gh-144259: Fix Windows EOL wrap by syncing real console cursor

[hyongtao-code]

Long log:

Windows VT terminals do not consistently wrap the cursor when a line
exactly fills the terminal width.
Previously we assumed a wrap always happened, which could desynchronize
the logical cursor from the real console cursor and break subsequent
cursor movement.
This change queries the real cursor position and updates posxy
accordingly, and adds regression tests for both wrap and no-wrap cases.

Video with the patch

fix-issue-144259.mp4

Signed-off-by: Yongtao Huang yongtaoh2022@gmail.com

• Issue: Inconsistent display of long multiline pasted content in the REPL #144259

[hyongtao-code]

Thanks for the review! I agree with your point, so I’ve removed the VT-specific references and updated the tests to exercise both VT and legacy console modes.

[chris-eibl]

As you've seen, many Windows tests fail now, because in headless mode GetConsoleScreenBufferInfo() fails with The handle is invalid.

Instead of swallowing the error, maybe introduce something like

_HasWrappedToNextRow(self, y)

Then, we can simply mock this for most of the tests.

And you can also write an explicit test for _HasWrappedToNextRow by mocking the "bare" GetConsoleScreenBufferInfo() just for this one test?

[hyongtao-code]

Thanks for the suggestion — these failed cases were indeed quite tricky. I decided to follow your proposed approach to fix the failing cases.

[chris-eibl]

Suggested change

	if not GetConsoleScreenBufferInfo(OutHandle, info):
	err = get_last_error()
	if err == 6: # ERROR_INVALID_HANDLE
	return None
	raise WinError(err)
	if not GetConsoleScreenBufferInfo(OutHandle, info):
	raise WinError(get_last_error())

Let's raise again? We can handle failing CI like for

cpython/Lib/test/test_pyrepl/test_windows_console.py

Line 43 in 1834741

console._getscrollbacksize = MagicMock(42)

by adding

console._has_wrapped_to_next_row = MagicMock(False)

That's what I had in mind in #144297 (comment)

Instead of swallowing the error [...] we can simply mock this for most of the tests.

In case you like this suggestion, please also adapt the annotation and docstring. test_returns_none_on_invalid_handle and test_raises_on_unexpected_error are then no longer needed.

Thanks for bearing with me.

[hyongtao-code]

No problem at all, thanks for the clarification and your help — I’ll keep working on these over the next day or two.