- From: T.V Raman <raman@google.com>
- Date: Thu, 10 Mar 2011 09:37:16 -0800
- To: nrm@arcanedomain.com
- Cc: raman@google.com, ashok.malhotra@oracle.com, www-tag@w3.org, jeni@jenitennison.com
Noah, I hate to say this -- but at this point, JavaScript is part of the Web platform, and trying to insist that URL references make sense outside a javascript evaluation context does not make sense on today's Web. Perhaps the bug was that we didn't (early enough) invent a URL mechanism that made it obvious as to what evaluation environment the URL required --- but that's water under the bridge. URL de-referencing has always relied on bits from the evaluation environment --- though admittedly, expecting a JS interpreter is a big leap from the past. Noah Mendelsohn writes: > > On 3/9/2011 4:47 PM, T.V Raman wrote: > > Itend to agree with Ashok here. The '?' is well understood as the > > server-side param. > > I think that, in the long run, having identifiers be either "client-side" > or "server side" is a mistake, though certainly that idiom is out there. To > the extent practical, we want the same identifiers to work everywhere. An > example of where this is breaking, is that Ajax-orginated identifiers with > #, such as #!, tend not to work when sent to non-Javascript clients. > Conversely, ? identifiers can be interpreted at both client and server; > Google maps URIs work OK from mobile devices that lack Javascript, and are > crawlable. > > > only lead us furthe rinto our original comfort zone of "state is > > managed on the server" > > Not where I'm going at all. I want a syntax that can be interpreted > transparently, server and client, with good Javascript control over when > you do or don't want server interaction. ? seems like the more promising > long term syntax; stated differently, I think it will be easier to start > building user agents that allow Javascript to set a ? URI at the client > without initiating server interaction, than it will be to get fragment ids > into the HTTP Request-URI when you need them interpreted at the server. > That being the case, if we want an identifier that works equally well at > both ends, with good control over when server interaction occurs, ? is the > more promising long term choice. > > Clearly that's not what implementor's are doing today, but I posit that's > mainly because the user agents they have do insist on contacting the server > when a ? URI is changed. Fix the user agents, and ? may start looking > better. Certainly none of the normative specs indicate that ? is > server-focused. > > Noah --
Received on Thursday, 10 March 2011 17:37:55 UTC