This is my first experience with JIRA, so apologies if it is not formatted properly. I raised this topic on the tapestry-users mailing list and was asked by a couple of people there to create an issue here.
internal.services.XMLTokenStream's openStream method contains the following lines:
InputStreamReader rawReader = new InputStreamReader(rawStream);
PrintWriter writer = new PrintWriter(bos);
Both of these implicitly rely on the default JVM charset. This poses a significant problem for non-ASCII text in templates on Google App Engine, where the default is "US-ASCII". In the interests of robustness, I think it would be nice if Tapestry was able to eliminate any reliance on default charsets. I am not confident enough in my understanding of Tapestry internals to know how to appropriately retrieve symbol properties (such as "tapestry.charset") via the IoC system in internal service implementations, but I have verified that explicitly specifying "UTF-8" as follows resolved my problem:
InputStreamReader rawReader = new InputStreamReader(rawStream, "UTF-8");
PrintWriter writer = new PrintWriter( new OutputStreamWriter(bos, "UTF-8") );