linewrap then backspace produces unexpected visuals, missing characters

[fredless]

Environment

Windows build number: Microsoft Windows [Version 10.0.18362.175]
Windows Terminal version (if applicable): built using d82eab4

Steps to reproduce

• open terminal with following profile:

{
    "globals" : 
    {
        "alwaysShowTabs" : false,
        "defaultProfile" : "{9d2406b9-7e40-4e2b-a20d-82dc61c4cc24}",
        "initialCols" : 80,
        "initialRows" : 60,
        "requestedTheme" : "dark",
        "showTabsInTitlebar" : true,
        "showTerminalTitleInTitlebar" : false
    },
    "profiles" : 
    [
        {
            "acrylicOpacity" : 0.80000001192092896,
            "closeOnExit" : true,
            "colorScheme" : "Solarized Dark",
            "commandline" : "cmd.exe",
            "cursorColor" : "#FFFFFF",
            "cursorShape" : "bar",
            "fontFace" : "Consolas",
            "fontSize" : 10,
            "guid" : "{9d2406b9-7e40-4e2b-a20d-82dc61c4cc24}",
            "historySize" : 9001,
            "name" : "cmd",
            "padding" : "0, 0, 0, 0",
            "snapOnInput" : true,
            "startingDirectory" : "%USERPROFILE%",
            "useAcrylic" : true
        }
    ]
}

• input characters until the line wraps, backspace once, and enter an additional character

Expected behavior

• Cursor should return to the previous position following the last character on the previous line after hitting backspace.
• Additional input character (sent after backspace) should show up after the last character on the previous line.

Actual behavior

• Cursor returns to previous line, but with a whitespace between the last character input and the cursor position
• Additional input character does not show up in terminal, although it is recognized

'..789ab' is received as input in the example below, even though it is not shown:

[DHowett-MSFT]

This is probably something down in the pseudoconsole layer; we probably have a duplicate for it, but we can't find it.

[fredless]

Should have also mentioned in case it may be helpful, identical behavior was observed on 73ad742

Didn't report then as that was my first exposure, assumed either I couldn't be the only one (or that I was). I just repulled and built in hopes it might have been squashed. I searched then and now before submitting but didn't see a specific match.

[fredless]

I can't be the only one.. can I?

[zadjii-msft]

@fredless you're not the only one - this is a reasonable bug report with clear repro steps. Doesn't seem like there's much that needs to be added until someone starts working on a fix.

[zadjii-msft]

After playing with this some more, this looks like it only repros in the Terminal. This might actually just be another variant of WriteCharsLegacy2ElectricBoogaloo. I doubt we're putting the cursor in the right spot when we backspace around a linewrap like this.

cc @miniksa.

[miniksa]

Backspace around a line is a WriteCharsLegacy handled thing in the conhost. So your reasoning is sound. This should be linked to #780. Maybe not resolved until we're sure that it will resolve this too.

[riverar]

Still seeing this issue in 0.8.10091.0.

[DHowett-MSFT]

Thank you for the report. If we ever think we have explicitly fixed this bug, we’ll be sure to close this issue.

[zadjii-msft]

For the record, I've merged a combo of branches for #4354, #4289, and dev/migrie/f/conpty-wrapping-003 all together, and this still repros. It only repros in cmd.exe, and not bash however. I'm thinking it's possible that it's related shockingly to another bug where the cursor doesn't get moved by conpty on a frame where only the cursor changed position - maybe some combo of #4102/#3202. I'll keep digging.

[zadjii-msft]

Hokay I've got a fix for this that I'm a little wary of, but I'm not sure it can be de-tangled from dev/migrie/f/conpty-wrapping-003. I'll investigate.

[zadjii-msft]

@<future me>:

This is the other side of the #357 coin actually.

Changing the fix for #357 from

terminal/src/renderer/vt/paint.cpp

Lines 452 to 467 in 814f4ca

	// Update our internal tracker of the cursor's position.
	// See MSFT:20266233
	// If the cursor is at the rightmost column of the terminal, and we write a
	// space, the cursor won't actually move to the next cell (which would
	// be {0, _lastText.Y++}). The cursor will stay visibly in that last
	// cell until then next character is output.
	// If in that case, we increment the cursor position here (such that the X
	// position would be one past the right of the terminal), when we come
	// back through to MoveCursor in the last PaintCursor of the frame,
	// we'll determine that we need to emit a \b to put the cursor in the
	// right position. This is wrong, and will cause us to move the cursor
	// back one character more than we wanted.
	if (_lastText.X < _lastViewport.RightInclusive())
	{
	_lastText.X += static_cast<short>(columnsActual);
	}

to

    if (_lastText.X < _lastViewport.RightExclusive())
    {
        _lastText.X += static_cast<short>(columnsActual);
    }

fixes the Terminal, but re-breaks using conhost as the Terminal (interop/vtpipeterm).

Maybe the Terminal is just wrong here (wouldn't surprise me). Especially considering this bug doesn't repro in conhost as a terminal. Perhaps this is delayed eol wrap? I'll look at the traces, see what's going on.

[ghost]

🎉This issue was addressed in #4403, which has now been successfully released as Windows Terminal Preview v0.9.433.0.:tada:

Handy links:

• Release Notes
• Store Download