Probably digressing here (from our previous digression), but I find a hybrid API best. E.g., the python pulldom API, which is event driven until you find a node you're interested in, whereupon it can give you an object. Should be even easier with json, since it can just give back a data structure. Sent from my iPhone On 04/08/2012, at 11:26 AM, Martin Thomson <martin.thomson@xxxxxxxxx> wrote: > On 4 August 2012 08:52, Mark Nottingham <mnot@xxxxxxxx> wrote: >>> The other interesting case is where large amounts of data arrive in a stream. >>> SAX and SAX-like libraries makes this easy to implement with XML. I hope >>> there's an equivalent for Json; if not there needs to be. >> >> Funny you mention that, I was just looking into that yesterday. >> >> This seems to be in the front running: >> http://lloyd.github.com/yajl/ > > The conclusion that I reached on this subject was that both XML and > JSON are poorly suited to streaming use cases for large datasets. Try > implementing XMPP "framing" some time and tell me afterwards how much > you love SAX. > > I prefer CSV or lots of small objects.