Limits and Threading

Bound parsing work

Limits are inclusive and count UTF-16 units rather than bytes. Zero disables an individual bound. new ParseLimits() retains a finite XML entity-expansion ceiling even when other bounds are omitted.

Property	Scope	Default
MaxInputCharacters	Original markup or CSS input	0
MaxTokenCharacters	One atomic lexical token	0
MaxNestingDepth	Nested CSS functions and simple blocks; not DOM depth	0
MaxEntityExpansionCharacters	XML general/parameter entity replacement units	10,000,000

Use HtmlParseOptions, XmlParseOptions or CssParseOptions to supply bounds. Public parse entry points also accept a cancellation token. With using Jint.HtmlParser;:

using var cancellation = new CancellationTokenSource(TimeSpan.FromSeconds(5));
var diagnostics = new ParseDiagnosticCollector(capacity: 20);
var options = new HtmlParseOptions
{
    Limits = new ParseLimits { MaxInputCharacters = 100_000, MaxTokenCharacters = 20_000 },
    Diagnostics = diagnostics
};
var document = MarkupParser.ParseHtml("<p>Hello</p>", options, cancellation.Token);
Console.WriteLine(MarkupSerializer.ToHtmlChildren(document.DocumentElement!.LastChild!));

ParseLimitException exposes Kind, Limit and Observed. Handle it separately from MarkupParseException (XML/SVG syntax) or CssParseException (a CSS construct that cannot be parsed). Cancellation raises OperationCanceledException. A failed operation returns no result; the context of fragment parsing is not replaced by a successful or failed parse.

The entity ceiling is not a complete allocation or execution-time budget. Input and token ceilings and cooperative cancellation should be chosen for the application's workload. ParseLimits.Unbounded disables every bound, including entity expansion; reserve it for inputs whose work your host already controls. Serialization has a separate output limit.

Collect recoverable diagnostics

HTML and CSS options accept a ParseDiagnosticCollector with a positive capacity. Its Items view is read-only and live, and IsTruncated records overflow. Diagnostics have domain-prefixed Code values and original UTF-16 offsets. Reuse after Clear() only when earlier readers no longer depend on the collector's live view. Recovery diagnostics do not turn HTML parsing into strict XML parsing.

Share immutable inspection safely

After parsing completes and a tree is safely published, concurrent immutable inspection of links, names, character data and attribute values is supported, including cold text/attribute getters, text-content reads and serialization with separate per-call state. Keep the tree unchanged throughout all of these operations.

Mutation, adoption, parsing sessions, live ranges/iterators and mutation subscriptions require external synchronization with all readers. One writer alone does not permit overlapping reads. Operations that initialize mutable state or indexes, including selector/id queries, form-control state and processing-instruction attribute views, also need synchronization. A method that looks like a read is not automatically safe concurrently.

Trees owned by Jint.Browser retain page-loop affinity. Use scheduled Page operations to access those trees; standalone immutable inspection does not authorize reading a live browser DOM from another thread while the page can run script, parse or mutate it.