[mwasiluk]

Base: modernize (5fd01a4, after #35 and #37). Third of the stacked sd-model PRs: A (#35,
merged) → B (#37, merged) → D (this, fixes) → C (lodash in the lockfile). One commit on top
of modernize.

Likely root cause of SilverDecisions/SilverDecisions#234 (see below).

Bugs

1. replaceNode (used by convertNode) did not set $parent of the children to the new node;
   they kept pointing to the removed node.
2. cloneSubtree / cloneNode copied nodes and edges shallowly: the copy shared $fieldStatus
   (and the nested computed values of the nodes) with the original.

Reproduced in the SilverDecisions app (2f77351, demo page, through the TreeDesigner methods the
UI calls) with sd-model 0.2.11 from npm:

• A: R → chance C → [decision D → T (payoff p*10), T2], global code p = 1; convert C to a
  decision node, change the code to p = 2, recompute: the payoff below the converted node stays
  10 (scope of D initialised from the removed node), expected 20.
• B: after the conversion, select the converted subtree (4 nodes), copy, paste: 7 nodes are
  pasted (the children were treated as separate subtrees).
• C: copy and paste a chance subtree, mark the probability of a pasted edge as invalid (what the
  sidebar does on input), recompute: the original edge is invalid too, its probability is not
  evaluated and the original tree reports invalidProbability, probabilityDoNotSumUpTo1.

Related issue: SilverDecisions/SilverDecisions#234

The issue reports that nodes in a copied part of the tree keep an old value of a global variable
(Day = 3200 after the global code was changed to Day=-3200) and do not see a variable added
later (Expert). In the attached file (saved by SilverDecisions 1.2.1) every node with the stale
scope is a descendant of one chance node ("Sam") whose own scope is current — the pattern bug 1
leaves when that node's type was converted after its children existed: the children keep the
removed node as $parent, and sd-computations initialises their scope from its frozen scope.

• With sd-model 0.2.11 in SilverDecisions 2f77351: global Day=3200, R → chance C → decision →
  terminal (payoff Day); convert C, then change the code to Day=-3200 / Expert=-20000 and
  recompute: the payoff stays 3200 and the scopes below C have Day: 3200 and no Expert — the
  symptom of the issue. With this PR: -3200, Expert present.
• Copy / paste alone (copy C, paste twice) does not reproduce it, with or without this PR, and
  the issue's steps mention only copying; so the conversion is the likely, not a proven, cause.
• Reopening the saved file and recomputing with code evaluation gives correct scopes in both
  versions (the links are rebuilt on load), so files are not damaged. The issue can be closed
  once SilverDecisions is released with this sd-model.

Fix

replaceNode sets e.childNode.$parent; cloneNode deep-copies computed and $fieldStatus,
cloneSubtree deep-copies the edges' $fieldStatus. Values stay the same, only the sharing ends.
Regression tests in test/contract/data-model-edit.test.js (fail on 0.2.11 and on this branch
without the fix, pass with it). CHANGELOG entry.

Results

• npm run test:build (Node 24 and 22): contract tests standalone(.min) 58, module(.min) 59,
  source 57, consumer bundle 57, Jasmine 8/8; also with the unreleased sd-utils of
  SilverDecisions/sd-utils modernize 2a3af7c (--sd-utils=<tgz>).
• SilverDecisions scenarios with this branch: A 20, B 4 nodes, C original edge valid, no errors;
  undo/redo and save/load of the edited diagram unchanged. Same with the sd-utils 2a3af7c
  tarball. App.serialize of 13 diagram files identical to 0.2.11 (the fix does not change
  loading/saving).
• Consumers, tarball installed with --no-save: sd-computations 196/196, sd-tree-designer 12/12,
  SilverDecisions build + Karma 37/37 (also with the sd-utils 2a3af7c tarball).

Not changed (open question): convertNode copies only the name of the node, not its code,
folded state or computed values.

Squash merge

[fix] keep parent links after convertNode, do not share state with subtree copies

convertNode left the children's $parent pointing to the replaced node (stale expression scope
in sd-computations, duplicated nodes on copy/paste in sd-tree-designer); cloneSubtree shared
$fieldStatus and computed values with the original (invalid copies marked the original invalid).