[cfe-dev] Should we build semantically invalid nodes?

[steve naroff]

On Oct 25, 2008, at 10:28 AM, Argiris Kirtzidis wrote:

> Hey Steve, breaking up Sema is a ridiculous awesome idea!
>

Sounds like I struck a chord:-)

> Here's a thought I'd like to throw around..
> Could there be something like "composable Actions" ? The ASTBuilder  
> would build the AST while Sema would do semantic checks and reject  
> invalid nodes.
> This will cleanly separate the semantic checks from the AST building  
> and, as you said, will make the code more maintainable.
>

Composable Actions is an idea I've always wanted to experiment with  
(but haven't had a compelling reason to do so). If what you are  
thinking of sounds compelling, then it might be a good place to  
experiment.

At this point, it's unclear if the semantics checks can always be  
separated from the AST.

Are you certain the current AST's are appropriate for what you  
envision doing?

snaroff

> -Argiris
>
> steve naroff wrote:
>> Hey Argiris,
>>
>> Since clang supports multiple "Action" modules, why not have an  
>> ASTBuilder class?
>>
>> My intuition says simply passing in a flag to Sema won't be rich  
>> enough. Over time, you would probably need several flags to get  
>> Sema to do what you actually want. It just doesn't feel right to me.
>>
>> On the other hand, AST's produced by an ASTBuilder class could be  
>> segregated and managed differently. It gives you much more  
>> flexibility to meet the needs of a more "fuzzy" environment (like  
>> an IDE), including replacing some of the AST nodes (if the one's  
>> were using in Sema aren't ideal). For example, the ASTBuilder class  
>> might choose to have "real" (i.e. non-unique types) types  
>> (something we've discussed before). Having the ability to get the  
>> actual source location for type specifiers seems like a "must have"  
>> feature for non-compiler clients.
>>
>> I believe my suggestion is fairly easy to implement for Stmts/ 
>> Exprs. It is more difficult to implement for Decls/Types (since  
>> Sema::ActOnDeclarator() does some "heavy lifting" to transform a  
>> Declarator into a Decl/Type). If this were the only obstacle, we  
>> could refactor the code into a Declarator library that both Action  
>> modules could share. Since Declarators in C/C++ are very complex,  
>> adding a layer might actually improve the maintenance of the code.
>>
>> I think your desire for more fuzzy AST's is spot on. My only  
>> concern is identifying clients for them and choosing an  
>> implementation strategy that doesn't disrupt our most central/ 
>> complex Action module (i.e. Sema). We knew this day would come,  
>> which is why the Action interface doesn't force you to use a  
>> particular data structure (or any at all).
>>
>> Thoughts?
>>
>> snaroff
>>
>> On Oct 25, 2008, at 4:21 AM, Argiris Kirtzidis wrote:
>>
>>> Here's an alternative idea:
>>> How about we pass a flag at Sema to indicate that Sema should just  
>>> build
>>> the AST without semantic checks ?
>>>
>>> The meaning of this flag is not that it guarantees that  
>>> *absolutely no*
>>> semantic checks are done, just that the focus is less on semantics  
>>> and
>>> more on getting a complete AST.
>>> The Action methods could be modified gradually to check this flag,  
>>> eg.
>>> on Sebastian's patch, if this flag is on, Sema::ActOnCXXCasts will  
>>> skip
>>> the CheckXXXCast call and go to creating the CXXCastExpr directly.
>>>
>>> This addresses Doug's concerns about allowing invalid nodes while  
>>> also
>>> enabling clients that only care about a full AST.
>>>
>>> Any thoughts ?
>>>
>>> -Argiris
>>> _______________________________________________
>>> cfe-dev mailing list
>>> cfe-dev at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>
>>