2017-04-25 19:08 GMT+02:00 Hilaire <hilaire@drgeo.eu>:
Can you produce tests exposing these fails?
|set| set := Set new. Smalltalk allClassesAndTraits select:[:tc | tc traitComposition notEmpty] thenDo:[:b | |tc nonLocal| tc:= b traits. nonLocal := b localMethods select:[:m | tc anySatisfy:[:t | t methods contains:[:tm| ((tm selector = m selector and:[tm bytecode = m bytecode]) and:[tm literals allButLast = m literals allButLast])]]]. nonLocal notEmpty ifTrue:[set addAll:( nonLocal)]]. MessageBrowser browse:set
Hilaire
Le 25/04/2017 à 10:47, Nicolai Hess a écrit :
Keep in mind. 1. the way new trait methods are handled is still wrong if non-local methods (methods from a trait) aren't added to the rpackage when creating the user of this trait, why are new trait methods "flattened" into this package 2. we already have some "corrupted" (flattened) packages in the image. (some methods of users of Traits are copied from the trait). For example, traits using the trait GLMBrickExtensionTrait or test case traits using other traits (TRemoveForMultiplenessTest / TRemoveTest, TDictionaryCopyingTest/TCloneTest)
-- Dr. Geo http://drgeo.eu