Vadim Gritsenko writes:
It's not concern of the resolver's user to fiddle with request / response. If change to be made, it should be done in cocoon source implementation.
This sounds right.
I think reader has a right to make this call unconditionally.
If the resolver shouldn't fiddle with the response, why do they have the right to set the content length of the response?
Because reader should care less who invoked it. Reader should not change its behavior depending on caller. Cocoon protocol implementation should care about request / response it provides to reader, not the reader.
PS Are you using reader to deliver xsl:import parts? Have you tried generate/serialize pair, which does work?
I'm not making the reader, just trying to import a stylesheet using a "cocoon:/path" source. The reader is created in the callback resolver passed down to xalan. Here's a (very long) stacktrace:
Thread-25[1] where [1] org.apache.coyote.Response.checkSpecialHeader (Response.java:386) [2] org.apache.coyote.Response.setHeader (Response.java:361) [3] org.apache.coyote.tomcat4.CoyoteResponse.setHeader (CoyoteResponse.java:1162) [4] org.apache.coyote.tomcat4.CoyoteResponseFacade.setHeader (CoyoteResponseFacade.java:375) [5] org.apache.cocoon.environment.http.HttpResponse.setHeader (HttpResponse.java:143) [6] org.apache.cocoon.reading.ResourceReader.processStream (ResourceReader.java:254) [7] org.apache.cocoon.reading.ResourceReader.generate (ResourceReader.java:292) [8] org.apache.cocoon.components.pipeline.AbstractProcessingPipeline.processReader (AbstractProcessingPipeline.java:615) [9] org.apache.cocoon.components.pipeline.AbstractProcessingPipeline.process (AbstractProcessingPipeline.java:486) [10] org.apache.cocoon.components.source.impl.SitemapSource.getInputStream (SitemapSource.java:276) [11] org.apache.excalibur.xml.xslt.XSLTProcessorImpl.getInputSource (XSLTProcessorImpl.java:777) [12] org.apache.excalibur.xml.xslt.XSLTProcessorImpl.resolve (XSLTProcessorImpl.java:704)
So, you *are* using a reader to read imported stylesheet. I'll repeat: have you tried generator/serializer pair? It *does* work.
<snip stacktrace/>
So should the ResourceReader not touch the response?
No, it can touch it.
Or should cocoon or excalibur be building a new internal-use response that ResourceReader can freely change since it will be discarded?
Yes, that's what should be happening (but it does not).
Or is the problem somewhere else I'm not seeing?
Try the suggestion above.
Vadim
