Good catch Mariano -- I have run your test on Squeak and fixed a
couple of instances there too.
On Wed, May 30, 2012 at 7:05 AM, Mariano Martinez Peck
<
[hidden email]> wrote:
> Hi guys. I found a problem while working with FuelPreview where we store
> objects of the graph in a Set. While doing a collect, there was an error
> because an implementation of #= was assuming that the argument was of a
> certain shape. Of course, that should answer false instead. So I wrote this
> test to see all existing wrong implementations:
>
> testAllClassesImplementSafeEqualsMethod
> "This tests that all classes in the system that implements #= do it in a way
> that they don't throw error when passing as an argument something different
> from expected. The correct behavior is that #= answers false."
> | wrongClasses |
> wrongClasses := IdentityDictionary new.
> ((SystemNavigation default allImplementorsOf: #=)
> collect: [:each | each methodClass])
> do: [:each |
> | instance |
> "Some classes like CompiledMethod override basicNew to throw
> an error.
> In any case, the comparison will be false, so no problem"
> [instance := each basicNew] on: Error do: [].
> [instance = Object new]
> on: Error
> do: [: err | wrongClasses at: each put: err ]].
> self assert: wrongClasses isEmpty
>
>
> The result is: {Magnitude. WideCharacterSet. KMKeymap. MCMockDefinition.
> ScaledDecimal. MCSnapshot}
>
> So, my I think we should change #= in those classes by adding:
>
> self == aKeymap
> ifTrue: [ ^ true ].
> self class = aKeymap class
> ifFalse: [ ^ false ].
>
> at the beginning of the #=
>
> what do you think? if agree, I open an issue.
>
>
> --
> Mariano
>
http://marianopeck.wordpress.com>