Smalltalk › Pharo › Pharo Smalltalk Developers

Question on AST

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

3 messages Options

Gustavo Santos

Question on AST

Hello,

I'm working on code manipulation on Pharo using the AST.

Recently, I ported my code to Pharo 5 and some of my tests are not running anymore.

Specifically, if I have a method as follows:

method

| variable |

variable := 'String'.

In Pharo 4, when I get the tree, I have a RBTemporaryNode representing the declaration of the variable, and a RBVariableNode for the assignment at the third line.

In Pharo 5, I have two RBVariableNode's for both cases. And when I call #isTemp I get a MNU because the return of #binding is nil.

How can I now differentiate these two nodes in the example? Is this a bug or a feature? :-P

Cheers,

Gustavo Santos

Marcus Denker-4

Re: Question on AST

> On 24 Sep 2015, at 14:40, Gustavo Santos <[hidden email]> wrote:
>
> Hello,
>
> I'm working on code manipulation on Pharo using the AST.
> Recently, I ported my code to Pharo 5 and some of my tests are not running anymore.
> Specifically, if I have a method as follows:
>
> method
> | variable |
> variable := 'String'.
>
> In Pharo 4, when I get the tree, I have a RBTemporaryNode representing the declaration of the variable, and a RBVariableNode for the assignment at the third line.
>
> In Pharo 5, I have two RBVariableNode's for both cases. And when I call #isTemp I get a MNU because the return of #binding is nil.
>
> How can I now differentiate these two nodes in the example? Is this a bug or a feature? :-P

You need to do semantic analysis first. The idea that the parser does that is just wrong, it was a mistake to add that.
So in Pharo5 I removed the hack in the parser.

-> Parser parses, it knows nothing about semantics
-> Semantic analysis looks at variables and decides what they mean.

The #binding is set by the semantic analysis.
(and, as it was nice, actually specialises the classes of variables)

Right now you have to call #doSemanticAnalysis on the MethodNode. I am planning to add some lazy init so this is done automaticaly
when accessing #binding of a variable…

Marcus

Gustavo Santos

Re: Question on AST

Thanks, it works.

And I understand the process now

On Thu, Sep 24, 2015 at 2:47 PM, Marcus Denker <[hidden email]> wrote:

> On 24 Sep 2015, at 14:40, Gustavo Santos <[hidden email]> wrote:
>
> Hello,
>
> I'm working on code manipulation on Pharo using the AST.
> Recently, I ported my code to Pharo 5 and some of my tests are not running anymore.
> Specifically, if I have a method as follows:
>
> method
> | variable |
> variable := 'String'.
>
> In Pharo 4, when I get the tree, I have a RBTemporaryNode representing the declaration of the variable, and a RBVariableNode for the assignment at the third line.
>
> In Pharo 5, I have two RBVariableNode's for both cases. And when I call #isTemp I get a MNU because the return of #binding is nil.
>
> How can I now differentiate these two nodes in the example? Is this a bug or a feature? :-P

You need to do semantic analysis first. The idea that the parser does that is just wrong, it was a mistake to add that.
So in Pharo5 I removed the hack in the parser.

-> Parser parses, it knows nothing about semantics
-> Semantic analysis looks at variables and decides what they mean.

The #binding is set by the semantic analysis.
(and, as it was nice, actually specialises the classes of variables)

Right now you have to call #doSemanticAnalysis on the MethodNode. I am planning to add some lazy init so this is done automaticaly
when accessing #binding of a variable…

Marcus

Gustavo Santos