Text creation — design

An SVG editor that can create rectangles, ellipses, and lines should also let the author create text. Text was the one primitive deferred from the first insertion pass, because it does not share the interaction model the geometric shapes use. This note states what text creation is and why it takes the shape it does.

Why text is not like the other shapes

The geometric primitives have intrinsic size: the author presses, drags, and the drag distance sets the width and height (or the radii, or the endpoints). The shape exists at whatever size the drag describes.

Text has no intrinsic size in SVG. A text element's extent is a consequence of its string content, its font, and its font size — not a quantity the author sets by dragging a box. There is no "empty text of size 200×80". The press-drag-size gesture is therefore meaningless for text: there is nothing for the drag distance to set.

The interaction model: click to place, type immediately

Creating text is a single action. The author indicates where, a caret appears there, and input begins at once. This matches every text tool the author already knows — point, then type. There is no intermediate "empty placeholder" the author must then fill; placing and editing are the same gesture.

1. The author selects the text tool and clicks a point on the canvas.
2. A text element is created at that point with the editor's default text appearance, and the inline content editor opens on it immediately with the caret active.
3. The author types. Editing ends on exit — clicking away or pressing Escape.

Why there is no drag-to-define-a-box

In other tools, a drag gesture defines a text box: a fixed-width region into which text wraps. The editor's text model is single-line positioned text — a text element anchored at a point, sized only by its content and font. A single-line element has no width to drag out, so the drag-to-box gesture has nothing to set; text creation is therefore click-only.

Auto-wrapped text is a genuinely different model. SVG 2 does define wrapping (the inline-size and shape-inside properties give a text element a wrapping width), so a drag-defined text box is expressible in the format — it is simply a separate creation affordance for a separate model, with its own caret, selection, and reflow behavior. That model is deliberately out of scope here; this note is about creating single-line text. If wrapped text is taken up later, it gets its own gesture (a drag that sets inline-size), orthogonal to this decision.

Invariant: empty text is not a valid persistent state

A text element with no content renders nothing. It cannot be seen, and it is effectively impossible to select on the canvas — there is no painted geometry to aim at. An empty text element is indistinguishable from "no element" to a viewer, yet it lingers in the file as noise and as an unselectable trap.

The editor enforces a single, unconditional rule:

On exit from text content-editing, a text element whose content is empty is removed. "Empty" means zero-length text content — a typed space is authored content and is kept.

This is a general fact about text, not a special case of the creation tool. It fires the same way regardless of how the element reached empty:

• a freshly created text element the author placed but never typed into,
• a pre-existing text element whose content the author cleared before exiting, and
• a pre-existing element that was already empty when the author entered it and left unchanged.

In every case the empty text is treated as a deletion intent.

The rule is deliberately unconditional rather than firing only on a transition to empty. Entering content-editing on an element is itself the author's "I am authoring this" signal; an empty text carries no visible content for the canvas author to act on, so the editor does not preserve one on exit. The accepted consequence: a source element that happened to be empty — even one carrying an id, an animation target, or a script-filled slot — is removed if the author opens it for editing and exits. This trades a corner of round-trip fidelity for a simpler, more predictable rule; it is a chosen tradeoff, not an oversight. (Loading a file never triggers it — the element is removed only if the author actually enters and exits content-editing on it.)

Undo contract

What undo restores depends only on whether the element existed before the gesture — not on how it became empty:

• Freshly placed, never filled. Creation and the empty exit are one abandoned gesture. After exit there is no element and no committed history step — undo does not resurrect an empty element, because from the document's perspective nothing happened. This mirrors how an abandoned drag of any insertion tool leaves no trace.
• Pre-existing element removed. The element existed before the gesture — whether it held content the author cleared, or was already empty on entry. Removing it on exit is a deletion: a single undoable step, and undo restores the element exactly as it was before the gesture (with its original content, empty or not, and all its attributes).

The throughline: undo returns the document to the state before the gesture began. For a creation that never produced visible content, that state is nothing. For the removal of a pre-existing element, that state is the original element.

Default appearance

A newly created text element carries the editor's default text appearance — a default font family, font size, and fill — so it is visible and editable the instant it is placed. These are editor defaults, the same class of decision as the default fill of a new rectangle, not parameters the author supplies through the gesture.

Font resolution — mapping a family name to real metrics and glyphs — remains a host-provided capability. The defaults here are only the initial attribute values the new element starts with.

Rejected alternatives

• Drag-to-define-a-box. Out of scope, not impossible — a drag box maps to SVG 2 wrapped text (inline-size / shape-inside), which is a separate model from the single-line text this tool creates. It gets its own gesture if and when that model is taken up.
• Placeholder text (drop the literal word "Text", let the author edit it later). Rejected — it writes content the author did not author, pollutes the document, and forces a select-then-edit second step that click-to-caret avoids.
• Keep empty text elements. Rejected — they are invisible, effectively unselectable, and accumulate as file noise (see the invariant above).

Relationship to existing content editing

Editing the content of an existing text element is already part of the editor. The creation tool does not add a second editing path: it creates the element and then enters that same content-editing flow. The empty-equals-delete invariant is shared by both entry points — creating new text and editing existing text exit through the same rule.

Status: implemented (click-only, single-line). Wrapped/multi-line text and the tool-axis cleanup remain deferred (see the package TODO.md).