On Oct 15, 2011, at 8:21 PM, John Freeman wrote:
>> + /// Declaration for the closure type.
>> + CXXRecordDecl *ClosureType;
>>
>> This is just getType()->getAsCXXRecordDecl(), no?
>
> To my knowledge, yes it is. I added the ClosureType member simply for
> caching. I saw you removed it. The same could be done for the Function member
> as well. Is the general policy to keep the AST classes as small as possible?
Yes, absolutely. Memory footprint is extremely important when compiling C++
code, so we try to keep our data structures small. We'll use extra storage if
it provides a significant performance improvement, of course, but that's not
likely the case here.
>> + /// The temporary object resulting from evaluating the lambda expression,
>> + /// represented by a call to the constructor of the closure type.
>> + CXXConstructExpr *ClosureObject;
>>
>> This makes me think that LambdaExpr should just be a subclass of
>> CXXConstructExpr, because there's a lot of behavior we get "for free" for
>> lambdas if it derives from CXXConstructExpr (such as proper handling of the
>> temporary). It would require some twisting-around of the logic for building
>> lambda expressions (in particular, the LambdaExpr wouldn't be built until
>> the whole body had been parsed), but I think the result would be better.
>
> How should we store the bits and pieces up until the body has been parsed?
> Should there be a separate data structure just to hold them that gets passed
> between Sema functions, or should there be members within Sema for things
> like CurLambdaIntroducer and CurLambdaMethod, or something else?
Check out BlockScopeInfo; you'll want something like that, to keep the stack of
active lambdas available. It'll become extremely important once you start
handling implicit captures.
- Doug
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
