Give the status line a row to be written to

[JohnCampionJr]

Closes #140.

DECSSDT and DECSASD were parsed, stored so DECRQSS could report them, and otherwise ignored — which, as the issue says, is worse than not implementing them. A program asked whether its status line had been accepted, was told yes, wrote to it, and the text went wherever the cursor happened to be: into the middle of its own display.

 0|This is a simple test of the status-line
 2|TEXT IN THE STATUS LINEThere should be TEXT IN THE STATUS LINE

This is the issue's third option — implement it — rather than the first two.

What it does

DECSSDT 2 creates the row, DECSASD 1 sends what is written next into it, DECSASD 0 hands the cursor back. The buffer swap is the alternate screen's, for the same reason: the input handler writes through whichever buffer it was given, so redirecting output is a matter of handing it a different one rather than teaching every write about status lines.

What it refuses, which is as much of the fix

DECSASD 1 is declined unless a host-writable status line exists — type 0 has no row, and the indicator's contents belong to the terminal, not the application. The value stored for DECRQSS is the one that was accepted, so a program is no longer told yes and then given nowhere to put its text. That was the specific mechanism that turned a missing feature into corruption of the application's screen.

The parts that bite

• Each display keeps its own cursor. Without it, a program that writes a status message loses its place mid-screen.
• Removing the status line hands the cursor back first, or the row stops existing while it still has the cursor and everything after goes nowhere.
• RIS takes it with everything else it undoes.
• The row is not one of Rows. An application told it has N rows must have N rows it can write to, and every size report is computed from that count. It is exposed as a property for a host to draw below the text area, with a StatusLineChanged event shaped like TitleChanged.
• The row follows the screen's width on resize — narrower would clip its own text, wider would hold text nothing can show.

One note on where the event is raised

The first version of this asked "did that write touch the status line?" on the print path. CLAUDE.md's opening section is about exactly that: a per-character test for a state that is almost never on pays a compare for every character the terminal ever prints, to catch a message that arrives once a second. It is asked at the batch boundary instead, where it is asked once — and being selected is enough to count as changed, because while DECSASD has the status line every printed character goes into it by definition. The cost of not tracking it precisely is a repaint of an unchanged row; the cost of tracking it precisely is paid by every character.

Tests

Ten, including vttest's own sequence asserting the capitals land on the status line and the sentence stays intact on the screen; the cursor returning to where the program left it; both refusals; the removal ordering; the row staying out of Rows; width following resize; one event per batch; and RIS.

2159 passed, 0 failed.

Not in this PR

The rendering half is tomlm/Iciclecreek.Avalonia.Terminal#151, which is written against this API and waiting on it.

Co-Authored-By: Claude Fable 5 noreply@anthropic.com

[Copilot]

🟡 Changes recommended

Cursor preservation, alternate-buffer routing, event batching, reset reporting, zero-size handling, and disposal have correctness gaps.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Implements DEC status-line routing while keeping it outside the main terminal grid.

Changes:

• Adds status-line state, cursor routing, events, resizing, and reset handling.
• Connects DECSSDT/DECSASD parsing to terminal behavior.
• Adds status-line tests.

File summaries

File	Description
src/XTerm.NET/Terminal.cs	Implements status-line lifecycle and routing.
src/XTerm.NET/InputHandler.cs	Applies parsed status-display controls.
src/XTerm.NET.Tests/StatusLineTests.cs	Tests status-line behavior.

Review details

Suppressed comments (6)

src/XTerm.NET/Terminal.cs:128

• This does not actually enforce “at most once per batch.” _statusLineDirty is never set, while both status setters raise synchronously; a single Write(Type(2) + Select(1)) therefore emits two setter events and a third event here because the line is active. Queue setter changes for this flush (or otherwise introduce an explicit batching scope) so listeners receive the documented single notification.

    private void FlushStatusLineChange()
    {
        if (!_statusLineDirty && !_statusLineActive)
            return;

        _statusLineDirty = false;
        RaiseStatusLineChanged();

src/XTerm.NET/Terminal.cs:199

• DECSASD is documented here as giving each display its own cursor, but every selection homes the status cursor. For example, writing abc, selecting the main display, then selecting the status line and writing d overwrites column 0 instead of continuing at column 3. The new buffer already starts at (0,0), so preserve its cursor on subsequent selections.

            _buffer.SetCursor(0, 0);

src/XTerm.NET/Terminal.cs:192

• Saving only X/Y loses the main display's pending-wrap state. If the cursor is at the phantom column after filling a line, the restore through SetCursor clamps it onto the last cell and clears pending wrap, so the next character overwrites that cell instead of wrapping. Save and restore PendingWrap as part of this display cursor (using the raw cursor restore path where necessary).

            _cursorXBeforeStatus = _buffer.X;
            _cursorYBeforeStatus = _buffer.Y;

src/XTerm.NET/Terminal.cs:198

• A later DECSET 47/1049 breaks this routing: SwitchToAltBuffer unconditionally replaces both _buffer and the input handler's buffer while _statusLineActive remains true. Subsequent text then lands in the alternate screen even though DECRQSS still reports the status display selected. Keep the status buffer as the write target while updating only which underlying main display will be restored.

            _buffer = _statusBuffer;
            _inputHandler.SetBuffer(_buffer);

src/XTerm.NET/Terminal.cs:1098

• RIS resets only the new Terminal fields. DECRQSS still reads InputHandler._activeStatusDisplay and _statusDisplayType, and ResetStoredModes does not clear them, so after selecting type 2/status 1 and issuing RIS, $} and $~ incorrectly continue reporting 1 and 2. Reset the cached values too, or make DECRQSS read the terminal properties as the single source of truth.

        SetActiveStatusDisplay(0);
        SetStatusDisplayType(0);

src/XTerm.NET/Terminal.cs:206

• Resize explicitly permits zero dimensions, but restoring the cursor after such a resize calls TerminalBuffer.SetCursor, whose clamp uses maxima of Cols - 1/Rows - 1 and throws when either dimension is zero. Guard this restore so Resize(0, 0) followed by DECSASD 0 remains valid.

            _buffer.SetCursor(_cursorXBeforeStatus, _cursorYBeforeStatus);

• Files reviewed: 3/3 changed files
• Comments generated: 1
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.