How is precedence of binary messages implemented in Pharo Smalltalk?

Question

I was looking at the syntax of Pharo Smalltalk, and was wondering how is precedence of binary messages implemented.

How does one go about declaring such a binary message?

How does the system figure out the precedence over unary messages?

Answer 1

For implementing binary messages:

Yes, you can look at the number class for example. It has a lot of binary messages there. Or consider this method of Object:

-> anObject
    ^ Association basicNew key: self value: anObject

This allows you to evaluate 'five' -> 5 and get an association.

For precedence:

This is done by a parser. So first of all it looks for keyword messages, then binary, then unary, then parents.

So if you have

collection add: 'five' -> 5

The parser will first of all parse add: with receiver collection then it parses 'five' -> 5 and puts it a a parameter of add:. This way the AST is composed in the way that keyword messages are more general and will be executed after their attributes are computed

Where to find

In the latest versions of Pharo the parsing is done with RBParser (same parser used previously for refactoring). Keyword messages are parsed with parseKeywordMessage and binary with parseBinaryMessage. The easies way to explore would be to put a one time breakpoint on parseBinaryMessage and execute something like

RBParser parseExpression: 'coll add: #five -> 5'

The debugger will stop on parseBinaryMessage so you can look at the stack and see how it is working. Pay attention, breaking once is important, otherwise you will get a debugger every time you compile a method or so on.