Re: refining the Synopsis APIs

Vladimir Prus wrote:

>On Tuesday 26 October 2004 08:16, Stefan Seefeld wrote:
>
>>The Abstract Semantic Graph (*ASG*) is a structure that is built from
>>the AST (or PTree), but where identifiers are cross-referenced to their
>>respective declarations (thus it's a graph, not a tree). 
>
>
>I think that "Abstract" and  "Semantic" words are all very confusing. 
>Basically, you description of ASG says it's just a set of interlinked C++ 
>objects. Semantic could mean anything, including operational semantic (how 
>i+1 is prorcessed and so on). I'd suggest to name this module just
"Objects", 
>or "Entities", to avoid further confusion.

I hoped the term would be in common use in the compiler / parser community,
and some search on google and citeseer seems to confirm this.
You are right, though, that if one doesn't consider the domain and the
context it provides, "abstract semantic" is quite difficult to coin.

>>Also, at this 
>>level of abstraction there is very little connection to the syntax, so
>>the type hierarchy can be simplified (compared to the ugly mix of type,
>>identifier, and other tokens that C++ declarations usually are made of).
>>Construction of the ASG is done by traversing the AST while looking up
>>symbols in the SymbolTable. It is done after the parsing has finished.
>
>
>Traversing AST or PTree? The difference between the too is confusing as
well. 
>If PTree is just a grouping of input tokens into blocks (class definitions,

>functions), and "Entities" is high-level classes and functions (no trace of

>syntax), then there's very little place left for any other layer.

A parse tree represents a concrete syntax, while an AST removes some of
the (now redundant) data.

>I'm still not sure that this reprsentation can be constructed after
parsing. 
>We've talked with you that during parsing you need to instantiate
templates, 
>and it's not clear if PTree representation is rich enough to handle such 
>functionality.

well, you need both, the parse tree as well as the symbol lookup table.

>In fact, now that you've sketched the diagram and explained they layers, I 
>think we need to agree what PTree does. For example:
>
>PTree -- just grouping of tokens, as little semantic information as
possible.
>Entities -- reach semantic information, including logic for instantiating 
>templates, template argument deduction and so on.
>
>Given that, it's not all clear that Entiries are constructed after parsing.

>Given an appropriate design of "Processor", the steps can be interleaved. 

yes indeed. That's something we will find out in the process :-)

FWIW, I very good summary of the process is :

http://www-adele.imag.fr/~jmfavre/ENSEIGNEMENT/TRANSPARENTS/IntroductionToSo
ftwareAnalysis/IntroductionToSoftwareAnalysis-1.pdf

That's what helped me understand the terms as I used them in my other mail.
Can we agree on this usage ?

Regards,
		Stefan