.py                          | 14 ++++++++++++
 Lib/test/test_pyrepl/test_windows_console.py                       | 14 ++++++++++++
 Misc/ACKS                                                          |  1 +
 .../next/Windows/2025-09-14-13-35-44.gh-issue-128067.BGdP_A.rst    |  1 +
 7 files changed, 76 insertions(+), 6 deletions(-)
 create mode 100644 Misc/NEWS.d/next/Windows/2025-09-14-13-35-44.gh-issue-128067.BGdP_A.rst

CPython versions tested on:

CPython main branch

Operating systems tested on:

macOS

Linked PRs

@ZeroIntensity

What are the failures you're seeing exactly? Which tests are failing exactly? @ZeroIntensity topic-repl ~~pending (I'd skip any other labels because we don't have enough information)~~

Sorry, forgot the key bit.

======================================================================
FAIL: test_no_newline (test.test_pyrepl.test_pyrepl.TestMain.test_no_newline)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/skip/src/python/cpython/Lib/test/support/__init__.py", line 3022, in wrapper
    return func(*args, **kwargs)
  File "/Users/skip/src/python/cpython/Lib/test/test_pyrepl/test_pyrepl.py", line 1909, in test_no_newline
    self.assertIn(expected_output_sequence, cleaned_output)
    ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: 'Something pretty long>>> exit()' not found in 'print(\'Something pretty long\', end=\'\')\r\nexit()\r\nPython 3.15.0a3+ free-threading build (bisect/good-18f3c59e57e5d30218210136060310310b10ff72-2-g09ce5924996:09ce5924996,) [Clang 21.1.8 ] on darwin\r\nType "help", "copyright", "credits" or "license" for more information.\r\n\x1b[?7l>>> print(\'Something pretty long\', end=\'\')\x1b[42D\n\r\x1b[?7hSomething pretty long\x1b[?7l>>> exit()\x1b[10D\n\r\x1b[?7h'

Notes: cleaned_output contained unexpected ANSI sequences, free-threaded build (unlikely to be related).

Failing on 3.14 branch as well.

======================================================================
FAIL: test_no_newline (test.test_pyrepl.test_pyrepl.TestMain.test_no_newline)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/skip/src/python/py3.14/Lib/test/support/__init__.py", line 2935, in wrapper
    return func(*args, **kwargs)
  File "/Users/skip/src/python/py3.14/Lib/test/test_pyrepl/test_pyrepl.py", line 1899, in test_no_newline
    self.assertIn(expected_output_sequence, cleaned_output)
    ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: 'Something pretty long>>> exit()' not found in 'print(\'Something pretty long\', end=\'\')\r\nexit()\r\nPython 3.14.2+ free-threading build (heads/3.14:6583951e7de, Dec 29 2025, 09:10:49) [Clang 21.1.8 ] on darwin\r\nType "help", "copyright", "credits" or "license" for more information.\r\n\x1b[?7l>>> print(\'Something pretty long\', end=\'\')\x1b[42D\n\r\x1b[?7hSomething pretty long\x1b[?7l>>> exit()\x1b[10D\n\r\x1b[?7h'

For the passing case, cleaned_output ends with

Something pretty long>>> exit()\x1b[10D\n\r

For the OP's case, cleaned_output ends with

Something pretty long\x1b[?7l>>> exit()\x1b[10D\n\r\x1b[?7h

The presence of \x1b[?7l prevented the test from passing.

@JanEricNitschke

As noted, this is related to GH-138732. CC @JanEricNitschke @chris-eibl

@ambv, this looks similar to the builbot failure from GH-138732 where \x1b[34h popped up. However, this issue appeared on MacOS and the sequence was \x1b[?7l.

Not sure whether this is tests or stdlib -- auto-wrap mode from ?7h and ?7l may technically change the visual output and cursor position. Please confirm me on that though.

@ZeroIntensity 3.13 3.14 3.15 (and topic-repl of course)

test_no_newline needs a lot of filtering, and that's why I suggested two easier tests additionally. Łukasz already hardened test_no_newline in #143356.

We can try to filter for \x1b[?7l as well, but if test_no_newline gets a maintenance burden, IMHO we can just remove it, too?

@smontanaro

@smontanaro does adding

safe_patterns.append(r'\x1b[?7l')  # auto wrap

fix the issue for you?

@picnixz

If that's the case, let's relabel this as tests @picnixz (not stdlib)

pythongh-143394: filter for turning off automatic margins in test_no_…

pythongh-143394: On macOS, run main PyREPL tests as "Apple Terminal" …

The test is only passing when the GNU readline is enabled in the CPython build.

Why readline matters

The test sends both commands at once via os.write(master_fd, b"print('Something pretty long', end='')\nexit()\n") to a PTY master.

With readline: readline puts the terminal in raw mode (ECHO=off) because it handles character echoing itself.
With kernel ECHO disabled, input written to the PTY master is not echoed back into the read stream. The output
is clean – just the REPL's own output - and "Something pretty long>>> exit()" is found.
Without readline: the basic REPL falls back to plain C fgets(), which doesn't touch terminal settings. The
kernel's ECHO flag stays on, so every byte written to the PTY master is immediately echoed back into the read stream.
Both commands (print(…)\r\nexit()\r\n) are echoed as a block before the REPL processes anything. exit() appears
early in this echo block, never after >>>, and the assertion fails.

How I confirmed this

I wrote a diagnostic script debug_pyrepl_issue.py that checks whether the REPL toggles the ECHO termios flag (indicating readline is active) and then reproduces the test scenario.

Example output on a machine without readline:

$ ./python debug_pyrepl_issue.py
Python: ./python
Platform: FreeBSD 14.1-RELEASE-p4 amd64
Version: 3.15.0a7+

--- Phase 1: detect readline ---
Initial:  ECHO=True  ICANON=True
At prompt: ECHO=True  ICANON=True
ECHO unchanged (readline NOT active)

--- Phase 2: reproduce test_no_newline ---
Test assertion: FAIL ('Something pretty long>>> exit()' NOT found)
result='print(\'Something pretty long\', end=\'\')\r\nexit()\r\nPython 3.15.0a5+ (heads/main:979d92fefc0, Jan 25 2026, 06:54:33) [GCC 14.2.0] on linux\r\nType "help", "copyright", "credits" or "license" for more information.\r\n>>> Something pretty long>>> '

readline MISSING -> ECHO on -> echo pollution -> test FAILS

Example output on a machine with readline:

$ ./python debug_pyrepl_issue.py 
Python: /home/arf/Desktop/cpython-pyrepl-issue/python
Platform: Linux 6.18.20-1-lts x86_64
Version: 3.15.0a5+ (heads/main:979d92fefc, Jan 25 2026, 16:09:18) [GCC 15.2.1 20260103]

--- Phase 1: detect readline ---
Initial:  ECHO=True  ICANON=True
At prompt: ECHO=False  ICANON=False
ECHO changed: True -> False (readline is active)

--- Phase 2: reproduce test_no_newline ---
Test assertion: PASS ('Something pretty long>>> exit()' found)

readline present -> ECHO off -> clean output -> test PASSES

This is directly visible in the buildbot configure logs:

FreeBSD 15 buildbot (passes): checking for stdlib extension module readline... yes
FreeBSD 16 buildbot (fails): checking for stdlib extension module readline... missing

On Linux, I confirm that test_pyrepl pass with the readline module, and fails without the readline module (ex: run find -name "readline*so" -delete in the Python source tree).

pythongh-143394: Skip pyrepl test_no_newline() basic REPL if readline…

I wrote #147973 to skip the test on the basic REPL if the readline module is missing.

gh-143394: Skip pyrepl test_no_newline() basic REPL if readline is mi…

pythongh-143394: Skip pyrepl test_no_newline() basic REPL if readline…

The latest known test_no_newline() failure should be fixed by 97babb8 (backport to 3.13 and 3.14 will follow once the CI pass on the PRs). I close the issue.

[3.13] gh-143394: Skip pyrepl test_no_newline() basic REPL if readlin…

@vstinner

Thanks @vstinner. Fixed for me, even on my very pokey Raspberry Pi (which I only test on "for fun").

[3.14] gh-143394: Skip pyrepl test_no_newline() basic REPL if readlin…

pythongh-143394: Skip pyrepl test_no_newline() basic REPL if readline…

Navigation Menu

Repository navigation

test_pyrepl.test_no_newline() failure on main #143394

Bug report

Bug description:

CPython versions tested on:

Operating systems tested on:

Linked PRs

24 remaining items

Why readline matters

How I confirmed this

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

test_pyrepl.test_no_newline() failure on main #143394

Description

Bug report

Bug description:

CPython versions tested on:

Operating systems tested on:

Linked PRs

Activity

johnslavik commented on Jan 3, 2026

smontanaro commented on Jan 3, 2026

johnslavik commented on Jan 3, 2026

smontanaro commented on Jan 3, 2026

johnslavik commented on Jan 3, 2026

johnslavik commented on Jan 3, 2026

chris-eibl commented on Jan 4, 2026

chris-eibl commented on Jan 4, 2026

johnslavik commented on Jan 4, 2026

24 remaining items

furkanonder commented on Mar 29, 2026

Why readline matters

How I confirmed this

vstinner commented on Apr 1, 2026

vstinner commented on Apr 1, 2026

vstinner commented on Apr 2, 2026

smontanaro commented on Apr 3, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions