tom_undo

Published on pub.dev as tom_undo 0.1.0.

One undo/redo stack, TomUndoStack<E>, that every undo history in the workspace runs on — Tom Forge's DefaultUndoManager, the text, Markdown, YAML and Dart editors, the file tabs' ForgeFileSession, and tom_flutter_ui's TomUndoHistory. Plain Dart with no dependency, so it reaches no dart:io and no Flutter.

dependencies:
  tom_undo:
    path: ../../tom_ai/basics/tom_undo

• Applied first, moved second. undo and redo apply an entry and only then move it between the stacks, so an application that throws leaves the entry where it was and the error reaches the caller. undoAsync and redoAsync do the same for an application that completes later.
• Bounded. A push clears the redo stack and drops the oldest entries beyond maxEntries; popNewest takes the newest entry back so a history can merge a burst of typing into one act.
• Atomic composite steps. tomApplyAll (and tomApplyAllAsync) apply a list of steps in order and, when one throws, put back the ones already applied, newest first, before rethrowing.

import 'package:tom_undo/tom_undo.dart';

void main() {
  final history = TomUndoStack<String>(maxEntries: 50);
  history.push('typed "hello"');
  history.undo((entry) => print('undo $entry')); // applied, then moved to redo
  history.redo((entry) => print('redo $entry'));
}

Forge's undo contract (ForgeUndoManager) and its default implementation stay in tom_forge_core, built on this stack.

One stack, kept one

_bin/check_undo_stacks.py at the workspace root finds an undo or redo history written by hand anywhere else — a field or local named like _undo or _redoStack that starts as an empty list or queue — and its fix is this package, not a rename:

./_bin/check_undo_stacks.py --strict   # exit 1 on a hand-rolled stack
python3 _bin/test_check_undo_stacks.py # the check's own tests

It is started by hand, like every check in the workspace.

License

BSD-3-Clause — See LICENSE for details.