Related
I am building a massive real-time collaboration web application. It is a Web IDE that has support for HTML, CSS & JS programming and a stage area that would reflect the results a la JSFiddle, Plunker etc.
Now, the twist is that, it would support multi-user real-time collaboration, where people viewing the same instance of the web application would be able to write code together that would reflect across all the open instances. I have figured out the race conditions, session management et al
What I am having trouble with is
how to reflect the typing and/or deleting along with caret positioning
across these multiple instances giving the illusion that when one
person is typing, he is actually typing on all the instances.
The other thing is that RactiveJS says that it
updates only those parts of the page that are out of date. Tedious DOM
manipulation is a thing of the past.
Which is a very nice effect in their Demo. Imagine, you are on JSFiddle and you don't have to re-run every time. So, my question is, how do they do it? What is the concept behind it?
I don't want to use any library for this. I am pretty good in JS. I am having a hard time figuring out the algorithm.
Things I have considered:
Dirty Checking [but, its dirty right?]
Delta Differencing [but like ReactJS, it would have to update the entire application state every time]
Object.observe [the browser compatibility is not realistic yet]
So, if you have anything that can help me go in the right direction, I would be really thankful.
Realtime-collaboration tools, that allow concurrent editing/manipulation of objects/texts etc, usually use a variant of the Operational Transformation algorithm.
OT is not trivial to understand, much rather implement so I'd suggest you take a look at already-cooked libs for this such as:
ShareJS
Racer for Node
OT works, in some very basic way, similar to GIT
As an update to what you posted in the comments.
You say you are using Python. You wouldn't rebuild your whole codebase, I guess, but keep in mind that real-time collaboration tools usually benefit a lot from using an event-driven server-side language.
Since you are using Python, you could check out the Twisted Framework
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 9 years ago.
Improve this question
I want to create a custom UI framework in JavaScript for web applications (like Google Docs ui) (do not confuse with web application that deploy using languages like PHP, Python, etc.). However, after reading several books about web development, I understand that the best website is layered as follows:
Structure in HTML
Presentation in CSS
Behaviour in JavaScript
So there are several approaches to creating my own HTML document and control it in JavaScript. However in this approach HTML and CSS will be mixed, like in case of extJS UI. I am confused now, and I need some answers from experienced developers on how to write this kind of framework.
If HTML, CSS, and JavaScript is mixed.
What was advantages?
What was disadvantages?
Is there are other methods?
What was the usual type of creating UI frameworks?
I apologize that this answer is extremely long and at times may seem somewhat off-topic, but please keep in mind that the question was not very specific. If it is improved, or made less general, then I will gladly remove the superfluous parts, but until then, just bear with me. This is somewhat of a compilation of the other answers here, in addition to my own thoughts and research. Hopefully my ramblings will be at least somewhat helpful for answering this question.
General Tips for Frameworks
Frameworks are a ton of work, so don't spend all of that time for nothing. Work Smarter, Not Harder. In general, you should remember these things when creating a framework:
Don't Reinvent the wheel: There are tons of great frameworks out there, and if you create a framework that does the exact same thing as another framework, you've wasted a ton of your time. Of course, understanding what goes on inside another library is a great way to make your own library better. Quoting #ShadowScripter, "knowledge -> innovation." However, don't try to rewrite jQuery by yourself. :)
Solve a Problem: All good frameworks (and programs) start with a problem they are trying to solve, then design an elegant solution to solve this problem. So don't just start writing code to "create a custom UI framework," but rather come up with a specific problem you want to solve, or something you want to make easier. jQuery makes selecting and manipulating the DOM easier. Modernizr helps developers identify the features supported by a browser. Knowing the purpose of your framework will make it more worthwhile, and may even give it a chance of becoming popular.
Fork, don't rewrite: If the problem you aim to solve is already partially solved by another framework, then fork that framework and modify it to fully fit your needs. There's no shame in building of the work of others.
Optimize and Test: This is kind of a no-brainer, but before publishing version 1.0 on your website, test every single part of the function, under every single possible scenario, to make sure it won't crash and burn in production. Also, another no-brainer, minify your code (after testing) for performance benefits.
DRY and KISS: Don't repeat yourself, and keep it simple, stupid. Pretty self-explanatory.
Stick to Stable: (This is mostly my personal opinion) Unless you're trying to create a framework specifically targetted to HTML5 in Chrome 31, using experimental features and unstable code will make your framework slower, uncompatible with older browsers, and harder to develop with.
Open Source: (Another of my opinions) It takes years for huge companies like Google with thousands of dollars invested to create frameworks (e.g. AngularJS) so it is an excellent idea to make your source openly available. Having a community of developers contributing to your project will make development faster, and will make the end product faster, bug-free, and better all around. Of course, if you're going to sell it, that's another story...
For more information about best practices when making libraries, take a look at these links:
Javascript Library Design
Javascript Module Pattern: In Depth
Best Practices in Javascript Library Design
Building a Javascript Library
Types of Frameworks
The first thing you need to think about, as mentioned above, is what functionality you want your framework to provide. Here are is the list of types of frameworks/libraries (thanks to #prong for the link). For a much more comprehensive list, see jster, which has 1478 libraries, put into 8 categories, each with several subcategories.
DOM (manipulation) related
GUI-related (Widget libraries)
Graphical/Visualization (Canvas or SVG related)
Web-application related (MVC, MVVM, or otherwise)
Pure Javascript/AJAX
Template Systems
Unit Testing
Other
As you can see from the link, there are already dozens of libraries and frameworks in each of these categories, so there's not exactly much room for something new. That being said, I don't want to discourage you- who knows, you could create the next bootstrap or jQuery. So, here are some resources about each type of framework.
Note: you can't say that any type is better than the others, they simply are designed for different goals. It's like comparing apples and oranges.
DOM (manipulation) related
These types of libraries are designed to interact with, modify, and control the DOM of a website. Here are just a few of the things they do:
Select Elements in the DOM ($("div#id .class"))
Add/Remove Elements in the DOM ($("div#id .class").remove())
Edit Attributes of Elements in the DOM ($(div#id .class).height("30px"))
Edit CSS of Elements in the DOM ($(div#id .class).css("property","value"))
Add listeners for various events taking place in the DOM ($(div#id .class).click(callback))
The most notable of these, of course, is jQuery, and it has one of the largest user bases of any Javascript library. However, it is by no means perfect, and if your library wants to compete, the best thing to do would be to make it excel in the areas that jQuery fails- speed, fragmentation, and "spaghetti" code. (The last two aren't completely in your control, but there are certainly things that you can do to make it easier for users to use the most update version, and keep their code maintainable)
GUI-related (Widget libraries)
I think that this may be the type of framework you're looking to create. These types of libraries provide widgets (datepickers, accordians, sliders, tabs, etc.), interactions (drag, drop, sort, etc.) and effects (show, hide, animations, etc.). For these, people are looking for quantity- the best frameworks out there have several useful widgets/effects that work well. This is one case where it's "the more, the merrier," of course, if it works properly.
Graphical/Visualization (Canvas or SVG related)
The purpose of these libraries is to control animations on the page, specifically on an HTML5 Canvas. These feature animations and sprites for games, interactive charts, and other animations. Again, successful graphical libraries have many, many sprites/animations. For example kineticjs has over 20 different sprites available. However, make sure that quantity does not compromise performance and overall quality.
Web-application related (MVC, MVVM, or otherwise)
Basically, the idea is to provide a layout for the users to put their code in, typically separating the model (data) from the view(what the user sees), with a small controller to provide an interface between these two. This is known as MVC. While it is by no means the only software pattern to base a framework off of, it has become quite popular recently, as it makes development much easier (that's why Rails is so popular).
Pure Javascript- AJAX
This should really be two categories. The first, AJAX libraries, are often paired with a server side library and/or database (though not always) and are designed to make connections with a server and get data asynchronously. The second, "Pure Javascript" are designed to make Javascript easier to program in, as a language, provide helpful functions and programming constructs.
Template Systems
This might also be the type of framework you're looking to create. The idea is to provide components that developers can use. There's a thin line between Template Frameworks and Widget Frameworks (which twitter bootstrap, one of the most popular template frameworks, crosses a lot). While widget frameworks just give a bunch of little elements that can be put in a site, template frameworks give structure to a website (e.g. responsive columns), in addition to making it look good.
Unit Testing
This type of framework is designed to let developers test, or systematically ensure the correctness, of their code. Pretty boring, but also really useful.
Other
This type of framework is for really specific purposes that don't really fit into any of these other categories. For example, MathQuill is designed for rendering math interactively in web pages. It doesn't really fit into any other category. However, these types of frameworks aren't bad or useless, they're just unique. A perfect example is Modernizr, a library for detecting a browser's support for features. While it doesn't really have any direct competitors, can't be put into any of the other categories, and does a very small task, it works very well, and is very popular as a result.
More Types
There are a bunch of other types of libraries. Below are the categories (I'm not listing subcategories because that would take half an hour to copy down) that JSter puts their 1478 libraries into:
Essentials
UI
Multimedia
Graphics
Data
Development
Utilities
Applications
It depends on what you really want. The first distinction that needs to be made is between a Javascript UI framework (which provides structure to the app), an HTML UI Framework (Presentation) and Widget Libs.
Javascript Frameworks such as backbone, angular, ember, and knockout provide MVC-like structure to the app.
UI frameworks such as YUI, bootstrap, and Foundation provide a consistent HTML and CSS base.
Widget Libraries such as jQuery UI, Fuel UX, and Kendo UI provide ready made widgets.
There are also fully-fledged frameworks which provide things across the board, such as Google Closure tools, Dojo with Dijit.
This Wikipedia list pretty much sums it up, and here is the comparison.
In order to find the best way to create a framework, first ask this question: Can any of the above frameworks/libraries solve all or some of the problems/requirements I have?
If something solves all the problems, you can use it right away.
If something solves your problem partially, you can start by extending/forking that project/framework.
Follow DRY and KISS rules. Only solve a problem which nobody has solved as of now.
Fortunately, there is already a good solution: Google Closure Library. This is what Google uses. It shows the approach of mixing HTML, CSS and JS. I wouldn't say it's perfect, but I believe it's one of the best ones at this moment. Its architectural principles rely on proven component based and OOP concepts, and it's accompanied with a static compiler for Javascript. It's definitely worth of studying before baking your own solution.
I'd like to say that cloudcoder2000's answer sums it up nicely. I'm just writing this answer because it didn't seem right in the comment section :P
If you are thinking of making another framework, my suggesting is to stop thinking.
First find the thing in current implementations which troubles you the most, and try to find how you can improve it. Contribute to existing projects, nearly all of them are open source anyways. Also, you don't really need to be a JS-ninja to get into their midst. Take a fork, and get started. Once you're done, and feel that you're code is good enough, make it known to the original repo's maintainers that you have done improvements, and are looking for it to be merged into the project.
Keep in mind here that I'm not discouraging you from solving the problem at all.
I'm just pointing out that there are so MANY frameworks out there, wouldn't it be better if you went ahead and contributed to one of them instead of going for complete glory and implementing a full framework yourself? Making a framework is hard, but getting people interested in your framework is HARD. Really Really HARD, even for Google! Only once Angular got a very good documentation (which itself took quite some time, and resources, of Angular evangelists), that it has gathered so much steam. So, I'm just saying that instead of creating your own monument, perhaps lending a hand to others would be a more worthwhile exercise.
Most importantly though, is the fact that since you are just starting out, I presume you wouldn't have much experience designing frameworks, or thinking in those design terms even. It would of immense value if you contribute to existing projects. Then you will be gathering knowledge and experience of how things are built. Then, you'll be more confident. Then, you can go ahead and make your own framework! Then you'll be more equipped to tackle mammoth projects such as designing a framework. And Then, my friend, will you make something which would be worth the time of countless developers.
Short answer
Build a skinny DOM and only focus on JS code to make it more efficient.
Long answer
A good architect always replies with "it depends." You can't have one single framework that enjoys all others' benefits and suffers from no disadvantages, all at once. There's always a trade-off.
In order to create a framework that is really lightweight, you would probably want the lightest DOM (HTML) structure. However, having a skinny DOM might have the cost of more JS code. So you would probably try to write the most efficient code. You can start here.
Needless to say, you should be keeping the open-close principle, and have the stylesheets separated from HTML, using only classes and never inline styling. I would suggest using less. It makes the implementation faster, and the result is pure css so you suffer from no performance issues around it.
I must respectfully disagree with cloudcoder2000,
From a young age I have been being told don't re-invent the wheel, but why?
During the last 3.5 years, I have re-invented almost all of my web controls using javascript/html/css. From the extremely complex; for example a grid, rich text editor, tree view, dialog. To the very simple, like a progress bar, dropdown. What I gained is a better understanding of html/js/css.
No pain, no gain. I'm so happy with what I was doing these years as I learned more than others.
Some of the controls I re-invented, I think, are much better than the popular ones, like telerik, jquery mobile, extJS. For example, my tree view is fully customizable, with it one can build very complex trees.
So, I encourage you re-invent the wheels, and you will definitely get more than you expected. But also, from the beginning, you need to learn the source code of the popular controls, and re-invent them. You will be very happy when you find yourself be able to make the controls better.
Then the tips on creating HTML controls:
1. use jquery;
2. write jquery plugins(jQuery.prototype...) for simple controls, while define classes for complex controls;
3. separate css from html and js files. Define the html template in js, don't use html files, which make using the controls hard.
Regards,
Leo
For best performance in your UI design, you need to use a lightweight JavaScript framework like angular or backbone, Twitter Bootstrap for the UI, AJAX for base script load and use gzip compression in your app to make it lightweight and help the UI render faster.
I've been looking at quite a few of the MVC for JavaScript frameworks lately and I quite like the overall feel of Spine.js as well as the fact it is using CoffeeScript. That said, my primary concern at this stage is I need to be able to drop an MVC framework iteratively into a fairly large existing application. Most of the documentation that I have seen on Spine.js focuses on creating new applications.
Does anyone have any suggestions on the best way to go about doing this; if there is a good tutorial somewhere that would be great too, but Google was not my friend on that one.
Or is this just a bad idea and I should go elsewhere when dealing with legacy? While Angular didn't click with me in the same way; it did however seem pretty straight forward on how I could use it iteratively with a legacy application.
I think it's rather a bad idea to do it "iteratively". I'm working as freelancer for a company where they decided to do a complete UI rework splited over a few releases. In the first stage it was planned to drop the existing JS and use spine js as MVC and require js as AMD. So far so good - but the process was "iteratively". The project/App is quite complex and it took over a half year (I'm involved since 2,5 months).
It was a big mistake to leave all the existing JS and "iteratively" replace the functionality for the release. We ran into multiple problems, because of that. If you have hundreds of js files, inline scripts, dependencies it's much easier to simply throw them away, because you can tell literally if your new code works or not. Otherwise you always have to distinguish whose fault is it: new or old code.
This was also responsible for trouble that could have been avoided, or at least alleviated, if it were a hard cut:
The product owner was not really aware of the status, because he couldn't see any visible progress. He only figured out bugs and misbehaviour. That was unsatisfying/frustrating for both parties. One quote I've heard over and over again: "It did work the other day/before - now it's not"
You start to update some outdated libs, plugins that sometimes will cause compatibility issues, because API's tend to change, so you have to fix that in your old code as well. Means: at some point it starts to feel like hacking around problems and not to develop a cool/good App. Thats's extremely frustrating too.
It took waaays longer than expected/estimated.
You've been developing in dependency of your old code that does not exist in the very end.
My personal advice is: don't do it iteratively. Do hot fixes for the live App and develop the new one on a separate branch.
I can't see how conforming any legacy app that didn't have an explicit mvc structure into a spine.js or backbone or angular etc. is likely to be a drop in experience.
Iteratively converting is going to be a lot like writing a new app minus the boilerplate stuff for defining the app GUI interaction. Start by focusing on models and controllers by pulling out components that fit well into the MVC structure. I would suggest early on to avoid certain app interaction components of Spine like stacks, and routes as those make a lot more sense once everything else is ready to be used according to spine convention.
Aside from the language differences Javascript vs. Objective-J what benefits does Cappuccino provide over SproutCore and vice-versa in your experiences?
In terms of a long-term forecast, is SproutCore more "supported" than Cappuccino because it is backed by Apple?
I am trying to choose between the two. I am both familiar with JavaScript and Objective-C.
This is an interesting question, and one that has been popping up fairly frequently on various messages groups, twitter, and even IRC. There's a couple of ways to evaluate SproutCore versus Cappuccino, but, perhaps, some of the immediate caparisons that people look at are the following:
1) Their respective feature set
2) Ease of use
3) Community support and documentation
Let's look at the first point -- there respective feature set. By "feature set" there's a couple of ways to look at it. From the number of UI widgets they have; the foundational support to connect things together and communicate with some kind of back-end; the framework's general architectural approach, although not necessarily a "feature", but still important; and, yes, even the language you can use.
Regarding language, I think it's important that you do not dismiss what is being used (JS versus Obj-J). Why? Because of adoption and where you are coming from. SproutCore came from the perspective that JavaScript is indeed the language of the web, so it's what you use to program against the framework. Where JavaScript lacks in language OO completeness (proper object-object inheritance, etc) it makes up for in the framework (e.g. MyApp.Foo = SC.Object.extend({...})). Cappuccino comes in from a different angle. They use Obj-J as a primary language enhancement to JS in order to inject language features that JS is missing; this instead of injecting those language features directly into the framework (Cappuccino) itself. Of course, as the folks over at Cappuccino have noted before, you can still use JS to program against Cappuccino proper, but, then, you miss out on what Obj-J provides. Note to the Cappuccino community: Please correct me if I'm wrong :-). Finally, if you're someone who is already familiar with Obj-C then Obj-J may be more your cup of tea. Hey, even Sony is apparently now jumping on the whole Obj-C bandwagon to develop against their mobile platform :-P.
Looking at the architecture of the two frameworks, they both looked at Apple's Cocoa framework for guidance/inspiration in one form or another. Cappuccino took Cocoa fully to heart and basically ported Cocoas API. Again, if you're coming from developing apps in Apple using Cocoa then you're probably going to feel right at home. SproutCore on the other hand took inspiration from Cocoa where it felt right. As for pure architecture, they both follow MVC, they both make use of Cocoa-style bindings, they both have a data store mechanism, and they both have their own respective style of rendering and composing UI widgets/views.
The rendering of views is, to me, a particular area of importance. Both frameworks have some level abstraction in order to remove you from directly dealing with CSS and HTML even though at the end of the day they have to render to what the web browser ultimately understands.
On the Cappuccino side, they completely abstract away CSS and HTML from you. Instead, you use the framework's various rendering primitives to "draw" your views. Because of this level of abstraction, Cappuccino can make use of the best rendering approach available instead of coupling you, to some degree, with CSS and HTML.
As for SproutCore, you are rendering closer to the "metal" so to speak. When doing a pure rendering of a view, you make use of a rendering context object that provides a certain degree of abstraction, but, ultimately, you are directly injecting HTML and adding class names to apply CSS. Even after your view has been rendered and you want to manipulate certain parts of the view based on an event, you can directly access to the DOM elements and manipulate their properties. Depending on where you are coming from this may seem good or bad. Good for those who are used to working with CSS and HTML and like the more direct control over how the views are rendered and styled. Bad if you want to generically render a view in order to make use of the best render approach based on what the browser allows (HTML/CSS, SVG, HTML5 canvas, etc). But, note, there are future plans to make SproutCore have a more abstract rendering approach but still allow you to directly work with HTML and CSS if you so choose. So you'll eventually get the best of both worlds.
Now, as for the stock UI widgets/views the two frameworks come with -- they both have a lot right out of the box in order to get you going. Buttons, labels, lists, segmented views, radio buttons, scrollers, etc -- they're all there. Therefore, it's safe to say you're fine in both camps.
Going all the way back, let's now discuss the ease of use. To me, ease of use is based on you own personal experience working with JavaScript, HTML, Obj-C, Cocoa, other MVC frameworks, documentation, and community support. If you've never worked with Cocoa, or never built a decktop- or iPad-like app, then it's fair to say you're going to have a bit of a learning curve no matter what framework you choose. That being said, what you don't know and want to learn can be acquired through each framework's respective community and docs. Both have active communities in one for or another, so you won't be left out in the cold if you get stuck somewhere. As for docs, Cappuccino, admittedly, has the upper hand. The docs for SproutCore are lacking, but the code base is at least fully commented. The SproutCore community is fully aware of the docs needing to be updated, and it is currently something that is being dealt with, so keep checking.
Finally, you mentioned the long-term forecast for the two frameworks. It's public knowledge that Motorola bought the Cappuccino framework, so you certainly have a big company backing its growth and longevity, or at least it seems like that way for now. As for Apple and SproutCore, I personally can't speak for them, but Apple does not own the framework. There are many companies and various individuals that all use and contribute back to the framework in some way. That might give some people and companies pause or discomfort for those who are looking at SproutCore due to the more organic nature of the framework's development, but I don't see that as a problem. My feeling is that both frameworks will be around for a long time, especially now that more are looking at developing next generation desktop and iPad apps using open source frameworks. And, hey, competition between the frameworks is good -- keeps everyone on their respective toes :-).
Hope this information helps you out with your decision!
Cheers,
Mike
I'd like to touch on the comments made about objective-j Michael.
You're not going to lose anything if you drop down to JavaScript instead of objective-j. In all actuality the distinction is kind of difficult to make, especially in cases where we have toll-free bridged classes (more on that in a bit).
Objective-j is really just a thin wrapper over js. It provides classical inheritance something that has traditionally been implemented as a language feature, which sproutcore implements as a framework feature, it also provides code importing, accessor generation, static scoping, and support for messaging nil.
Objective-j instance variables are accessible via the traditional dot syntax if you want... I like to think of it like this: once you start writing a method, you're mostly writing JavaScript. That is, loops, variables, functions, closures, etc are all just javascript. You're not losing anything by dropping down, that's exactly how the language is designed.
We take it a step further by "toll-free bridging" some of our classes CPDate, CPArray, CPException, CPString and perhaps more that I can't recall. Toll free bridging just means a CPArray IS a native js array, and a native js array is a CPArray, so you can use methods and functions of both world interchangeably.
So for example would could do:
var foo = [];
[foo addObject:"bar"];
foo.push("2nd push");
var value = foo[0];
var value2 = [foo objectAtIndex:0];
alert(value === value2); //true
As you can see I'm using objective-j syntax and js syntax together... You can imagine the power if this.
The final thing I want to put out ther, just to make sure there is no confusion: objective-j gets parsed in the browser. It doesn't need to be compiled before hand (although we provide compilation tools for when you're ready to deploy your app).
I think some people are needlessly put off by objective-j as if it's some monstrous beast that will take time to learn, and while objective-j adds a lot of great features to js, to actually learn them won't really take you the better part of a day if you're already familiar with object oriented programming, and obviously if you're coming from cocoa you'll be able to jump right in.
I wrote a blog article exactly about "cappuccino vs. sproutcore". It is not a technical comparison but compares other interesting data.
http://elii.info/2010/11/cappuccino-vs-sproutcore/
From the Cappuccino website:
"On the other end of the existing frameworks are technologies like SproutCore. While SproutCore set out with similar goals to Cappuccino, it takes a distincly different approach. It still relies on HTML, CSS, JavaScript, Prototype, and an entirely new and unique set of APIs. It also requires special development software and a cumbersome compilation step. We think this is the wrong approach.
With Cappuccino, you don't need to know HTML. You'll never write a line of CSS. You don't ever have interact with DOM. We only ask developers to learn one technology, Objective-J, and one set of APIs. Plus, these technologies are implementations of well known and well understood existing ones. Developers can leverage decades of collective experience to really accelerate the pace of building rich web applications."
So it seems that Cappuccino does not have/need any build tools, and completely abstracts the browser away from the developer. Whereas in Sproutcore you get build tools (a development server, for example) and the developer should be somewhat aware of what DOM is.
Michael Cohens answer pretty much covered everything since it was extremely detailed.
I have been struggling with a decision for the past 3 weeks. I have read everything there is out on the web about both frameworks and I have written a lot of source samples with both and still cannot make a decision. The following issues have me jumping from one framework to the other and keep making my decision tougher.
Sproutcore has a better data store api than the one cappuccino has.
Sproutcore makes use of bindings better than cappuccino currently does. Cappuccino does also have kvc/kvo support but bindings are not totally there yet. For example in sproutcore you can implement incremental loading with bindings and ArrayController very easily where on the other hand in cappuccino its not as straightforward. Of course cappuccino offers the CPTableView DataStore api which is pretty clean and can achieve similar results just not with bindings. Its what cocoa did before core data. Bindings are constantly being worked on in cappuccino though.
Cappuccino has a better view api according to my personal taste. Although I am used to developing html and the DOM I much prefer the idea of abstracting the DOM completely away and getting rid of css.
One issue that is really important to me is the lack of a good TableView in sproutcore. Currently SC.TableView is in alpha and it is not performant at all. I dont know of a timeline for the tableview in sproutcore. I tried asking on the irc sproutcore channel but got no satisfying answer. Cappuccino on the other hand has a great and very optimized table view.
I have found more real world applications written on cappuccino than on sproutcore. There is also a pretty nice full blown application that is provided by cappuccino as a source sample and is very helpful. Check out http://githubissues.heroku.com/.
Despite the fact that I have no experience in objective-c and I much prefer the pure js syntax I will probably go with cappuccino on my current project and hope sproutcore comes out with a better table view in the future.
Disclaimer & Long Winding Question Approaching
I know topics like this have been beaten to death here so suffice to say I'm not asking about which framework is better, I don't really care about opinions on the better framework. They all do pretty amazing things.
The Question
Given that I have an existing web application, made of mostly regular HTML+CSS (jQuery where needed), which is the optimal framework to integrate a few "rich" pages into typically a regular stream of HTML.
Reason
I am trying to bring our proven application into the realm of awesome desktop like UI but I want to do it one small piece, one screen at time. But for our users, support personel and especially me taking it slow is the only option.
Also, with our branding requirements having a framework that just takes over the viewport isn't an option, it has to play nice with other HTML on the screen.
Imagine the example being a rich user manager in an otherwise plain HTML+CSS environment.
Experience Thus Far
Dojo + Dijit
Pros: The new 1.5 widgets plus the claro theme is the cure for what ails us. Dojo seems to be able to use markup to create the UI which is very appealing and has a fair amount of widgets.
Cons: Holy bloated lib Batman! Dojo seems to be enormous and I have to learn a custom build system to get it to stop requesting 4,800 javascript files. This complex empire of Javascript makes me believe I won't be able to create much that isn't already there.
ExtJS
Pros: Amazing set of widgets, does everything we could possibly want. Seems quick, every version brings new improvements.
Cons: I'm not sure how to use this without the entire display being EXT. I'm still building a web site, so I would prefer something that could integrate into what we already have. Some pointers here would be great.
YUI
Pros: Well, it's Yahoo isn't it? AWS console is downright wicked. Plenty of support and a giant community.
Cons: Well, it's Yahoo isn't it? AWS console is the only wicked thing. Complex for someone who's used to jQuery.
Help Me
I am willing to accept experience, links to ways to solve problems I've outlined, new toolkits (even though I'm pretty sure I've seen most by now) or even just advice.
Regarding ExtJS, it's pretty easy to start it in an existing div with something like this:
Ext.onReady(function() {
App = new Ext.Panel({...})
App.render('div-id')
});
The App panel can then have it's own layout manager.
This might be useful if you're familiar with jQuery, but not yet familiar with YUI 3 syntax: http://www.jsrosettastone.com/
Each of the libs you listed is excellent. When embarking on a larger scale project, the quality of a lib's documentation, community, and commitment to support become more relevant.
With Dojo, keep in mind that outside of dojo base, it only ever loads what you tell it to. But yes, without a built layer, that means it could easily end up requesting 50 JS files at startup for a large application using a bunch of widgets.
There are several pages in the reference guide documenting the build script: http://www.dojotoolkit.org/reference-guide/build/index.html
Rebecca Murphey wrote a nice blog post outlining an example app and build profile that you might find illuminative: http://blog.rebeccamurphey.com/scaffolding-a-buildable-dojo-application
If you get stuck, there's likely to be people in the Dojo IRC channel that can help.
RE ExtJS: I'm not sure what your exact situation is, but keep in mind that if you're intending to use it in commercial non-open-source software, you need to pay for licenses: http://www.sencha.com/store/js/
I'm a little curious as to why you think the size / number of requests is specifically an issue with Dojo though. I haven't used the others, but I'd expect it to be somewhat of a potential concern with any of them.