So is the test BS? Should we modify it to expect the cascade node instead?

On Dec 21, 2010, at 3:07 AM, Frank Shearar <frank.shearar at angband.za.org> wrote:

> One of the failing Decompiler tests is when we check MethodPragmaTest>>testCompileCharacter:
> 
> testCompileCharacter
>    self assertPragma: 'foo: $a' givesKeyword: #foo: arguments: #( $a ).
>    self assertPragma: 'foo: $ ' givesKeyword: #foo: arguments: #( $  ).
> 
> As bytecodes, this method looks like this:
> 
> 37 <70> self
> 38 <21> pushConstant: 'foo: $a'
> 39 <22> pushConstant: #foo:
> 40 <23> pushConstant: #($a)
> 41 <83 60> send: assertPragma:givesKeyword:arguments:
> 43 <87> pop
> 44 <70> self
> 45 <24> pushConstant: 'foo: $ '
> 46 <22> pushConstant: #foo:
> 47 <25> pushConstant: {Character space}
> 48 <83 60> send: assertPragma:givesKeyword:arguments:
> 50 <87> pop
> 51 <78> returnSelf
> 
> And when decompiled, we have:
> 
> testCompileCharacter
>    self
>        assertPragma: 'foo: $a'
>        givesKeyword: #foo:
>        arguments: #($a ).
>    self
>        assertPragma: 'foo: $ '
>        givesKeyword: #foo:
>        arguments: ((Array new: 1) at: 1 put: Character space; yourself)
> 
> Notice that the literal #( $  ) has expanded into an Array declaration etc.
> 
> When we compile this decompiled string, we end up with the rather different
> 
>    self
>    pushConstant: ''foo: $a''
>    pushConstant: #foo:
>    pushConstant: #($a)
>    send: #assertPragma:givesKeyword:arguments: (3 args)
>    pop
>    self
>    pushConstant: ''foo: $ ''
>    pushConstant: #foo:
>    pushLit: Array
>    pushConstant: 1
>    send: #new: (1 arg)
>    dup
>    pushConstant: 1
>    pushLit: Character
>    send: #space (0 args)
>    send: #at:put: (2 args)
>    pop
>    send: #yourself (0 args)
>    send: #assertPragma:givesKeyword:arguments: (3 args)
>    pop
>    returnSelf
> 
> The decompiler turns that pushConstant: into a whole bunch of stuff because it interprets #( $  ) as (Array new: 1) at: 1 put: Character space; yourself.
> 
> (The same thing happens for SyntaxMorph>>replaceSel:menuItem: and SyntaxMorph>>replaceKeyWord:menuItem. It looks like $  is being interpreted as Character space, and then decompiled to pushLit: Character send: #space.)
> 
> That's all fine, but what happens is that when you compile the decompiled method (and then decompile _that_ method), you can't have the newly-decompiled method match the original decompiled method. The parse tree are different: where the original has a LiteralNode, the new version has a CascadeNode.
> 
> frank
>