Thursday, April 20, 2006

First Experience with Ruby

It seems Ruby is becoming omnipresent these days. A lot of hype, hope and buzz seems to surround it. A few of my friends moved into ruby and went ooh-aah over the new fanged toy that does wonders. But when i tried to get them tell me what was so intersting and powerful, the common answer i got was that it was so simple to write code. Sufficiently piqued i decided to get some first hand experience.

So i borrowed a PragmaticProgrammer book on Ruby (online version can be found here) and started reading thru it. It is a good book with the only grouse being some of the topics were not covered in depth as i would have preferred.

Anyways after poring through half the book the list of features that spring out are

1. Fully OO

Everything seems to be a method connected to an object including new, loops etc. For e.g. to create a new instance of my object you do MyObject.new(). Numbers (java equivalent of int's) are all instances of FixedNum class so you can do things like 3.times, 3.step(30, 3). Both the 3.X methods are looping structures with the first one looping 3 times and the second behaving like a java for loop.

2. Iterators and Blocks

Ruby defines a bunch of iterators that operate over Ruby containers (Arrays, lists and hashes). Coming from the java world, to me an iterator just meant a way to loop through the contents of a collection. But Ruby iterator is a different beast that no only loops through the contents but also executes a 'Block' of code for each object in the collection. To make it seem less daunting, lets look at an example..

[ 1,2,3,4,5 ].each {|i| print i } This would print 1 2 3 4 5 on the console.

In Java, we'd have to do..

int numArr[] = {1,2,3,4,5}
for ( int i =0; i < style="font-weight: bold;">3. Closures


A closure is a block of code that retains the values of the instance/local variables it accesses or uses. So we can create a closure, which accesses a few local variables defined and set above it. This closure when passed to a method in a different class will retain the local variable values as it was in the original context.

Martin Fowler has a very good article on closures. Read it to know abt the power of closures.

4. Assignments and Operator Overloading

Assignments can be chained since each assignment returns that value as the result of the assignment expression. So doing a = (b = 1 + 2) + 3 will set a to 6 and b to 3. Also both if and case return the value of the last expression executed. so we can do i = if (num <>

Also operator overloading is supported. Damn java seems to be the only one not supporting it these days.

5. Getter/Setter

We can have getters/setters for object variables without writing methods. In ruby classes attr_reader and attr_writer followed by variable names will make the attributes readable/writeable.

6. Mixins

Ruby like Java supports only single inheritance. Mixin is a powerful concept where you define multiple modules and when u include them in your class all the methods in the mixins are accessible in your class. This mixin does not copy the code into the local class just references it, so any change to the module methods/variables in any class will affect globally.

For e.g. ruby comes with a standard module Comparable which defines comparision operators. For comparable to work we have to have implemented the method <=> in our class which the module uses. This is similar to an abstract base class in java but only we are allowed to have multiple abstract base classes for a single class. Pretty powerful !!

So Ruby has some features i like (all the ones above) but its arrays and collections are not strongly typed and i have a general aversion to such things having some bad memories from a VB/ASP project long ago.

Overall i have to think about some real life problems that ruby makes it easy to solve and actually scales well before actually going nuts over it.

Subscribe to comments for this post

Wednesday, April 19, 2006

10 Tips for good API Design

All along my life, i had complete control over my code. I could change method/class names, signatures at will and all users would be in the same code base. I could do endless refactoring. Now I have to maintain and keep enhancing an API. Sounds easy ? Its actually big trouble, every method you publish would end up being called by thousands of people and you lose your ability to refactor in a jiffy.

Public API's are forever and there is only one chance to get them right - Joshua Bloch

I learnt good API design is an art in itself and i also picked up some basic rules that should be followed. They are

1.Intuitive APIs

When designing an API, always consider specific examples of what code clients should write to use it. Model api's after commonly used usage patterns/api's. This will help us avoid cumbersome or unintuitive APIs.

2.Internal Code

All internal code that should not be called by public should go in a separate package marked internal. Like all public classes in com.company.product.* and internal classes in com.company.product.internal.*.

3.Expose only what's needed

Use the most stringent access specifiers possible (private/package provate/protected/public). The idea is to prevent unintended usage of the api's. Some common tips
1) All classes/functions that should be called by other classes in the same package should be declared as package private. Note there is no such thing as package private interfaces since all methods in a package private interface have to be declared public.
2) If a method needs to be called by classes in a different package then the tendency is to make it public even though it is not a true public api. In such cases, a good way to proceed is to make the class as abstract, then subclass this class in the private packages and use static factory methods to return an instance of the internal class. This way the methods would be exposed in private packages only.
3) Declaring a method as protected is as good as declaring it as public since any one can subclass and be able to call the method. It should be used only when it is sure that clients should be able to override them or subclasses are in different packages.
4) Make classes/methods final such that they cannot be overridden. Turning it around, assuming clients will extend any of the public classes which are not final it is unpredicatable to let clients overide selected methods without understanding the big picture.
5) No member variables should be public except constants (public static final)

4.Javadoc

All public methods/constants in a public class must be documented. The ease of use of your api depends on how good the javadoc is. A good clean javadoc means there is no implementation detail specified. All operations on data that are visibile to clients is well documented as are error conditions. Specify clearly what each method will and more importantly will not do.

5.Static Factory creation

Prefer using Static factory methods to create objects than using new. This allows us to create any object that implements the specified contract than exposing a truckload of classes to the public.
Use of a constructor forces the implementation to return an instance of the class itself rather than a subclass (or one of a set of subclasses). This would force the class to have knowledge of all variant behaviours.

6.Contexts

Dont store contexts in objects. Prefer to pass them as method parameters. If we store contexts in objects and pool those objects, then we have to ensure that the context is valid throughout object lifecycle. This is a big pain.

7.Thread safe classes

Immutable classes and fly weights are easily thread safe. Prefer to use them judiciously.

8.Helpers & Utils

Any util class that needs to be associated to a state should be made a helper that has to be instantiated with the state. Utils should be final and non instantiable and all methods should be static.

9.Class Names and Method names

Make names consistent throughout the api, length() method returns length of both String/StringBuffer. If you get the naming correct most people would be able to use it intuitively without poring over the docs. This makes the api's very simple to use.

10. Exceptions

Never throw a single type of exception with different messages. Use lots of different exception classes with the rule being roughly for each kind of error throw an exception. Be sure to put them in proper hierarchy like java.io.IOException so that clients have the flexibility to declare a single block for a general category of exceptions.
Always ensure that the exception message is picked up from a resource bundle or some other file so that messages can be easily translated. Hey, you never know who is going to use your api's !

When in doubt leave it out. We can always go in and add something later after more delibration.

Subscribe to comments for this post

 
Clicky Web Analytics