Ok, base64 encoding looks like the way to go. This works
fine for small XML documents, however, I have notced that
when I try to send a large base64-encoded XML document
(about 1.5 MB) in this manner, it takes an extremely long
time (several minutes). I am using the loopback
interface for my testing, so it is not limited by network
bandwidth. Is there a more direct way to transfer large
ammounts of data between the client and the server?
Given that RPC::XML::Server uses HTTP::Daemon and
HTTP::Response for its transport, is it possible to
directly use HTTP methods such as 'GET' to transfer large
documents?
Thanks,
-brian
| [reply] |
Doesn't look like it as such, but peeking under the hood what you could do is to create an RPC::XML::Server instance with no_http set to true, then have your own HTTP::Daemon code which passes the body of any XML-RPC calls off to the dispatch() method. Anything which needs to return an XML file would return a URL which your HTTP::Daemon handle itself (rather than passing off to the RPC::XML::Server).
Hopefully that made some sort of sense. :) Read the docs for RPC::XML::Server, specifically the arguments for new() and the description of dispatch().
| [reply]
[d/l]
[select] |
Ok, the large ammount of time spent transferring base64
encoded data appears to be spent in the client, receiving
(but not decoding) the encoded data. It appears to be
related to the number of newlines in the data, because if
I strip them, the data is transferred in a couple seconds.
Is there a way to encode the data such that there are no
newlines?
| [reply] |
Ok, this is a little bit gross, but it is possible to
use the MIME::Base64::encode_base64() method to do the
encoding, and you can specify that you don't want it to
insert newlines. This works, but you have to be careful
to decode the data on the other end (i.e. it doesn't
happen automagically), because RPC::XML doesn't know that
its payload is base64 encoded. It does know if you use
the XML::RPC::base64->new() method, which wraps the
data in <base64> tags). Not totally ideal, but I
think I have my solution. :^)
Thanks to all for your suggestions!
-brian
| [reply] |