The current pty library has the following issues:

1. Does not set slave termios. Documented in the source.

2. Does not set initial slave window size. Documented in the source. Does not handle SIGWINCH. See bpo-41494, bpo-41541. This is essential in the following practical scenarios: i. creating split windows/panes while using a terminal multiplexer; ii. when resizing GUI terminal emulator window, especially relevant when using tiling window managers; iii. resizing an ansi-term window created inside a GNU Emacs frame.

3. Does not perform signal handling. Signals must be blocked during sensitive portions of code.

4. Hangs on FreeBSD. See bpo-26228.

5. Includes deprecated functions pty.master_open(), pty.slave_open().

6. In pty.fork(), the fallback code should try using TIOCSCTTY first. It is still using the old method of opening a tty to make it the controlling tty. Currently even SysV based systems provide TIOCSCTTY. See https://stackoverflow.com/questions/51593530/code-explanation-for-glibc-login-tty-function-openttyname-immediately-f

The current version of pty.spawn() uses pty.fork() internally. However, pty.fork() closes slave and only returns (pid, master_fd). To update winsize, access to slave is necessary. Further, slave termios must be properly set. The proposed modifications do this by implementing a login_tty(3) based function ( tty.login() ), and using that in pty.spawn() instead of pty.fork(). tty.login() tries TIOCSCTTY before falling back to the old SysV method because Python currently does not provide an interface to the native login_tty(3).

7. tty.setraw() is called right after tty.tcgetattr(). This increases redundancy of code because tty.setraw() itself makes an identical tty.tcgetattr() call.

8. Requires testing/porting to more platforms. Solaris, Illumos, macOS, Cygwin, etc. Windows ConPTY?

9. There should be an option in pty.spawn() to turn off slave's ECHO flag. For example, when it is being used in a pipe. See util-linux/util-linux@1eee1ac and util-linux/util-linux@75ccd75#diff-3834a3d25eeaf20d9d0dcb05a46995f6

10. Tests are incomplete. Tests consider OSes such as Tru64 but not {Free/Net/Open/...}BSD.

Find ongoing work here: https://github.com/8vasu/pypty2

In addition to the above, if a major revision is made to pty, I'd suggest also addressing the issue of "master/slave" terminology, and replace it with something comparable like "parent/child". There's an open devguide issue (python/devguide#605) to more explicitly state terms to avoid, and support for avoiding usage of "slave/master" seems uncontroversial (especially in any new code).

Makes sense. I will happily make a change of terminology in the pypty2 repository after the most desirable alternative is determined based on the choice of the majority. I think 'mother/son' sounds cute while still retaining the same initials as before; people used to the older
terminology will find this easy to remember. Terminology such as parent/child and server/client might make it a little confusing.

New changeset c13d899 by Soumendra Ganguly in branch 'master':
bpo-41818: Updated tests for the standard pty library (GH-22962)
c13d899

This change broke x86 Gentoo buildbots: bpo-42463.

#23514 has the fix, waiting for all buildbots finish before pressing "Merge" button.
Gentoo bots are green.

#23514 has the fix, waiting for all buildbots finish before pressing "Merge" button.

Thanks for working on a fix :-)

In addition to the above, if a major revision is made to pty, I'd suggest also addressing the issue of "master/slave" terminology

In bpo-34605, I chose to leave the pty module unchange since it *seems* like "master_fd" and "slave_fd" is part of the API. See my rejected PR 9100.

More recently, I discussed with a glibc maintainer who is open to change the openpty() manual page to avoid "master" and "slave" terms.

In fact, these terms are not part of the API. They are just variable names in a manual page.

"parent" and "child" terms would work here. I'm not sure of the exact relationship between the two file descriptors.

New changeset 87f7ab5 by Andrew Svetlov in branch 'master':
bpo-41818: test_openpty succeed on Gentoo, don't expect to fail on this platform (GH-23514)
87f7ab5

Moving my notes from PR23514 to here, the issue that that PR addressed is not Gentoo-specific; here's a simple reproducer on Ubuntu:

./python -c "import pty;pty.spawn('./python -m test -v test_pty'.split())"

The reproducer was helpful. #23526 should fix this issue.

This change also broke Solaris (SunOS), where (similarly to BSDs and Darwin) OSError is not raised in the new test_master_read() test.

Adding or PLATFORM == "SunOS" into the expectedFailureOnBSD function fixes this issue, but it's no longer just BSDs and it's derivates.

New changeset f5a19ea by Soumendra Ganguly in branch 'master':
bpo-41818: Make test_openpty() avoid unexpected success due to number of rows and/or number of columns being == 0. (GH-23526)
f5a19ea

This is actually good news. I had not tested the code on Solaris; this confirms my suspicion that Linux is the only platform that "behaves differently". Sadly, the current Lib/pty.py code depends on such "different behavior" to exit from pty.spawn()'s copy loop, which is why it hangs on the BSDs, macOS, and now we know that it (probably) also hangs on Solaris. Adding 'or PLATFORM == "SunOS"' is the correct thing to do. I am working on this now.

PR-23533 should fix the test_master_read() issue on Solaris. Actually, instead of adding 'or PLATFORM == "SunOS"', I ended up removing the whole line and replaced it with 'if platform.system() != "Linux"' because I expect that test to fail on every platform that is not Linux.

Update: I closed PR 23533. PR 23536 is much better. It will help us detect exact behavior on each platform, which is necessary for making pty.spawn() successfully exit its copy loop.

New changeset 74311ae by Soumendra Ganguly in branch 'master':
bpo-41818: Fix test_master_read() so that it succeeds on all platforms that either raise OSError or return b"" upon reading from master (GH-23536)
74311ae

I noticed an issue in one of the newly added tests; see #68307

Thank you for the fix. That test was created by modifying an existing test which already had that issue; it is documented in a comment by user nnorwitz in the file. If your solution resolves the problem for all the tests in "class PtyTest", then can you please also remove that comment?

New changeset 65cf1ad by Petr Viktorin in branch 'master':
bpo-41818: Close file descriptors in test_openpty (#GH-24119)
65cf1ad

New changeset ae224bb by Soumendra Ganguly in branch 'main':
bpo-41818: Add termios.tcgetwinsize(), termios.tcsetwinsize(). (GH-23686)
ae224bb

New changeset 245f1f2 by Gregory P. Smith in branch 'main':
bpo-41818: ++ termios versionadded markers. (GH-27987)
245f1f2

#29658 merged to add os.login_tty(fd) in 3.11.