lexical_flutter 1.4.0

Flutter rendering, selection and input layer for lexical_core: a dirty-set reconciler, one render object per block, InlineSpan building with a bidirectional offset map, theming and IME integration.

Changelog

1.4.0

The caret no longer shrinks and rides up when a space is typed at the end of a line. getFullHeightForCaret reports the height of the glyph run at the position, and for trailing whitespace the engine reports that run without the style's height multiplier — 16 logical pixels where the line is 27. The visible result was a caret that jumped the moment a space was typed, which reads as the space having broken something rather than as a measuring artefact. RenderLexicalBlock.caretRect now never returns a caret shorter than the line it sits on.

1.3.0

Dragging a selection no longer stutters. Four things ran on every pointer move, and none of them had to:

• The caret was scrolled into view on every move. While a pointer is dragging, the user is looking at where they are pointing — and this was the most expensive thing on the path, because it walks to the enclosing scrollable and repaints everything in between.
• The block's flat text and offset map were rebuilt from the model on every move, to hand the platform an editing value whose text had not changed. $buildEditingWindow now takes reuseOffsets, and LexicalInput.syncToModel passes the previous map when the commit moved only the selection.
• The caret blink timer was cancelled and re-armed on every move, for a caret that is not drawn at all while a range is selected.
• Resolving the selection into per-block spans walked every leaf it covered rather than every block (fixed in lexical_core 1.3.0).

parseCssDeclarations now defers to lexical_core's getStyleObjectFromCss. Splitting the string on ; and : — which is what it did — truncates background: url(a;b) and mangles a quoted font stack, and it was a second answer to a question the core already answers for the value that has to round-trip through Lexical web.

1.2.0

No library changes. The packages version in lockstep — they are one library split for pay-for-what-you-use, not a set of independent projects — so this release keeps the set on one version.

1.1.0

• Fixed: a press that lands inside a block's padding — a cell's, or the empty area below the last block — now places the caret in the nearest block instead of being ignored. Ignoring it was not neutral: the selection kept the anchor it already had, so the next drag spanned from wherever the user had last clicked. In a table that showed up as a merge covering rows nobody selected.
• LexicalEditorScope: the editor drawing a document, reachable from anything drawn inside it — a decorator, a block layout, an overlay.

1.0.0

First stable release; semantic versioning applies from here. The Flutter layer: per-block render objects, span and offset map built in one walk, IME, selection handles, token widgets and pointer interaction.

The entries below record how it got here.

0.1.0-dev.2

• LexicalInteraction: hover and tap for links, mentions and hashtags, keyed on node type string so this package still imports none of them. Reports the interactive ancestor rather than the text node under the pointer, confirms the hit against the node's own line boxes, and hands callbacks a snapshot — key, type, text, serialized fields and global bounds — that stays valid outside the read it came from.
• LexicalTheme.tokenBuilders: render a token text node as a widget, so a mention can be a rounded chip. The node stays text in the model — same JSON, same atomic delete — and only token-mode nodes qualify, because the widget occupies one position while the node holds a whole label.
• Selection handles, the platform context menu and the magnifier, built on Flutter's own TextSelectionControls and AdaptiveTextSelectionToolbar.
• RemoteSelection: other people's carets and ranges, painted in their own colour.
• LexicalBuilder, for rebuilding a widget after every commit without tripping over a commit that lands during a build.
• Fixed: a text node's type is styled again, so a theme entry for mention or hashtag reaches the node it names.

0.1.0-dev.1

Milestone M2 — the read-only renderer.

Added

• LexicalDocument, a widget rendering an editor's document, with rebuilds driven by each commit's dirty set rather than by rebuilding from the state. Untouched blocks are reused by reference.
• RenderLexicalBlock: one render object per block, with inline widget children for decorators, selection painting and a caret that repaints without relayout.
• SpanBuilder, producing an InlineSpan and a BlockOffsetMap in a single walk so the two cannot drift apart.
• BlockOffsetMap, the bidirectional map between a block's flat text offsets and (NodeKey, offset, type) model points, with explicit boundary rules.
• LexicalTheme: text-format styles, per-type block styles keyed on the type string so this package never imports the feature packages, marker builders for list bullets and checkboxes, and an injectable CSS resolver.
• BlockRegistry, tracking mounted blocks so the input layer can locate a block's geometry without keeping a key per document node.
• Viewport culling through a lazily built list.

Known limitations

• Editing arrives in M3: no IME, no keyboard handling, no selection gestures.
• Semantics are not yet emitted per block.

Added — editable

• LexicalEditable: focus, pointer, keyboard, caret, selection painting and the composing underline, with caret and selection pushed straight onto render objects so a blink never rebuilds a widget.
• LexicalInput, a DeltaTextInputClient. The platform is given a bounded window around the caret rather than the document, with newline sentinels standing in for the block boundaries — without them, backspace at the start of a paragraph produces no delta at all and blocks can never be merged from a soft keyboard.
• The model stays authoritative about the caret: after a text-changing delta the corrected value is pushed back, which is what lets an atomic token delete whole when the platform thinks one character went.
• A delta computed against a value we no longer hold is refused and our value re-stated, rather than corrupting the document over a transient race.
• $resolveDocumentSelection and flatSelectionFor, mapping the model's selection onto the blocks that paint it.
• buildModelOffsets: the model-side offset map, deliberately distinct from the renderer's — handing a presentational uppercase to an input method would have it autocorrect text that does not exist.
• LexicalTheme.blockStyleResolver, for types whose presentation depends on a field rather than only on their type string.