Backbone.js, Rails 3 and asynchronous interfaces


Backbone.js and RailsCurrently at Monterail we’re building a frontend-heavy app. With a lot of cool stuff. Like preloading data for daily views so that you won’t have to wait for every surrounding day to load. Many, many interactive elements that should be really interactive – not in an oldschool, ajax “spinner” way. And a lot of other stuff we shouldn’t talk about yet.

We decided to put more emphasis on Backbone.js with Rails and build most of the app with this combination.

Below you can find some of our findings on the topic of making those two speak to each other in a nice manner.

Asset pipeline

Really makes things easier. Organizing your JS code before it meant a lot of things: handling dependencies, minifying & compressing the code, organizing it into a lot of different files and building connections between them. With asset pipeline and it’s awesome “require” directive, you can forget about handling it manually. You just create the structure in app/assets/javascripts. Or app/assets/javascripts/backbone. Your choice. You can have many different files in whatever directories you like – in the end you’ll get one (or more, if you decide to) JS file that has everything your webapp need.

Don’t forget to:

  • gzip your JS/css assets
  • minify them!

Rails-backbone gem

Also makes things easier. It puts backbone.js (& underscore.js) library code into your asset pipeline. It gives you some great generators to get started if you don’t know where to start. Definitely worth including in your Gemfile.

http://rubygems.org/gems/rails-backbone

Coffeescript & underscore.js

Obviously. It makes Javascript code feel fresh. You can fold, iterate and do a lot of stuff. Be sure to check out underscore.js docs. And coffescript docs, of course, but I bet you know it already.

“V” in MVC done right

Views are classes. You handle logic inside them using the hardcore code, which is receiving so much hate in Rails views. They can have many methods, they do not have to render a single “action”. And templates are templates – and nothing else. There is a lot of discussion recently on the matter of Views/Templates in Rails, but you can start implementing it right away. Go and try it. You’ll fall in love.

You have many options for templates – eco, Handlebars.js, {{ mustache }}, but we recommend haml_coffee_assets. It gives you the HAML you (probably) already know with the ability to use coffeescript within it.

Single-page app

Yes. You will use Rails to render only the core view of the application. Of course, you will have to handle all things around the core of your app.

For example, you should probably leave out:

  • authentication logic: sign up / in, reset password
  • steps necessary to use the app: billing pages
But, other than that, you can use the views rendered in JS. What does that mean?
  • if they do not include any dynamic data, they are lightning fast
  • if they include dynamic data, you can preload it and they still will be lightning fast
Backbone comes with it’s own router. You can utilise pushState without any effort. How cool is that?

Asynchronous interface

Since you’ll be doing most of the association/validation work in background, you don’t have to worry too much about server responses. It’s the same with rendering elements: you’ll be doing it in JS, not Rails. You don’t have to wait for the request to return the rendered element to actually display them. If you create something, the UI should respond right away and the ajax request can persist the data with it’s own pace.

You should, however, prepare a global “something went wrong, reload the page” request rescuer – in case of network outages, backend app failures and so on. But it’s way easier to manage.

API

You get it for free. It doesn’t have to be public, but if you decide to, it’s a matter of adding OAuth as the way of authenticating requests and it’s done. (Or you could go with http basic auth, which is obviously bad for a lot of reasons).

With smart caching (give redis-store a try), you’ll save tons of server power. You know, relating to the old ruby saying, “views are most resource-consuming”.

Conclusion

There are, obviously, downsides, which I’ll cover in future (I silently hope that there won’t be any reason to do that). And topics I haven’t covered. I’m curious about other stacks and ways of doing things. Share in the comments or ping me on Twitter!

Testing AJAX-reloaded elements with Capybara

Capybara is a great tool for testing application flow and user interfaces. Thanks to Culerity and Selenium web drivers you can test javascript and AJAX features of your apps.

However testing AJAX is not that straightforward because of asynchonous nature of these requests. In order to test effects of AJAX calls a web driver must wait until they finish which is difficult if not impossible to detect. However this can be worked around using some tricks.

Scenario 1

Let’s say you want to test a link on the page that triggers an AJAX request which eventually inserts some element in DOM. This could be an “Edit” link that retrieves object’s data from database (via AJAX call) and then shows an HTML form on the page. Test scenario might look like this:

@selenium
Scenario: Clicking "Edit" should show edit form
  Given I am on product's page
  And I click "Edit" link
  Then I should see form

The form will not appear immediately after the link is clicked because it takes some time for your app to process the AJAX request and return the response. Capybara is intelligent enough to acknowledge this and instead of expecting a form to appear immediately it periodically looks for it in page’s DOM. You can define for how many seconds it should keep looking by setting following option:

Capybara.default_wait_time = 5

If the form doesn’t appear in this specified time frame, the test fails. Note that Capybara doesn’t always wait full 5 seconds, it simply moves on to the next step as soon as the form appears.

 Scenario 2

Now you want to test a link that removes an element from DOM. This could be a “Save” link that saves object to database (via AJAX call) and then removes the form. Test scenario:

@selenium
Scenario: Clicking "Save" should remove edit form
  Given I am on product's page
  And I click "Save"
  Then I should not see form

This test fails even if your app works as expected. I guess you see the problem. Capybara finds the form on first lookup which is performed immediately after the click, but at that point the AJAX call has not completed yet, so the form is still there.

Popular solution is to explicitely tell Capybara to wait until it starts looking for changes in DOM. In other words, give the AJAX calls chance to complete. Here’s an adequate cucumber step:

Given /^I wait (\d+) seconds?$/ do |sec|
  sleep(sec.to_i)
end

Rebuilt scenario would look like this:

@selenium
Scenario: Clicking "Save" should remove edit form
  Given I am on product's page
  And I click "Save"
  And I wait 5 seconds
  Then I should not see form

The downside is that it will always wait 5 full seconds now.

Other aspects

Neither explicit waiting with sleep, nor Capybara’s default_wait_time option guarantees 100% success. Your tests may still ocassionally fail if the AJAX requests they perform takes longer than you assume. The time it takes for the app to process such request may be quite random as it depends on many aspects like machine load or external services it hits (database, search engines, etc.). So remember to set the option to a value high enough for your app.


Fixtures without validation with Factory Girl

Factory Girl is my fixture replacement library of choice. It improves tests readability and maintainability. It’s also customizable.

There are sometimes situations when you want to create test scenarios that checks how your app is handling invalid data (not user input, but invalid records that already sit in your database). To do this you first need to put this invalid data to your db.

You could accomplish this with such line:

@user = Factory(:user, :email => "not a correct email address")

However, factory girl would raise an exception here, because that’s the default strategy of creating new fixtures – raise exception if save fails (because of validation errors for example).

Thankfully we can use our own strategy of creating new fixtures, such that does save records without validation.

First let’s define our new strategy:

class Factory::Proxy::CreateWithoutValidation < Factory::Proxy::Build
  def result
    @instance.save(false)
    @instance
  end
end
 
class Factory
  def self.create_without_validation (name, overrides = {})
    factory_by_name(name).run(Proxy::CreateWithoutValidation, overrides)
  end
end

Now we can use it while defining new factory:

Factory.define :invalid_user, :class => User, :default_strategy => :create_without_validation do |f|
  ...
end

And then we can happily create invalid fixtures without any exceptions raised.

Turning off auto timestamping for testing in Rails

Suppose that you implemented a functionality that depends on values of created_at or updated_at fields of your models. How do you test it?

If you use fixtures that reside in test/fixture/*.yml files then there is no problem, because the values you set there for created_at and updated_at fields are saved to the database ‘as is’. So you can easily have an article created one week ago:

article:
  title: What a great day
  created_at:
  updated_at:

However, I don’t use fixtures files myself. I feel a bit dirty using them ;) I find fixture replacement tools far more maintainable. Namely, I love thoughtbot’s Factory Girl. But here comes the problem. This won’t work as expected with Factory Girl:

Factory(:article, :created_at => 1.week.ago, :updated_at => 1.week.ago)

That’s because ActiveRecord’s automatic timestamping feature sets Time.now for created_at and updated_at fields overriding our values. At least that’s ActiveRecord’s default behavior. Fortunately it can be disabled with:

Article.record_timestamps = false

Chances are that after creating a model with a custom timestamp we’ll want to turn automatic timestamping back on. But turning it off and on in many places in your unit tests would be pretty cumbersome. Wouldn’t it be cool if you could achieve all of this with a snippet below?

without_timestamping_of Article do
  Factory(:article, :created_at => 1.week.ago, :updated_at => 1.week.ago)
end

It turns timestamping off, executes the block and turns timestamping back on. I find it clean and dry. Here’s the code to place in your test_helper.rb:

# test_helper.rb
class Test::Unit::TestCase # or class ActiveSupport::TestCase in Rails 2.3.x
  def without_timestamping_of(*klasses)
    if block_given?
      klasses.delete_if { |klass| !klass.record_timestamps }
      klasses.each { |klass| klass.record_timestamps = false }
      begin
        yield
      ensure
        klasses.each { |klass| klass.record_timestamps = true }
      end
    end
  end
end

Of course you can turn off timestamping for many models at once:

without_timestamping_of Article, Comment, User do
  Factory(:article, :created_at => 1.week.ago, :updated_at => 1.week.ago)
  Factory(:comment, :created_at => 1.day.ago)
  Factory(:user, :updated_at => 5.hours.ago)
end

Hope you like it. If so, share :)

Dynamic cookie domains with Rack’s middleware

Handling sessions in multi-domain environment is not the simplest things to do, because of the fact that cookies are scoped to a domain they were set by.

Recently we were developing an application with such an idea in mind:

  • Application will work as a base for other mini-applications (which we call sites)
  • Each site can be accessed via different url types: site.example.org and example.org/site
  • We want the users to remain logged in when switching from one url type to another

I won’t be covering application structure, routing, etc. here, I will only write about maintaing the sessions is such an environment.

So this is pretty simple here – all that we needed to do was to set cookie domain to .example.org (note the “dot” at the beginning). This could be done via:

ActionController::Base.session = {
  :domain => ".example.org"
}

However there was an additional requirement that we need to deal with:

  • Each site can be accessed via custom domain – site.com
  • Of course there’s no way here to keep the user logged in when he’s switching from site.com to example.org/site or site.example.org, at least it cannot be done with setting cookie domain to whatever value

Technically, to access the site via site.com, that domain must point to our IP address. Then we need to detect that the site is being accessed via custom domain and set cookie domain respectively.

This could be done via some funky before_filters in an Application Controller, however we found much better and cleaner way.

Rack’s middleware to the rescue

Rack itself is a minimal interface between web server and your ruby framework. It’s used by Ruby on Rails (since 2.3) and Merb. The request comes from web server, goes through middleware layers and enters the application.

So we wrote a middleware layer that detects the host with which our application is accessed and sets cookie domain for the request. Here it is:

app/middlewares/set_cookie_domain.rb

class SetCookieDomain
  def initialize(app, default_domain)
    @app = app
    @default_domain = default_domain
  end
 
  def call(env)
    host = env["HTTP_HOST"].split(':').first
    env["rack.session.options"][:domain] = custom_domain?(host) ? ".#{host}" : "#{@default_domain}"
    @app.call(env)
  end
 
  def custom_domain?(host)
    domain = @default_domain.sub(/^\./, '')
    host !~ Regexp.new("#{domain}$", Regexp::IGNORECASE)
  end
end

Now we need to turn it on:

environment.rb

config.load_paths += %W( #{RAILS_ROOT}/app/middlewares )

production.rb

config.middleware.use "SetCookieDomain", ".example.org"

.example.org is the default domain that will be used unless the application is accessed via custom domain (like site.com), we give it different values depending on environment (production/staging/development etc).

And since we’re fans of test driven development, here’s the test that ensures us that everything works as expected:

tests/integration/set_cookie_domain_test.rb

require 'test_helper'
 
class SetCookieDomainTest < ActionController::IntegrationTest
 
  context "when accessing site at example.org" do
    setup do
      host! 'example.org'
      visit '/'
    end
 
    should "set cookie_domain to .example.org" do
      assert_equal '.example.org', @integration_session.controller.request.session_options[:domain]
    end
  end
 
  context "when accessing site at site.com" do
    setup do
      host! 'site.com'
      visit '/'
    end
 
    should "set cookie_domain to .site.com" do
      assert_equal '.site.com', @integration_session.controller.request.session_options[:domain]
    end
  end
 
  context "when accessing site at site.example.org" do
    setup do
      host! 'site.example.org'
      visit '/'
    end
 
    should "set cookie_domain to .example.org" do
      assert_equal '.example.org', @integration_session.controller.request.session_options[:domain]
    end
  end
 
end

Test is sponsored by great Shoulda and Webrat gems.

Feel free to comment and share.

Rails 2.2.2, Ajax and respond_to

As I wrote some time ago in the article about Rails, Ajax and jQuery, sometimes there are problems with Rails not interpreting correctly content type headers of ajax requests. It’s because not all web browsers send that header in the same way. 

What I proposed was to sort the request.accepts array (array containing content type headers sent by browser) so that xml content type would be the first element. That would then trigger format.xml in our respond_to block.

However that approach does not work in Rails 2.2.2, because now the request.accepts array is frozen and it cannot be modified. I spent some time googling for the solution, but with no effect. So I dived into the API and Rails’ source code and came up with pretty nice and simple solution to the problem.

class ApplicationController < ActionController::Base
  before_filter :xhr_to_xml
 
  def xhr_to_xml
    request.format = :xml if request.xhr?
  end
end

This piece of code is an equivalent of the snippet I proposed in the article I referred to at the beginning. Now all ajax request will trigger format.xml in respond_to blocks.

Materials from AIESEC Conference in Częstochowa, Poland

Last month the netguru team has been invited to Corporate Responsibility and Information Technology Conference organized by AIESEC (international student organization) which took place in Częstochowa, Poland. We went there and gave some talks on various topics and I think some of you may find materials from this conference quite interesting.

First, we led an unofficial discussion on “The (fading) differences between desktop and web applications”. We introduced the topic to our hearers and then let them express themselves, it was quite successful.

On Day 2, we organized Ruby on Rails workshops which consisted of “Introduction to RoR” talk and one-hour live-coding session during which a simple twitter like application has been developed. You can find the results on our Github account.

We finished Day 2 with a talk on Enterprise 2.0 and on Day 3 we talked about “Developing a web-development company (startup)”, which was quite different from what people heard on earlier talks gave by the representatives of big corporations. I have no materials to share on this, though. Sorry.

Rails, Ajax and jQuery

The more ajaxified application, the more fun it is to use. But it is also more painful do develop. What is written below is my approach to pairing Rails and Ajax. It’s a mix of tips I found over the net on blogs and forums. I use jQuery for JavaScript, but I don’t use jRails or any JS/Ajax helper methods provided by Rails. Note that all Javascript/HTML code presented here can be used even if you dont use Rails or Ruby as your web development platform. Let’s begin.

Rails is RESTful

Thanks to Rails’ RESTfulness the only thing to take care of server side is setting proper response in controllers’ actons.

class PostsController < ActionController::Base
  def index
    @posts = Post.find :all
    respond_to do |format|
      format.html
      format.xml { render :xml => @posts.to_xml }
    end
  end
end

Rails decides which format block to call basing on routes defined in routes.rb file (map.connect ‘:controller/:action/:id.:format’) and accept headers sent with request by the client.

In most cases we want Ajax requests to trigger format.xml blocks in our controllers’ actions, so we need to set proper accept headers. Let’s do it just once with application-wide setting.

// All ajax requests will trigger the format.xml block
// of +respond_to do |format|+ declarations
$.ajaxSetup({
  'beforeSend': function(xhr) {xhr.setRequestHeader("Accept", "text/xml")}
});

Browsers’ quirks

There is something worth noting here, a problem I had once with IE and Safari. The code above may work differently in various browsers. Browser set text/html accept header by default. Here IE and Safari will append text/xml to it so you’ll get something like ‘text/html; text/xml’, while Firefox will replace text/html with text/xml and you’ll get ‘text/xml’ only. This is very important because Rails will take the first format it detects in accept header and trigger respective block in controller’s action, which will be html for IE and Safari. Here’s a fix for this that shifts application/xml (if it is present) to the beggining of accept headers array.

class ApplicationController &lt; ActionController::Base
  before_filter :correct_safari_and_ie_accept_headers
 
  def correct_safari_and_ie_accept_headers
    request.accepts.sort!{ |x, y| y.to_s == 'application/xml' ? 1 : -1 } if request.xhr?
  end
end

Ajaxify your links

Here’s a quick way to ajaxify your existing links. Add this JavaScript to your application.js file.

jQuery(document).ready(function() {
  // All A tags with class 'get', 'post', 'put' or 'delete' will perform an ajax call
  jQuery('a.get').livequery('click', function() {
    var link = jQuery(this);
    $.get(link.attr('href'), function(data) {
      if (link.attr('ajaxtarget'))
        jQuery(link.attr('ajaxtarget')).html(data);
    });
    return false;
  }).attr("rel", "nofollow");
 
  jQuery('a.post').livequery('click', function() {
    var link = jQuery(this);
    $.post(jQuery(this).attr('href'), "_method=post", function(data) {
      if (link.attr('ajaxtarget'))
        jQuery(link.attr('ajaxtarget')).html(data);
    });
    return false;
  }).attr("rel", "nofollow");
 
  jQuery('a.put').livequery('click', function() {
    var link = jQuery(this);
    $.post(jQuery(this).attr('href'), "_method=put", function(data) {
      if (link.attr('ajaxtarget'))
        jQuery(link.attr('ajaxtarget')).html(data);
    });
    return false;
  }).attr("rel", "nofollow");
 
  jQuery('a.delete').livequery('click', function() {
    var link = jQuery(this);
    $.post(jQuery(this).attr('href'), "_method=delete", function(data) {
      if (link.attr('ajaxtarget'))
        jQuery(link.attr('ajaxtarget')).html(data);
    });
    return false;
  }).attr("rel", "nofollow");
 
  jQuery('a.get, a.post, a.put, a.delete').removeAttr('onclick');
});

Just add a CSS class .get, .post, .delete, or .put to a link to make turn it into an ajax-link. I recommend you use LiveQuery plugin which will automatically bind click events to new links that appear on the page (loaded with Ajax call for-example). You can optionally set ajaxtarget attibute of the link. It expects a selector of a container in which you want to place the response.

link_to 'my cool article', article_path(@article), :class => 'get', :ajaxtarget => '#article_container'

Ajaxify your forms

For this you’d need jQuery Form Plugin.

  jQuery('form.ajax').livequery('submit', function() {
    jQuery(this).ajaxSubmit();
    return false;
  });

Now all your forms that have “ajax” class will be submitted via Ajax.

<form class="ajax">
  ...
</form>

CSRF and authenticity token

Rails has built-in protection from cross-site request forgery attacks. It relies on an authenticity token which Rails look for when dealing with POST, PUT or DELETE requests, so this token needs to be sent by the browser together with the request. The token is automatically added as a hidden field to any form you create with form_for method, it is also attached to links that have :method param set to :post, :put or :delete. In fact the token is added dynamically by Javascript code placed in link’s onclick attribute. However in one of code snippets above we stripped that onclick attribute from links to prevent the page reload after we click the link. Now we need to attack that token ourselves. First we will alter our application layout:

<head>
  <% if protect_against_forgery? %>
    <script type='text/javascript'>
    //<![CDATA[
      window._auth_token_name = "#{request_forgery_protection_token}";
      window._auth_token = "#{form_authenticity_token}";
    //]]>
    </script>
  <% end %>
</head>

Now we need to ensure that the token is sent together with ajax requests.

jQuery(document).ready(function() {
  // All non-GET requests will add the authenticity token
  // if not already present in the data packet
  jQuery("body").bind("ajaxSend", function(elm, xhr, s) {
    if (s.type == "GET") return;
    if (s.data && s.data.match(new RegExp("\\b" + window._auth_token_name + "="))) return;
    if (s.data) {
      s.data = s.data + "&";
    } else {
      s.data = "";
      // if there was no data, $ didn't set the content-type
      xhr.setRequestHeader("Content-Type", s.contentType);
    }
    s.data = s.data + encodeURIComponent(window._auth_token_name)
                    + "=" + encodeURIComponent(window._auth_token);
  });
});

We’re done, we have our ajax requests protected from CSRF attacks.

Modifing page after Ajax calls

Standard way to do page modification after Ajax call is to use Javascript code that inserts content returned by the call somewhere on the page. The other method is to put the modifying code in views that are returned by the server and just execute it in the browser. For this I’d recommend another jQuery plugin – Taconite. As the author says: “The jQuery Taconite Plugin allows you to easily make multiple DOM updates using the results of a single AJAX call. It processes an XML command document that contain instructions for updating the DOM”. Thanks to this you can for example easily use flash messages in your Ajax views.

Let this be a part of your usual layout:

<div id="flash_notice" class="flash"<%= ' style="display: none"' unless flash[:notice] %>><%= flash[:notice]  %></div>

Now let this be your taconite layout you’d use when returning views for Ajax requests:

<taconite>
  <hide select="#flash_notice" />
  <% if flash[:notice] %>
    <replaceContent select="#flash_notice">
      <%= flash[:notice] %>
    </replaceContent>
    <fadeIn select="#flash_notice" arg1="slow" />
  <% end %>
  <%= yield %>
</taconite>

This will display flash notice messages with fade-in effect after Ajax requests. Similarly you can update other elements of the page.

What’s in your toolbox?

I would love to hear from you on how you deal with Ajax in your web applications. What libraries/plugins do you use?

Uploading to multiple S3 buckets with Paperclip and Rails

When your application uses many static files (photos for example), you should consider placing these files on different hosts to improve the speed at which they are downloaded by web browsers.

Most web browsers have a limit on how many simultaneous connections can be made to a single named host. And this limit is usually two. It means that if you let’s say display many photos on a single page, user’s browser can only download two at a time. The broadband internet connection will be no help here. The browser will not fully use it, because it will keep opening and closing connections.

With little server and Rails configuration you pretend to be serving your files from different hosts and trick the browser (read more). This is easy if you keep all the files on your own server, but when you use Paperclip to upload files to Amazon’s S3 servers it gets more trickier. In fact Paperclip doesn’t support uploading to different hosts (buckets).

First, notice that these addresses means the same for Amazon’s S3:

  • http://s3.amazonaws.com/bucket_name/filename.ext
  • http://bucket_name.s3.amazonaws.com/filename.ext

So all we need to do is to make Paperclip upload to different buckets and return attachment’s url of the latter type. I wrote an extension that accomplishes this and included it in PaperclipExtended. All you need to do is download both Paperclip and PaperclipExtended and change your model definition.

class User &lt; ActiveRecord::Base
    has_attached_file :avatar,
                      :storage =&gt; :s3,
                      :s3_credentials =&gt; "#{RAILS_ROOT}/config/s3.yml",
                      :path =&gt; "avatars/:id/:style_:extension",
                      :bucket =&gt; lambda do |attachment|
                        i = attachment.instance.id % 4
                        "bucket_#{i}"
                      end
end

This will place each avatar in one of four buckets: bucket_0, bucket_1, bucket_2 or bucket_3. The exact bucket is chosen at runtime and in this case it’s based on models id.

Getting attachment’s url:

puts User.find(1).avatar.url(:original)
# =&gt; http://bucket_1.s3.amazonaws.com/avatars/1/original.jpg

If you have a page on your website that displays many uploaded photos, placing them in different buckets should make noticeable difference to user – they will load faster.

SwfUpload and Flash 10


SwfUpload is a graceful combination of flash and javascript that provides a mechanism for uploading files which is far more user-friendly than standard <input type=”file” />. It allows to upload many files at once and displays nice-looking progress bar.

Not everything works fine in every environment, though. Some problems are caused by the differences between some versions of flash player or even by bugs in flash players. Many of them can be solved, I wrote a post about it some time ago. But now when the Flash 10 came out, we have new obstacles to overcome.

Flash 10 is more restrictive about security issues that it’s predecessors. Now you cannot trigger opening file-select window from outside of the flash and it affects SwfUpload which used to do it from Javascript.

Thankfully, developers reacted quickly and are now working on new version (2.2) of the library. Alpha version is already out there and it deals with the mentioned problem. Since it is only alpha I recommend upgrading to stable version as soon as it comes out.

Here are inctructions how to reconfigure your app to work with SwfUpload 2.2.

First, download the most recent version of the library from SVN repository:

svn co http://swfupload.googlecode.com/svn/swfupload/trunk/core swfupload

You’ll only need two files from this: swfupload/swfupload.js and swfupload/Flash/swfupload.swf. These are the new version of files you should already have in your application. Replace old ones with these.

The new SwfUpload creates “Select files” button on it’s own, you only need to provide a placeholder for it.

<div id="swfuploadButtonPlaceHolder"></div>

The last thing is to update some Javascript.

swfu = new SWFUpload({
...
button_placeholder_id: "swfuploadButtonPlaceHolder",
button_image_url: "../images/swfupload/button.jpg",
button_width: "216",
button_height: "25"
...});

“button_image_url” accepts a path relative to the swf file. The image it points to, should look somewhat like this:

One image containing three states of a button (normal, on mouse over, on mouse click). You can have text instead of image if you want:

swfu = new SWFUpload({
...
button_text: '<span class="theFont">Upload</span>',
button_text_style: ".theFont { font-size: 16; }"
...});

That’s it. You can see working example I based on here.

If you find it useful, please share.