hi, i have a testsuite that is a subclass of TestCase with this method:
testStupid |a| a :=#(1). self assert: a first isNumber . a :=a at:1 put: nil.
if i run the test once, it succeeds, but but when i run it a second time it fails. this seems to be sligthly counterintuitive, or how do i have to understand it?
Interesting, I would never have guessed the test would be broken at first sight. So maybe some people enlightened in the ways of the compiler can provide a better explanation than me. Here is my guess:
#(1) is a special construct which creates a compile-time array. The compile-time makes the trick: in short the array is more or less encoded directly into the bytecodes of the method as a constant. But modifying the array with #at:put: directly modifies the bytecodes in the compiled method itself. So you create a side-effect directly into your method.
Am I the only one to find this a bit dangerous and not intuitive? Is it a well-known behavior?
It is clearly ugly. Here what you see is a modification of the literal object that is stored in the compiled method literal frame. This is a well know limit of Smalltalk implementation. Having immutable literal objects may solve this problem. You get the same if you modify a string since this is a container of characters. This is like the optimization of similar literal strings that breaks the fact that 'foo' == 'foo' return false. normally yes this is false but when on the same method you get the same literal. Stef