On Mon, 7 Jan 2002 11:07:39 -0500, Casey Duncan <c.duncan@nlada.org> wrote:

Quick question for you HTTP gurus out there.

I am writing a product that marshals Zope objects into an XML data stream. In implementing the publication of this data through Zope, I have two options:

- Stream the data to the client as it is returned by the marshaler in chunks. I do not know the Content-Length in advance, so I cannot set that HTTP header.

Using RESPONSE.write, ZPublisher will use chunked encoding if it can, or Connection:Close if not.

- Get the entire XML document as a string, set the Content-Length and let ZPublisher take care of it.

If your method simply returns the XML document as a string then you dont need to set Content-Length; ZPublisher will do it for you. However, it will all stay in memory until the last byte is transmitted. Also, the first byte will not be transmitted until the last byte has been calculated.

I have initially chosen the former as it seems to be the best use of memory resources for what could be an arbitrarily large data stream.

You will want to have a look at the HTTP implementation. Without a manually-set Content-Length header, RESPONSE.write doesnt know that the response is big and will keep every untransmitted chunk in memory. I used to have some patches in the Collector that used to make this a little better behaved, but I dont have time to rebuild those right now. I suggest you have a look at the implementation and hack it a little. Toby Dickenson tdickenson@geminidataloggers.com