================
Comment at: utils/TableGen/ClangAttrEmitter.cpp:811
@@ +810,3 @@
+ if (Variety == "CXX11")
+ OS << " if (LK == 0 || LK == 1 || LK == 3) {\n";
+
----------------
Richard Smith wrote:
> Each attribute should only be printed at one location.
This seems not possible using current approach since an attribute could
legitimately appear at several source locations, with each source location
being encoded as a LocationKind enum. For example, C++11 attribute could appear
at start of a statement, or appear between class-head and class-name. So unless
we use a single enum value to encode both positions, we have to check both enum
value and print appropriately.
================
Comment at: lib/AST/DeclPrinter.cpp:365
@@ -364,3 +364,3 @@
}
- prettyPrintAttributes(D);
+ prettyPrintAttributes(D, Attr::LK_AfterDeclaratorID);
}
----------------
Richard Smith wrote:
> This one looks really odd. We should be printing attributes after the
> enum/enum class/enum struct keywords (the closest match is
> LK_AfterDeclaratorID for those, even though they're before the
> declarator-id). I don't think there are *any* attributes which appertain to
> an enum, and go here, after the enum body.
Indeed.
================
Comment at: lib/AST/DeclPrinter.cpp:338
@@ -337,3 +337,3 @@
D->getUnderlyingType().print(Out, Policy, D->getName());
- prettyPrintAttributes(D);
+ prettyPrintAttributes(D, Attr::LK_AfterDeclaratorID);
}
----------------
Richard Smith wrote:
> Isn't this LK_EndOfDecl?
Technically yes. I use 'LK_AfterDeclaratorID' because that is the only enum
value that allows print all kinds of attributes. 'LK_EndOfDecl' will not print
C++11 attribute, since in the cases of a function declaration, a C++11
attribute at end of declaration appertain to function types. I need to think
about this and use a better approach to deal with such cases.
================
Comment at: lib/AST/DeclPrinter.cpp:391
@@ -390,2 +390,3 @@
void DeclPrinter::VisitFunctionDecl(FunctionDecl *D) {
+ prettyPrintAttributes(D, Attr::LK_StartOfDecl);
CXXConstructorDecl *CDecl = dyn_cast<CXXConstructorDecl>(D);
----------------
Richard Smith wrote:
> Can we generalize this to print LK_StartOfDecl attributes at the start of
> *every* declaration?
Sounds good.
http://llvm-reviews.chandlerc.com/D395
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
