Everything I've read about Smalltalk makes an effort to stress objects are capable of doing three things: maintaining state, receiving messages, and sending messages. The understanding that an object can maintains its own state is common across many languages, but message passing may be unfamiliar to programmers working with other languages (though message passing does play an important role in Objective-C). Alan Kay has said that message passing is the most important concept to understand in Smalltalk, and that objects have been over-emphasized.
Smalltalk's object model follows the classical (class-based) style; the programmer writes a class which is then used by the runtime as a blueprint for instantiating an object. But unlike C++ or Java, the programmer never works with objects by invoking their methods directly in Smalltalk. Instead, messages are sent to the object which receives a message, looks up the appropriate method to perform, and then executes the method. This is more than a conceptual difference; it's a concrete implementation detail as explained by the following quote from Wikipedia's Objective-C article:
There are three types of messages in Smalltalk: unary messages that take no arguments, binary messages which have one argument, and keyword messages that can take one or more arguments.
Unary messages appear after the object instance they're sent to. The cr message is an example of a unary message that can be sent to a Transcript object, which responds by displaying a newline character.
anIntegerArray.atPut(5, 42) .
In statements that contain multiple messages, the order of precedence is from left to right with unary messages being sent first, then binary messages, and finally keyword messages, though the order can be influenced by parenthesis. Recall the example2 + 1 / 4 from Post 3. The expression contains two binary operators, + and /. Following the precedence rules, 2 is sent the message + with the argument 1 and then the message / with the argument 4. Contrast this with how the statement would be evaluated in Java following the mathematical order of operations. To force that execution in Smalltalk, the expression would need to be parenthesized as 2 + (1 / 4) .
It seems the biggest benefit of the approach is that its late/dynamic binding affords the flexibility of duck typing in a compiled language... but now that dynamic languages are widely used, and something like an inferred typing system can give the "feel" of dynamic programming in a strongly-typed language, I wonder if this is only historically important now. There is a runtime performance hit since method lookups happens at execution instead of compilation, and some Smalltalk implementations have explored method caching to help mitigate the impact, but it seems there must be a whole class of compile-time optimizations that cannot be performed by the compiler that may could affect performance as well.
I asked about the advantages and drawbacks of message passing on the PiLuD mailing list. Some of the respondents made pretty good points, and I recommend reading the exchange if you're interested in hearing some other opinions from a language-design perspective.
Smalltalk's object model follows the classical (class-based) style; the programmer writes a class which is then used by the runtime as a blueprint for instantiating an object. But unlike C++ or Java, the programmer never works with objects by invoking their methods directly in Smalltalk. Instead, messages are sent to the object which receives a message, looks up the appropriate method to perform, and then executes the method. This is more than a conceptual difference; it's a concrete implementation detail as explained by the following quote from Wikipedia's Objective-C article:
In a Simula-style language [like C++, Java, etc.], the method name is in most cases bound to a section of code in the target class by the compiler. In Smalltalk and Objective-C, the target of a message is resolved at runtime, with the receiving object itself interpreting the message.Perhaps a suitable way of looking at it is that all of the properties and methods of an object are private. The only way to interact with the object is by sending it the appropriate messages.
There are three types of messages in Smalltalk: unary messages that take no arguments, binary messages which have one argument, and keyword messages that can take one or more arguments.
Unary messages appear after the object instance they're sent to. The cr message is an example of a unary message that can be sent to a Transcript object, which responds by displaying a newline character.
Transcript cr.Binary messages appear with an argument. For example, the binary message = can be sent to foo with the argument 42, and foo responds by comparing its value with the argument and returning a Boolean object set to whether the values were equal.
foo = 42.Keyword messages end with a colon and can take multiple arguments. The at:put: message is an example of a keyword message which can be sent to an IntegerArray object. The array responds by placing the value 42 at index 5.
anIntegerArray at: 5 put: 42.Keyword messages are a single message. That is, at:put is not an at: message followed by a put: message. In Java, it might be written like
In statements that contain multiple messages, the order of precedence is from left to right with unary messages being sent first, then binary messages, and finally keyword messages, though the order can be influenced by parenthesis. Recall the example
It seems the biggest benefit of the approach is that its late/dynamic binding affords the flexibility of duck typing in a compiled language... but now that dynamic languages are widely used, and something like an inferred typing system can give the "feel" of dynamic programming in a strongly-typed language, I wonder if this is only historically important now. There is a runtime performance hit since method lookups happens at execution instead of compilation, and some Smalltalk implementations have explored method caching to help mitigate the impact, but it seems there must be a whole class of compile-time optimizations that cannot be performed by the compiler that may could affect performance as well.
I asked about the advantages and drawbacks of message passing on the PiLuD mailing list. Some of the respondents made pretty good points, and I recommend reading the exchange if you're interested in hearing some other opinions from a language-design perspective.
Comments
Post a Comment