[michaelthyregod]

Not benchmarked yet: I'll run the benchmarks once my pending AngleSharp core change (part of an AngleSharp.Css fix) is settled, since it may affect the results. The check per start tag is one comparison.

Follows up on #39: the XML parser gets the same nesting limit as core's HTML parser.

Change

• XmlParserOptions.MaximumDepth mirrors core's HtmlParserOptions.MaximumDepth:
  • the deepest level the parser nests elements at, counting the root as level 1;
  • default 512 (XmlParserOptions.DefaultMaximumDepth);
  • zero or a negative value restores the default, and Int32.MaxValue lifts the limit.
• Beyond the limit:
  • In strict mode, the parser throws XmlParseError.MaximumDepthExceeded (0x314), because nesting the element elsewhere would change an XML document.
  • With IsSuppressingErrors, the element is attached at the limit as a sibling of the deepest element, as the HTML parser does. Text that follows a closed sibling stays in its own element.
• Breaking for documents nested deeper than 512 levels, which parsed before.
• Tests and benchmarks that go deeper: the deep-nesting tests (20,000 levels) and the deep benchmark shapes still need deep documents, so they lift the limit. Their documents and assertions are unchanged. The benchmark sets the option via reflection, so the same benchmark code still compiles against older builds.
• Docs: the API tutorial, AGENTS.md and the CHANGELOG are updated.

Tests

XmlMaximumDepth covers:

• the default and how the property is resolved;
• 512 levels parse and 513 throw, both synchronously and asynchronously;
• custom limits, including 1;
• Int32.MaxValue;
• self-closing elements beyond the limit;
• attaching at the limit in recovery mode, including keeping the root as the parent.

621 tests pass.

🤖 Generated with Claude Code

[FlorianRappl]

Would remove the (breaking for documents nested deeper than 512 levels). Those are most likely hostile documents.

In any other case somebody needs to change the default, which is imho fine as it uncovers that there are some potentially ill-formed documents being parsed.