@adecore/editor-core

The document model of a code editor, with no DOM and no dependencies. DocumentModel holds the text, the selections, the undo history and the editing commands: typing with bracket pairing, smart Enter, comment toggles, line moves, multiple carets, search and folding. It runs in a browser, Bun and Node.

bun add @adecore/editor-core

import { DocumentModel } from '@adecore/editor-core';

const document = new DocumentModel('const greeting = "hello";\n');

document.setSelections([{ anchor: 26, head: 26 }]);
document.typeText('(', { language: 'typescript' });
document.getText(); // 'const greeting = "hello";\n()'
document.undo();

@adecore/editor draws a model on screen and turns keys into its commands. Use this package alone for editing logic without a view, such as in a test or a worker.

What is in it

• Documents and edits: text, lines, selections, simultaneous edit batches, revisions, undo history and change events.
• Typing and commands: typeText, paste and the 45 commands of execute, with the options that turn each behavior off.
• Search and folding: find and replace with regular expressions and preserved case, and folding ranges with roles.
• API reference: the helpers next to the model, such as changedSpans, scanBrackets and the word predicates.

Coordinates

Every offset is a UTF-16 code unit, the same as a JavaScript string index. Lines and columns are zero-based. An emoji takes two code units and a tab takes one, however wide it is drawn.

Shape	Fields	Meaning
Selection	anchor, head	Offsets; anchor > head is a backward selection
TextEdit	from, to, text	Replaces the half-open range [from, to)
Position	line, column	Zero-based line and UTF-16 column
ContentEdit	start, end, text	Two positions, in the order a language server applies them
FoldingRange	startLine, endLine, ..	Zero-based inclusive lines plus offsets

The view and the LSP packages call the column character instead of column. The unit is the same.

Limits

• The scanners behind typing, Enter, comments and folding are lexical, not parsers. A / after an ambiguous construct can read as a regular expression or a division. JSX text, Python triple quotes and HTML tag pairs are not modeled. The markup around PHP tags and heredoc and nowdoc bodies are lexer modes of their own.
• A language is a string id, such as typescript or php. An id the lexer does not know has no comments and no brackets, so nothing in it pairs, toggles or indents.
• Auto-indent works from brackets, keywords and the previous line, not from a formatter. It leaves alone languages without braces, JSX elements, Vue templates, PHP markup and the bodies of strings and comments.
• Search, folding and selection expansion read the whole document, synchronously, on each call. A regular expression has no timeout, so run untrusted patterns in a worker.
• Word movement reads every character it crosses, so one very long word is slow to cross.
• History keeps the last 200 steps. The text has no size cap.
• Columns are UTF-16 code units, not graphemes. The model has no notion of bidirectional text.
• The package creates an Intl.Segmenter when it loads, to delete by grapheme, so the runtime needs one.

License and provenance

The package is FSL-1.1-MIT, except the word-boundary predicates in src/words.ts. Those are adapted from EditorActionUtil.java of the IntelliJ Platform (JetBrains/intellij-community at 7184b03, lines 934 through 982) under Apache 2.0, and the package ships NOTICE and LICENSE.apache-2.0.txt for them. Everything else is written for this package: the rope, history, commands, search, folding and the lexical scanners. Bracket and quote typing were checked against the documented behavior of IntelliJ's handlers; none of their code is used.