Sammy.js, CouchDB, and the new web architecture
quirkey.com
quirkey.com
So, here we have the return of the fat client, episode 20.
And as with all the other episodes the standard question remains: If you're going fat-client then why on earth stick to a platform as horrible as HTML/JS?
That being said, I share much of your amusement at/disdain for the attitude of "OMG, look at this brand new thing I have made that is completely different from all those old things!"
decent, free development tools
Well, you can have these for any mature platform. Be it QT, Flex or even Java. I'm not sure what's more decent about the HTML toolchain. The often cited FireBug is, to me, merely an emergency bandaid that makes development at least possible but still far from enjoyable.
the ability to interpose proxies and application middleware between the client and server transparently
There's nothing stopping any fat-client to speak HTTP. You don't need to a browser for that.
and a very, very large install base for the client runtime.
I'd say installation base is a red herring when talking about cross-platform apps but I'll give you that "instant-on" is an argument. For that I'd point to Air, which can run hybrid in and outside the browser.
I'm not arguing that HTML doesn't have it's place, mind you. But the extent to which it is abused nowadays is absurd.
You are underetimating this. For evrything else, (air, java, Qt, C#) I have to start a new app. My firefox is always on.
Java too has a very large install base. It is a shame that SUN didn't do much to promote it but you can still take advantage of the fact that many many people have the JVM installed.
Programs are temporarily installed for a session by visiting a specific URL (Chrome's app shortcut even makes it seem like a real installation). You update to new versions by refreshing the page.
Old school, new school, like it or not, this is the foreseeable future.
I'm sorry but I fail to see what 3d technology in the browser has to do with application UIs. The two fundamental problems with HTML are the complete failure that is the box model, and the W3Cs complete ignorance of real world requirements.
As an example for the latter just look at the pityful HTML widget-set that hasn't changed one bit over the last 15 years. Every OS GUI toolkit in existence has native widgets for comboboxes, date pickers, sliders and various other amazingly useful primitives. I don't see any plans for finally bringing any of that to HTML. Instead we'll continue to see every new javascript framework reinvent them to varying degrees of success, along with a long list of other baseline functionality (client-side form validation, anyone? WebSocket really soon now?).
All the while the W3C invents fancy audio/video tags, 3d widgets, a questionable canvas API and lots of other stuff that hardly solves any real problem we developers are fighting every day, in any app.
I don't think the evolution of RIAs will wait for the W3C to get their act together. I rather think the alternatives to the browser will see a raise in popularity in the foreseeable future, as the HTML/JS combo gets maxed out and more people start looking for ways to notch the expirience up further than possible within the browser constraints...
While it might be the new old thing, it's happening in the context of a different tech env, different market. The limitations of the old old thing might be different for the new old thing.
The advantages of a fat-client client/server app are low latency of the UI, maybe development speed given the right toolset, plus definitely flexibility of the UI for graphics and complicated interactions. Disadvantages though are difficulty of deployment, client platform restrictions, distributing updates/patches.
Whereas the traditional web thin client is easy deployment to your users across different platforms, via your single installation on the server. The disadvantage of traditional web clients is that you're restricted in UI complexity and latency is high. But this hasn't stopped so many enterprise and web apps being written (and rewritten) for the web, just to get deployment/update simplicity in place of a fat client.
So if you can build a complex UI fat client with low latency and the deployment ease of a web app, you're having your cake and eating it too. I think it really could be different, this time.
jQuery is not a good architecture for a full client application. jQuery is meant for flipping some dom attributes here and there. It does a poor job of modeling the UI state. Sinatra is also a poor architecture choice for the client since the client has persistent state.
We already have a well designed client architecture. It's called Cocoa and its implemented in the browser as Cappuccino.
> I find myself hesitant about putting
> my business logic essentially on the front-end
I think that's the core argument, if you omit the "CouchDB as an app server" part (which is orthogonal).
Business logic was on the server in stored procedures, which were exposed in a simple web service api so they could be called directly by the browser script logic. This also made it very testable because the stored procedures could be invoked by testing scripts that did not depend on the presence of any middle tier or UI at all. It also would have been a simple matter to build other client front-ends, as long a platform could work with XML and HTTP it could in theory be a client.
The XSLT to render the UI from XML could get a bit verbose but despite that it turned out to be a very productive way to work.
The downsides at the time were that IE was slow to render when you injected the HTML output of the transforms into the body.innerHTML property; this became noticeable if the generated page was large, but on the plus side this was a constraint against developing overly-complex pages.
That statement kind of troubles me. It shows that you're really just offloading all of the work from a normal app server to the data storage layer. From a "Mom look what I can do!" perspective this is cool, but, again, you're not eliminating your app server. Couchdb allows this because it uses HTTP as it's communication protocol, but one could do the same with MySQL, via an HTTP communication frontend, which sounds a lot like what Rails, Django, etc effectively are with some optimizations to make other non-data-storage operations faster.
This will not fool anyone. You can't fold the business logic into the database and then pretend that you've eliminated the business-logic layer on the server. It's still there. You've just driven it into hiding.
However, I could see some of the code, such as the code that actually builds the page the users sees, being pushed into the browser. But then, don't we already have that with XML and XSLT (or whatever?)
There's always the option of using HTTP authentication.
Doing per-user security directly in the DB with simple CRUD permissions per table would be enough of a headache, but most applications eventually require finer-grained security than that, and for performance reasons you also don't want a client to be able to execute just any arbitrary query.
It seems like this is pretty close to same trap a lot of people fall into of only enforcing data validation client-side, or only enforcing things like view permissions client-side (by not rendering links), which leaves all sorts of holes open.
It's cool and all, but it sounds like you're still going to basically need a server stack there to enforce permissions and some logic and consistency things. For example, say users have some private data. What stops me from modifying the javascript to insert a record that points to someone else's email address in my record? Unless the database can see this and prevent it I might then be looking at or even able to change someone else's email on their account. It can get very hairy if you don't have full control over the database interaction because you have to start guarding against arbitrary database operations.
http://wiki.apache.org/couchdb/Nginx_As_a_Reverse_Proxy
Then as long as the client can keep its identification token secret from others, you can implement "user can only query their own blog entry"
I suggest you look into couchDB a bit further. CouchDB is a different beast altogether from relational databases. Therefore, the assumptions you make about what a database can and cannot do coming from a relational world doesn't always apply.
Example of how he does authentication: http://github.com/jchris/toast/blob/master/validate_doc_upda...
http auth + document update validation function
It's unintuitive, but this setup actually supports a huge number of connections and runs really fast. Go figure.
Now, that serverside code can be javascript serverd on something akin to 10gen if javascript is their tool of choice, but exposing any of those to the user so that they may edit them without any secondary sanity check is unacceptable.
Of course, it also means that the newly-empowered database has to scale out just as easily as an application server. For starters, request authentication becomes the responsibility of the database, and that could mean heavy volume.
The client-side script could read who to follow from the server and by itself performed the task of assembling the timeline which was the CPU and IO extensive task that twitter had problems scaling with. Not sure the extra bandwidth cost of performing this client side would be worth it, but there's almost no logic at all server side so there's not much scaling of processor power at least. Programming with prototype was fun, but then I found jQuery...