On Aug 14, 2011, at 1:50 PM, John Freeman wrote:
> On 8/12/11 5:26 PM, Douglas Gregor wrote:
>> +/// LambdaDecl - Represents a C++0x lambda expression.
>> +class LambdaDecl : public Decl, public DeclContext {
>>
>> It would help if the comment here showed an example lambda expression, and
>> pointed out the main pieces (capture list, body, etc.).
>>
>> Please add more source-location information to LambdaDecl, so we have at
>> least enough source-location information to implement getSourceRange() for
>> LambdaDecl, covering the entire declaration. We'll need this source
>> information for diagnostics and indexing.
>>
>> +/// LambdaExpr - Adaptor class for mixing a LambdaDecl with expressions.
>> +class LambdaExpr : public Expr {
>> + LambdaDecl *Lambda;
>> +
>>
>> It's really not just an "adaptor", is it? It's the actual lambda expression.
>> Please expand the comment with an example lambda expression. The commenting
>> here is actually more important than for LambdaDecl, since lambdas are an
>> expression in the language.
>
>
> Yeah, I'm still trying to figure out the LambdaDecl/LambdaExpr distinction
> myself. I based it off of the BlockDecl/BlockExpr design. Here is my current
> theory: LambdaExpr should be the AST node returned from
> ParseLambdaExpression. It holds a LambdaDecl that contains all the
> declarations related to the lambda expression: Captures, CXXMethodDecl,
> CXXRecordDecl. Location information for the whole expression will be stored
> in LambdaExpr. Components will have their location information stored in
> their respective Captures, ParmVarDecls, Stmt, etc.
>
> I'm starting to wonder, though, if we even need a LambdaDecl. Why not just
> put everything in the LambdaExpr? I have not come across a need for a Decl or
> DeclContext for lambdas, but I might have just not hit it yet.
Interesting. The reasons for BlockDecl that I can think of really don't apply
to lambdas, because lambdas already have a Decl and DeclContext associated with
them (the generated CXXMethodDecl). It would certainly feel simpler to just
have a LambdaExpr.
- Doug
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
