I understand that Web Workers cannot access the main document DOM, but has anyone successfully tried to build a partial DOM in a Web Worker using jQuery, output the resulting HTML and then attach that to the document DOM?
Did it provide much of a performance improvement over rendering on the UI thread that is worth the extra pain of implementing this in a thread-safe way?
Would this be trying to use Web Workers for something they shouldn't be used for?
It depends a lot on what kind of HTML are you trying to generate. However there is one big problem you might have overseen. It's not just that you can't access main thread's DOM.
You cannot create HTMLElement object instances in worker at all.
So even if you go through the trouble of generating the HTML, you will then have to convert it to HTML string or JSON, parse it in the main thread and convert it to DOM object.
The only case when this is worth it is when generating the data for HTML is CPU intensive. And in that case, you can just generate the data, send to main thread and display in HTML.
This is also good approach to keep data processing and visual rendering separate.
Related
is it possible to parse a HTML page that was fetched via a jQuery GET call in a web worker? I know that you can't access the DOM in a web worker, but I think it should be possible to pass the html to the worker and process it with a selector engine like sizzle. I don't need to manipulate the main page's DOM here. Or is sizzle also not available for use in this case?
The problem is that my processing/parsing is so heavy, that the UI hangs for about 10 seconds, so doing this stuff in a separate Thread would be a good idea. So if this is possible, how could I implement this?
I have a site written in Asp.net webforms. It uses ajax heavily.
Most forms on the site are submited with javascript. Javascript validates the input and sends it to /ajax.ashx on the server. The server processes the request and sends back a JSON response. My javascript uses the JSON to create html, which it inserts into the Dom.
I'm making a new version of my website written using asp.net MVC3. I've been looking at tutorials on this subject, and some of them recommend doing ajax in a different way. Rather than sending data and then building + inserting html with javascript, they create html at the server, and use javascript only to insert it into the Dom. For instance, in this tutorial.
Which way should I use? Using the new method will be quicker, but is it better?
That's a subjective question. Both approaches are possible and there is no better way. There are pros and cons of each approach.
Building the HTML on the server is easier and will require you less efforts but consumes more bandwidth compared to the first approach.
If you decide to go the first way you could use some client side templating framework which might help you simplify the generation of DOM elements on the client.
Creating html code directly into the server and injecting it with an ajax call is very fast and simple, the real problem is that in that way your service is bound to be used with that specific application. By sending RAW data you allow any app to use that data in any way, without bounding it to a specific application.
returning json feels more flexible to me; you can change what happens with your json response, like the layout it results in. If you return html you return data mixed with layout. This doesn't feel right to me.
I believe it is better to separate the layout from the actual data. That is why you should pass data between your scripts and not HTML.
If you go about it sending HTML, consider that you would have to build valid HTML and CSS, which might not sound hard at first but then you'll start using CSS that is not loaded in the file calling the ajax, etc.
Always separate content (data) from layout. That's why there is HTML and CSS, to separate layout from data. So why mess things up by mingling HTML between data?
Building the html serverside will probably be faster and not bog down the client which is important. Rendering data into HTML with javascript takes time and not every browser is fast with js (i.e. older versions of IE) so things can slow down if you're doing a lot of this.
Like previous posters said, it's kinda subjective because it depends on how much you're offloading to the client. I'm of the opinion that if you can do things serverside, you should.
If you are going to be using this service to return JSON to other applications/clients, then it's probably a good idea to just leave it as JSON and let the client do what it needs on their side.
I recent came across Javascript templates and have become quite intrigued.
I am building a large PHP application using the MVC pattern. Templating is handled by the rather awesome Twig.
I recently came across a javascript implementation of twig.
I have also read quite a bit about using javascipt template engines.
Now, in my application, the application generates a the full page for standard requests as fallback for users without javascript. For AJAX requests, it can generate the content part of the page (no <head>, <body> etc).
The ajax response object is currently just a the rendered HTML content, which is then inserted into DOM.
Should I instead return a response object containing a compiled javascript template and the objects to be inserted into the template? What are the benefits of doing that?
From the posts I have read, the javascript templates were only small snippets representing a small part of a page, for example displaying a comment on a blog post during the instant that the user has submitted it.
Are javascript templates only useful for inserting these sort of small "pieces" in a page?
Yes
A recent project I was on got "client-side template fever" and we used the dang things for every single template.
With every template library I've used (which is two or three), the error messages you get are not very good. If you have a huge template that operates on a fair amount of data, you'll quickly find the u.foo is null or not an object error message increasingly frustrating.
The best-practices I've settled on is:
Return a full HTML snippet (from the server) if it's a template that is seldom loaded. If you are only loading that HTML once on a page, then you might as well send it down populated right? This also encourages you to keep your logic up on the server, where it probably belongs.
Use client-side templates for small, repeated templates. Your blog comment example is probably a good one. I've found the most success when my client-side templates are pretty small (< 10 lines)
Use a logic-less template engine. The more logic allowable in a template, the harder they are to read/maintain. Plus, some of that logic should probably be in your business layer, not down in some JavaScript template. In other words, they force you to separate your presentation from your logic (which is good).
PS: It is cool that both your client and server-side templates could use the same templating engine. This will make developers on your project much more productive.
Depending on the scale and requirements of your application, you should take into considerations the following:
don't go rampant on Ajax; Ajax is not WebSockets, so use it sparingly. Plus, client-side execution speed is always key; AJAX is slow when compared to dumping as much resources as possible + using them when you need them; for example, you can send to javascript userdata = {name:'xxxx',address;'yyyy', ...} and using that, instead of requesting name and address via AJAX only when you need them.
it is recommended to use a global PHP var $sendData (or something like that) and right after plasting HTML, you send the $sendData with an easy <script>data = <?php echo json_encode($sendData); ?></script>
javascript templates add up to execution speed. which makes it reasonable to do the alternative, that is, separating everything that is dynamic and caching static resources like javascript functions
you can't, and I quote return a response object containing a compiled javascript template, not without residing to some server-side javascript engine that does the compile job prior to returning it
for your personal welfare, it will always boil down to how fast, easy and painlessly you respond to maintaining the application; there's no point in using super beta experimental frameworks, ports and kits over which you have limited control
Work smart, not hard.
Good luck mate.
You could also look into the Distal http://code.google.com/p/distal templating engine.
I was wondering what would be a good approach (or recommended approach) to render partial views in a web application.
I have a requirement where I need to load data into an already rendered page using AJAX, sort of like a "Load more..." link at the end of the page which grabs more information from the server and renders it to the bottom of the page.
The two options I am playing with at the moment for the AJAX response are
Return a JSON representation of the data, and use a client side template library, (e.g. jQuery templates) or just plain javascript to turn the JSON into HTML and append to the bottom of the page
Render the partial view on the server (in my case using grails' render template:'tmplt_name') and send it across the wire and just append the result to the bottom of the page
Are there other ways to do this? If not, given the above options which would be better in terms of maintenance, performance and testability? One thing I am certain about is that the JSON route will (in most cases) use up less bandwidth than sending html across the wire.
This is actually a really interesting question, because it exposes some interesting design decisions.
I prefer rendering partial templates because it gives my applications the ability to change with time. If I need to change from a <table> to a <div> with a chart, it's easy to encapsulate that in a template. With that in mind, I view almost every page as a aggregate of many small templates, which could change. Grails 2.0 default scaffolding has moved towards this type of approach, and it's a good idea.
The question as to whether they should be client-side templates or server-side is the crux of the issue.
Server side templates keep your markup cleaner on initial page load. Even if you use something like Mustache of ICanHazJS, you kinda sorta need to have an empty element in the page (something with your template), style it appropriately, and work with it in Javascript in order to get update your info.
Drawbacks
"chatty" applications
larger envelopes across the wire (responses include HTML, which may be considered static)
slower UI response time
Benefits
Good for people with lots of server side language experience
Maybe easier to manipulate or modify in a server-side environment (e.g. return a page embedded with multiple templates, which are programmatically loaded and included)
Keeps most of the application stuff "in one place"
Client-side templates can really reduce server-load, however. They make an application "less -chatty", in that potentially you could minimize the number of calls back to the server by sending larger collections of JSON back (in the same number of bytes, or fewer, that would be taken up by HTML in a server-side template scheme). They also make UI experience really fast for users, as clicking an "update" link doesn't have to do an AJAX round trip. Some have said:
Anthony Eden #aeden 10 Dec Reply Retweeted Favorite ยท Open
the future of web apps: requests are handled by functions, logic is always asynchronous, and HTML is never generated on the server.
Drawbacks
Not great for SEO (un-semantic extraneous UI elements on initial page load)
Requires some Javascript foo (to manipulate elements)
Benefits
- Responsive
- Smaller envelopes
Trends seem to be moving towards client side templates, especially with the power exposed by HTML5 additions (like <canvas>)...but if leveraging them requires you to rely on technologies you're not extremely familiar with, and you feel more comfortable with Grails partials, it may be worthwhile to start with those and investigate refactoring towards client side templates based on performance and other concerns later.
in my view the second option rendering partial is better because if you get a json data you are supposed to manipulate it like set style, create elements, set values and similar thing as per the need just after you get ajax response in javascript but if you render partial, you can design your view in that partial before and keep ready and just import using ajax call so there is no responsibility of handling the response data.
I would say it depends on how much data you're sending across the wire as you put it. If you're implementing a "load more" feature then it seems like you wouldn't want to take 5 seconds to load something. Google images is a good example of how quick that feature should be. If you know you'll never have that much data then rendering the partial on the server might be cleaner, but if you're requirements change then it's a hassle to go back to the first approach. So in short, I'd say the first approach allows for more disaster control as far as how long a load more takes to load a large amount of data. I'd say it's also general good practice to take load off server when you can, even if it's a little inconvenient on the client side for the developer.
I'm building a Django web application which has a lot of html generated on the fly by ajax requests. Right now I use Django's templating language to build up html and then pass this new HTML as a string in JSON object which is then injected into the page with jQuery.
This works fairly well, but with Javascript being so fast in modern browsers and with so many javascript template libraries being made I'm wondering if I should push everything clientside.
So my question is: Given the that my average "page" with all requests to and from it has to compile around ~300 templates (each of about 15 or so lines with 5 or so substitutions) out into HTML during its lifetime is there significant performance advantage to doing templating in the browser?
In addition can anybody reccomend a 'fast' Javascript templating library? I've heard good things about underscore.js, mustache.js and jQuery template.
The (massive) advantage of sticking with Django templates is that you only need to use one templating language, which retains the same capabilities regardless of the page you wish to generate. If you find that you're having performance issues then you should consider looking into caching template fragments.
I don't think an hybrid client and server templating architecture is bad. As long as you code a template only in one of the environment.
Every time you generate a page server side, an amount of processing time and some network bandwidth is consumed. This is what you pay if you use hosted servers.
While the user's browser is awaiting idle on a generally idle computer for the response.
If you send the template on the client(HTML + JS), they can be cached, for the session or even days, if the user do not delete them.This reduces the network traffic to deliver the same content various times. As the data are generally smaller than their equivalent rendered HTML.
As you point today's Javascript engines are really fast, as well as the computer they run it. Each time you sent the rendering work to the client, you save some processing time for your server and deliver faster the data.
We're at the other extreme side, as we run all on the client and that's why we created PURE for an ultra-fast client rendering. Our app looks very fast as a result of this decentralisation.
Why do you pass the HTML as JSON? Just send back the HTML and use jQuery's $.html() function to put it in the <div> or whatever.
As for templating in Javascript, there's Pure. If you're using jQuery (I'd recommend it), it already has a template engine.
It may be possible to make templates of dual purpose - to that they could be rendered into placeholders for the replacement by js and at the same time they could be rendered normally for the server output. Only a few templates will need to be of dual purpose like that - the fragments that are to be replaced by the js.
I agree with Ignacio, it is much better to keep just one copy of each template, so that you don't have to write a separate one for the javascript, however there is definitely room for improvement from the approach that I've mentioned above.
Ideally you might want to have templates compiled into robust javascript function code as well as plain string for the output by the server.
Closure templates called Soy, solve the problem nicely but do not (perhaps just yet) work with python, but they do work with Java and Javascript. Hopefully one day there will be python support for that.
But even if that happens, the templating language will be probably more restricted as it will be tough to make things like .get_absolute_url(), filters, etc work both in python and javascript - all automatically.