frog and code

Consider a simple list of to-do tasks:

<ol class="tasks">
  <li id="task-1">run</li>
  <li id="task-2">vote</li>
  <li id="task-3">code</li>
</ol>

There are three task instances (with ids 1, 2, 3), each of which is rendered as a list element referenceable by DOM id. Note the id format: <model>-<id>, e.g. "task-1". This pattern is frequently used to avoid collisions with ids for other models because DOM ids are global and cannot be qualified to a particular object type or namespace (e.g. Task v. Group). When using the <model>-<id> format, you'll often want to obtain just the model id portion, say, to send to the server to edit or delete a task. After you've written code like this a few times:

var modelId = element.id.split('-').last();

you'll be overcome by a strong urge to extract a method that hides the ugly and standardizes the approach. I use the Prototype library at work, so I've implemented a utility method as an extension to Prototype's Element, making it available to any html element obtained via $() or $$(). However, the idea is easily implemented in jQuery or a library-agnostic manner. Here's the code:

/*
 * Prototype Element extensions
 */
Element.addMethods({
  /**
  * Returns last delimited portion of dom id for id or element passed where element id is of the form:
  *   example-<id> OR example_<id>
  * Supports hyphen '-' or underscore '_' as delimiter. Returns null if element id is missing or does not
  * contain a delimiter. Some example ids:
  *   task-group-25 // Returns 25
  *   25 // Returns null; use element.id instead
  */
  modelId: function(element) {
    var id = $(element).id;
    if (id == null) return null; // Just in case; browsers tested return empty string for missing id.	

    var idParts = id.split(/[-_]/g);
    return (idParts.length > 1) ? idParts.last() : null;
  }
});

/*
 * Usage
 */

// Get single task id
$('task-1').modelId(); // Returns "1"

// Get all task ids
$$('.tasks li').invoke('modelId'); // Returns ["1","2","3"]

JavaScript, I love you, but I will never understand you. From the Firefox console:

>>> typeof 'test'
"string"

>>> 'test' instanceof String
false

>>> s = new String('test')
>>> typeof s
"object"

>>> s instanceof String
true

So the string "test" can be a "string", not a String, an "object" or a String. Good times. Then there's my personal favorite:

>>> typeof null
"object"

Null is an object? I object. In Ruby, nil is a full-blown object but in JavaScript, null is not. So when it comes to typeof, be careful what you ask for.

Douglas Crockford, perhaps the Godfather of JavaScript, comments further on typeof weirdness and offers an improved implementation.

Any developer who's had to deal with Date/Time calculations knows there are myriad issues to navigate, particularly around time zone and Daylight Saving handling. Today (Mon 2008.11.03), I ran our tests and stumbled upon the following code (simplified for this post) that was breaking:

# relapse_test.rb
def test_middle_time
  t = Time.now
  r = Relapse.new(:start_time => t-5.days, :end_time => t-1.day)
  middle = t-3.days
  assert_equal middle, r.middle_time # Breaking on Mon 2008.11.03
end

# Relapse class in relapse.rb
def middle_time
  return start_time + (end_time - start_time)/2
end

Results of the test:

<Fri Oct 31 11:27:18 -0400 2008> expected but was
<Fri Oct 31 11:57:18 -0400 2008>.

The purpose of middle_time() is to return the time halfway between the start and end times. The code seemed straightforward, but for some reason, the expected and returned values differed by 30 minutes.

As I stared at the code, there was another nagging question - why was this happening now when it wasn't prior, before the weekend. Hmmm ... a Halloween trick? To get more insight, I added some print statements to echo the times used by the middle_time calculation and got the following:

Start: Wed Oct 29 11:27:18 -0400 2008
  End: Sun Nov 02 11:27:18 -0500 2008

Bingo. Check out the time zone offset: -0400 for start time and -0500 for end. What happened over the weekend? Halloween in Salem was a good guess but the culprit was the switch from Daylight Saving back to Standard Time. Because of the time zone disparity, there is an extra hour between the start time, taking place in Daylight Saving time, and the end time in Standard Time. So (end_time - start time) in middle_time() adds an extra hour (4 days plus one hour) and, when divided by 2, produces the 30 minute discrepancy between expected and actual time.

Proper handling depends on your situation. In ours, we were interested in the middle DATE rather than the time, so the implementation was changed to use dates v. times. For situations where time does matter, the 30 minute discrepancy may, in fact, be desired as it factors in the number of hours that have elapsed. But if you want the calculation to be time zone-consistent (or agnostic), you can use the UTC (GMT) time zone; Time.now.utc returns the current UTC time. Calculations can be performed in UTC, which does not shift for Daylight Saving (or any other reason), and converted back to a zone-specific time as needed.

One of boons of using Ruby is its expressiveness and readability. A small detail that struck me when I started writing Ruby code was the ability to use words like and and or as keywords in conditional expressions (in lieu of && and ||). A nice touch that allows code to read more like English.

However, it turns out that the English forms of these conditional operators are not quite equivalent to their symbol counterparts. Consider the following exercise on the console:

>> username = 'frog'
>> password = nil
>> login_valid = !username.blank? && !password.blank?
>> puts login_valid
false

Login is valid if both username and password are not blank. Since password is nil, the value of login_valid is false as we would expect. Now let's try the same code, replacing && with and to represent the AND conditional operator.

>> username = 'frog'
>> password = nil
>> login_valid = !username.blank? and !password.blank?
>> puts login_valid
true

Whoa!? The login_valid variable now evaluates to true. What happened? Well, and is not simply an alias for &&, it's a different operator with a different level of precedence. Programming Ruby by Dave Thomas contains a table of operators in order of precedence and notes that "the word forms of these operators (and, or, and not) have a lower precedence than the corresponding symbol forms (&&, ||, and !)." More specifically, the word forms are lower in precedence than assignment operators (= in the example above), while the symbol forms are higher in precedence.

So in the second example, the variable login_valid is assigned the value of !username.blank? before the and condition !password.blank? is evaluated. Crazy, and a little scary - I'd expect both forms to work the same way, or at least both to have precedence over assignment operators. The solution in this case is either to use && as in the first example or wrap the full condition in parentheses to specify precedence explicitly:

>> username = 'frog'
>> password = nil
>> login_valid = (!username.blank? and !password.blank?)
>> puts login_valid
false

Rails 2.1 was released last week at RailsConf with a lot fanfare and sexy features like timestamped migrations, named scopes and better timezone handling. There was also a minor fix that I was happy to see: String#titleize now works with acronyms. Prior to 2.1:

>> "ALS community".titleize
=> "Als Community"

Not terribly desirable that "ALS" suddenly has lowercase letters because it's used in a title. But now in 2.1:

>> "ALS community".titleize
=> "ALS Community"

Ah, much better. I had patched String in my projects to correct this. Now the band-aid can come off. Interestingly, it looks like this bug was not reported (shame on me); instead the fix was made in response to an issue where the character following an apostrophe was being incorrectly capitalized:

>> "frog's code".titleize
=> "Frog'S Code"

The change to address this case also solves the acronym problem. Details of the fix can be found here:

http://dev.rubyonrails.org/changeset/8533