The answer is, I have to provide my own XmlResolver. I don't think this is built-in to .NET 3.5. That's baffling. It's also baffling that it has taken me this long to stumble onto this problem. It's also baffling that I couldn't find someone else who solved this problem already?
Ok, so.. the XmlResolver. I created a new class, derived from XmlResolver and over-rode three key things: Credentials (set), ResolveUri and GetEntity.
public sealed class XhtmlResolver : XmlResolver
{
public override System.Net.ICredentials Credentials
{
set { throw new NotSupportedException();}
}
public override object GetEntity(Uri absoluteUri, string role, Type t)
{
...
}
public override Uri ResolveUri(Uri baseUri, string relativeUri)
{
...
}
}
The documentation on this stuff is pretty skimpy, so I'll tell you what I learned. The operation of this class is like so: the XmlReader will call ResolveUri first, then, given a resolved Uri, will then call GetEntity. That method is expected to return an object of Type t (passed as a param). I have only seen it request a System.IO.Stream.
My idea is to embed local copies of the DTD and its dependencies for XHTML1.0 into the assembly, using the csc.exe /resource option, and then retrieve the stream for that resouce.
private System.IO.Stream GetStreamForNamedResource(string resourceName)
{
Assembly a = Assembly.GetExecutingAssembly();
return a.GetManifestResourceStream(resourceName);
}
Pretty simple. This gets called from GetEntity().
But I can improve on that. instead of embedding the DTDs in plaintext, I gzipped them first. Then modify the above method like so:
private System.IO.Stream GetStreamForNamedResource(string resourceName)
{
Assembly a = Assembly.GetExecutingAssembly();
return new System.IO.Compression.GZipStream(a.GetManifestResourceStream(resourceName), System.IO.Compression.CompressionMode.Decompress);
}
What I wanted to do is resolve only URIs for DTDs from Xhtml 1.0. So I wrote the ResolveUri and GetEntity to look for those specific DTDs, and respond affirmatively only for them.
For an XHTML document with the DTD statement, the flow is like this;
XmlReader calls ResolveUri with the public URI for the XHTML DTD, which is "-//W3C//DTD XHTML 1.0 Transitional//EN". If the XmlResolver can resolve, it should return... anything. If it cannot resolve, it should throw. My implementation just throws.
XmlReader then calls ResolveUri with the System Identifier for the DTD, which in this case is "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd". In this case, the XhtmlResolver returns a valid Uri.
XmlReader then calls GetEntity with that URI. XhtmlResolver grabs the embedded resource stream and returns it.
The same thing happens for the dependencies - xhtml_lat1.ent, and so on. In order for the resolver to work, all those things need to be embedded.
And yes, if the Resolver cannot resolve a URI, it is expected to throw an Exception. This isn't officially documented as far as I could see. It also isn't expected. (An egregious violation of the principle of least astonishment). If instead, ResolveUri returns null, the XmlReader will call GetEntity on the null URI, which .... ah, is hopeless.
This works for me. It should work for anyone who does XML processing on XHTML from .NET. If you want to use this in your own applications, grab the DLL. Licensed under the MS Public License.
You can plug it into your XML apps that fiddle with XHTML. Use it like this:
// for an XmlDocument...
System.Xml.XmlDocument doc = new System.Xml.XmlDocument();
doc.XmlResolver = new Ionic.Xml.XhtmlResolver();
doc.Load(xhtmlFile);
// for an XmlReader...
var xmlReaderSettings = new XmlReaderSettings
{
ProhibitDtd = false,
XmlResolver = new XhtmlResolver()
};
using (var stream = File.OpenRead(fileToRead))
{
XmlReader reader = XmlReader.Create(stream, xmlReaderSettings);
while (reader.Read())
{
...
}