AddToTagVisitor: skip insert when a semantically-equal child already exists

[steve-aom-elliott]

Summary

Make AddToTagVisitor idempotent by default: if the target scope already contains a child for which SemanticallyEqual.areEqual(existingChild, tagToAdd) holds, return without mutating the tree. SemanticallyEqual ignores comments and attribute order, so this catches semantic duplicates that only differ in whitespace, attribute ordering, or interleaved comments.

An allowDuplicates opt-out flag is provided for the rare caller that wants the prior unconditional-append behavior. Two new constructor overloads accept the flag without forcing callers to pass null for tagComparator just to opt out:

• (Xml.Tag, Xml.Tag, boolean)
• (Xml.Tag, Xml.Tag, @Nullable Comparator<Content>, boolean)

Why default-on

I audited every in-tree call site of AddToTagVisitor (~22, across rewrite-maven, rewrite-csharp, and the static addToTag(...) helpers used by rewrite-xml's AddOrUpdateChild and rewrite-maven's AddPluginDependency). None want duplicate sibling tags:

• rewrite-maven: Maven POM schema forbids duplicate <dependency>, <plugin>, <property>, etc.
• rewrite-csharp (AddNuGetPackageReferenceVisitor): csproj schema forbids duplicate <PackageReference>.
• rewrite-xml/AddOrUpdateChild: already gates externally by checking for an existing child before delegating to addToTag, so the new check is redundant rather than behavior-changing for it.

Centralizing the check here removes existing per-site duplicate prevention and prevents the semantically-equal-but-different-identity output that has tripped up downstream tree-identity checks.

Why default-on with an opt-out

Raw XML technically allows duplicate sibling tags. Although no in-tree caller wants them, an unknown third-party consumer of rewrite-xml could in principle. allowDuplicates=true keeps that path available without forcing every in-tree consumer to flip a flag.

Behavior change

Default-off-then-on. Anyone passing a tagToAdd that is SemanticallyEqual to a child already present in scope will now see a no-op instead of a duplicate insert. A literal "I want a duplicate" use case isn't expected for in-tree callers (XML duplicates of <dependency>/<plugin>/<property> etc. are usually bugs); flagging here so reviewers can call out anything that needs adjustment.

Test plan

• New doesNotAddSemanticallyEqualDuplicate test exercises the basic skip path
• New doesNotAddDuplicateIgnoringAttributeOrder test exercises attribute-order tolerance
• New addsWhenChildrenShareNameButDifferentAttributes test confirms additions still happen when children only share a tag name
• Existing AddToTagTest cases still pass
• Targeted maven recipe tests using AddToTagVisitor (Add{Dependency,Property,Repository,ManagedDependency,Profile,Plugin}, ExcludeDependency) all pass