FreeBSD: pty.sh fail: POSIX sh 101(C)

[McDutchie]

On this system:

$ uname -vm
FreeBSD 12.1-RELEASE-p6 GENERIC  amd64

The following regression test failure occurs:

        pty.sh[291]: POSIX sh 101(C): line 323: expected "^backslash-erase=:\r?\n$", got EOF

The test in question:

ksh/src/cmd/ksh93/tests/pty.sh

Lines 291 to 333 in f811482

	tst $LINENO <<"!"
	L POSIX sh 101(C)
	# If the User Portability Utilities Option is supported and shell
	# command line editing is supported: When in insert mode a control-V
	# causes the next character to be inserted even in the case that the
	# character is a special insert mode character.
	# Testing Requirements: The assertion must be tested with at least the
	# following set of characters: <newline>, erase, interrupt, kill,
	# control-V, control-W, end-of-file, backslash \ (followed by erase or
	# kill) and <ESC>.
	d 10
	p :test-1:
	w stty erase ^H intr ^C kill ^X
	p :test-2:
	w echo erase=:\cV\cH:
	u ^erase=:\r?\n$
	p :test-3:
	w echo kill=:\cV\cX:
	u ^kill=:\cX:\r?\n$
	p :test-4:
	w echo control-V=:\cV\cV:
	u ^control-V=:\cV:\r?\n$
	p :test-5:
	w echo control-W:\cV\cW:
	u ^control-W:\cW:\r?\n$
	p :test-6:
	w echo EOF=:\cV\cD:
	u ^EOF=:\004:\r?\n$
	p :test-7:
	w echo backslash-erase=:\\\cH:
	u ^backslash-erase=:\r?\n$
	p :test-8:
	w echo backslash-kill=:\\\cX:
	u ^backslash-kill=:\cX:\r?\n$
	p :test-9:
	w echo ESC=:\cV\E:
	u ^ESC=:\E:\r?\n$
	p :test-10:
	w echo interrupt=:\cV\cC:
	u ^interrupt=:\cC:\r?\n$
	!

[JohnoKing]

The test always fails in vi mode or when there isn't an editing mode enabled, but it will pass in emacs mode. Placing the following line at the start of the regression test allows it to pass:

w set -o emacs

Replacing set -o emacs with set -o vi or set +o vi; set +o emacs causes the test failure to occur on Linux as well.

[JohnoKing]

The regression test that fails is testing the escaping backslash feature (see ksh-community/ksh#8 (comment) for what it does). If escaping backslashes are disabled with a patch (like the one appended below) the test will always fail on Linux.

ksh93-escbackslash.patch.txt

[McDutchie]

Thanks, @JohnoKing, for figuring out what that test does exactly.

Just found this. After doing

stty erase ^H kill ^X

on the real terminal from which you're running the regression tests, this regression test failure (and another one that was introduced in a0dcdee) goes away on FreeBSD. The regression test itself already issues that command, but apparently, FreeBSD's stty does not work correctly on a pty pseudoterminal…