I gave the blog a new domain all of its own as it's focus is mostly rails and associated tech, and not much to do with my family and personal stuff.

The RSS feed is here.

Yes, the layout is dull, I'll get to it.

Most of the recent posts are already there, I'll keep this blog around as an archive of the older stuff.

What we need to do is to add an Expires header to our static content - one that gives a time a long way in the future so that the browser knows that the content should stays in the cache -and not be fetched again, or even checked (if it has been updated) from the server. But what if we want to change a css file or a js file? The user will get the old one from their own cache. No good.

Well Rails has a mechanism to prevent this - it adds a 10 digit number on the end of each url (as generated by image_tag, stylesheet_link_tag, or javascript_include_tag for instance) in the query string. That number is based on the file modification time - so when the file is updated then the URL will change. It's like having a remote way to expire the item in the browser's cache.

So it's simple right? Just turn on mod_expires in Apache? Not so fast. What about those images and items that do not have the query string? We don't want to send an expires header for those. Definitely not. (And you will likely have some - for instance images that are referenced from your css files.) If you do, the user is stuck with their cached version even if you change the file on the server.

So we need some way of selectively turning on the expires header in apache.

One way [Danny Burkes] (look at the update at the bottom) is to segregate by directory what you want to expire and what not. But this seems a bit clumsy to manage. (Also, side issue, I am not totally convinced that the munging of the urls provided by the plugin is needed - at least section 13.9 of the HTTP 1.1 spec seems to suggest that content with query strings will be cached fine if an explicit expires header is given.)

Actually you can do what you need with one symbolic link and some apache magic. The obvious magic won't work - you can't detect what's in a query string using a LocationMatch or FilesMatch container. But we can get around this with a rewrite rule, and a directory container. This is from my apache httpd.conf file:

  # add something we can do a directory match on
  RewriteCond %{QUERY_STRING} ^[0-9]{10}$
  RewriteRule ^(.*)$ /add_expires_header%{REQUEST_URI} [QSA]

  # the add_expires_header directory is just a symlink to public
  <Directory "/path/to/rails_app/public/add_expires_header">
    ExpiresActive On
    ExpiresDefault "access plus 10 years"
  </Directory>

We use this as something we can detect in a Directory container. And in there we turn on the expires header.

We need one more thing:

cd /path/to/rails_app/public
ln -s . add_expires_header

It's a big benefit - we have literally gone from 50 HTTP requests per page to about 16. And with some tweaking we'll get it down more - some of those are from references to images in css, but a few are due to us not using urls generated by rails for static content. And we can fix those.

I wanted to find where the system was spending its time, and the DB query logging provided by ActiveRecord was not enough to show me where the performance needed tuning.

The rails benchmark method to the rescue. And a little meta-programming. I added this to application.rb:

  if ENV['EXTRA_BENCHMARKING']
    @@added_methods = []
    def self.method_added(id)  # wrap a benchmark around all controller method calls
      name = id.id2name
      if name =~ /^old_(.*)$/
        return if @@added_methods.include?($1)
      else
        return if @@added_methods.include?(name)
      end
      @@added_methods << name
      alias_method "old_#{name}".intern, name.intern
      eval %{
        def #{name}(*args, &block) 
          logger.info("*=*=*=*=*=*=*=*=* #{name} START")
          self.class.benchmark("*=*=*=*=*=*=*=*=* #{name}", Logger::DEBUG, false) do
            send("old_#{name}", *args, &block)
          end
        end
      }
    end
  end

It's important to add this pretty much at the top of application.rb, or at least before all the methods you want to time.

Method_added gets called every time a method is added to the class. As each method is added we make a note of it in our array @@added_methods, and then alias it with the prefix "old_". Then we define a new method adding the logger calls and the benchmark call.

The output looks like this:

...
*=*=*=*=*=*=*=*=* save_url_history START
*=*=*=*=*=*=*=*=* save_url_history (0.00004)
*=*=*=*=*=*=*=*=* missing_page_catcher START
*=*=*=*=*=*=*=*=* get_categories START
*=*=*=*=*=*=*=*=* get_categories (0.00028)
*=*=*=*=*=*=*=*=* get_continents START
*=*=*=*=*=*=*=*=* get_continents (0.00004)
...

Also you will get all your regular debug level logging in between these new lines in your log.

Note that the @@added_methods array is important because both alias_method and the eval'd def result in a call to method_added - so we use the array to prevent infinite recursion.

Just run rails up with the EXTRA_BENCHMARKING environment variable set, and you are away:

EXTRA_BENCHMARKING=1 script/server -e production

ActiveRecord::Base.logger.level = Logger::DEBUG

Failing tests. We'd like them to pass of course. But there is something strange going on - the error is not a test error, it seems to come from rails itself:

  1) Error:
test_searcher_email(OrderMailerTest):
NoMethodError: You have a nil object when you didn't expect it!
You might have expected an instance of Array.
The error occurred while evaluating nil.[]
    /usr/local/lib/ruby/gems/1.8/gems/activerecord-1.15.4/lib/active_record/fixtures.rb:498:in `orders'
    ./test/unit/order_mailer_test.rb:22:in `test_searcher_email'

1 tests, 0 assertions, 0 failures, 1 errors

What does it mean? The clue here is that the error comes from fixtures.rb. The other thing to note is that I have a setup in my test:

  def setup
    ActionMailer::Base.delivery_method = :test
    ActionMailer::Base.perform_deliveries = true
    ActionMailer::Base.deliveries = []
   ...
  end

  def setup
    ActionMailer::Base.delivery_method = :test
    ActionMailer::Base.perform_deliveries = true
    ActionMailer::Base.deliveries = []
    ...
    super
  end

Started
.
Finished in 0.160284 seconds.

1 tests, 1 assertions, 0 failures, 0 errors

     alias_method :setup, :setup_with_fixtures

Gotcha 2

Ever tried to test your routes? You should. But if you have routes that depend on what's in your database you've got a problem. Why? Because the routing is initialised before any fixtures are loaded. In fact none of the test related code is loaded at that point. So your tests just won't work.

I don't know a good way to fix this. In the end I opted for a fudge at the beginning of routes.rb:

  if ENV['RAILS_ENV'] == "test" && Category.count == 0
    Category.create!(:name=>"Families and Friends", :alternative_name=>"Family")
  end

But at least the tests run ok now.

debug1: Remote: No xauth program; cannot forward with spoofing

It turns out that sshd has a hard wired idea of where xauth is supposed to be:

$ strings /usr/sbin/sshd|grep xauth
/usr/bin/xauth
xauthlocation
maxauthtries
No xauth program; cannot forward with spoofing.

# ln -s /usr/X11R6/bin/xauth /usr/bin/xauth

On with business though - Ruby has some serious wrinkles that produce counter-intuative and plain strange behaviour. One might say that they violate the Principle of Least Surprise. But POLS is somewhat personal - surprising to whom? (just me maybe!)

Surprising result #1

b=1
=> 1
a if a=b
NameError: undefined local variable or method `a' for main:Object
        from (irb):2

This is odd indeed. It seems that the Ruby parser thinks that a is a method at the beginning of the line. But when it parses a=b it will know (and remember) that it is a variable. However, when the condition passes and it comes to execute 'a', it does not apply its new found knowledge about the nature of a - it still goes for the method. And hence the error.

Surprising result #2

c=1 unless defined? c
=> nil
c
=> nil

This is a bit easier to explain. Once the parser has seen c=1 it enters c in the symbol table, it knows that c exists as a variable. Hence it is defined (as nil). So c=1 is never executed.

Surprising result #3

class A
  attr_accessor :a
  def b
    @a=0
    x=a
    a=x+1
    puts("a is #{a}")
    puts("@a is #{@a}")
  end
end
=> nil
c = A.new
=> #<A:0x32fb7c>
c.b
a is 1
@a is 0

No mystery here really - it's a nice gotcha though - a= does not call self.a=, it just assigns a local variable. It is just a little confusing unless you think carefully.

Surprising result #4

[0.0, 1/0.0].sort
=> [0.0, Infinity]
[0.0, 0/0.0].sort
ArgumentError: comparison of Float with Float failed

Thanks to Evan Weaver for pointing this out to me. This is due to 0/0.0 returning the value NaN (as opposed to 1/0.0 which returns Infinity). And the comparison with NaN by sort gives the strange error message.

Let me know if you spot any more oddities.

This helps, but actually all I needed to do was to simply say

gem update

gem install rails --include-dependencies

• Mysql full text search
• Lucene
• SOLR (server based on lucene)
• Ferret
• Sphinx
• Zebra (if you can ever figure out how to use it)

The idea is to write the shortest interpreter for the ETA programming language possible. Previously I had got the program down to 227 chars of tight and obfuscated Ruby.

With these updates, either 4 or 16 chars have been knocked off the old best effort, depending on whether you want warnings or not. And both the newer versions are compatible with Ruby 1.8.6, which is a good thing seeing as the older versions were failing to run.

Although obfuscated programming is on the whole a fairly useless pastime, one certainly learns all the strange little creepy corners of a language. And Ruby has a few. (And not just this). I'll explain more in a later post.

The first was a solver for the Bedlam cube. Here are the remarks, the makefile and the program itself.

The second was a one liner digital clock program. Here are the remarks, the makefile, the program itself, and finally the script for running it.

You can check out my previous entries here (2004) and here (2005).

This year the winners are being presented again at the Vintage Computer Festival on the 3rd and 4th November 2007 in Mountain View, California. It was a great event last year, and I'll go there again if I can this time. I recommend it if you are nearby.

Go have a look. And see if you can guess why I wrote about it :)

Truly, all Dave Watanabe's stuff is good though, & I have been using Newsfire as my main newsreader for a long time.

You want to write something like:

<% yellow_boxed do %>
some text and html and stuff
it's probably only worth doing it this way
if this bit is longer than a line or two
<% end %>

  def yellow_boxed(&block)
    concat(render(:partial=>"yellow_boxtop"), block.binding)
    content = capture(&block)
    concat(content, block.binding)
    concat(render(:partial=>"yellow_boxbottom"), block.binding)
  end

Note that I render a partial for the top and the bottom of my yellow box. The partials just contain table, tr and td tags and so on, but I wanted to keep the html strings out of my helper. But for short lengths of html it's not a problem, you can put text strings there instead of the much more expensive calls to render.