[Zope3-dev] PEP 333 and Zope

[Phillip J. Eby]

At 05:38 PM 11/9/04 +0100, Eckart Hertzler wrote:
>On Tuesday 09 November 2004 17:04, Phillip J. Eby wrote:
> > The second issue is that your implementation won't work properly for
> > streaming output or large files, because you're buffering all the data in
> > memory, duplicating the buffering that already exists in a publisher
> > response.  Instead of a StringIO, you need to create a simple file-like
> > object that wraps the WSGI 'write()' functionality.  The only case in 
> which
> > PublisherApp.__iter__ should yield a value is if the response wasn't
> > actually sent to the output stream.
> >
> >
>This was due to my limited understanding of the publisher internals.
>It seemed to me, that the publisher produced the body in one block, so
>I did not see how streaming might happen...

It's in the 'write()' method of the response.  The simplest way to make 
this work, I think, would be to pass the WsgiHeaderOutput object as the 
outstream, and add a 'write()' method to it that calls the WSGI 'write' 
callable.  Finally, the application object should call 'publish()' in 
'__init__', rather than in '__iter__'.  You could in fact probably drop the 
iteration altogether at that point.

On a side note, I've just noticed that there appears to be a bug in the 
trunk version of zope.publisher.http.HTTPResponse class; it looks like the 
'output()' method will incorrectly set the Content-Length header to the 
size of the first block of streamed data, overriding any manually-set 
content length, and causing variable-length streams to have an incorrect, 
fixed-length header.  Somebody should probably add a test for that.