[cfe-commits] r167363 - in /cfe/trunk: include/clang/Parse/Parser.h lib/Parse/ParseStmt.cpp

[Chandler Carruth]

On Sat, Nov 3, 2012 at 3:29 PM, Lang Hames <lhames at gmail.com> wrote:
> Author: lhames
> Date: Sat Nov  3 17:29:05 2012
> New Revision: 167363
>
> URL: http://llvm.org/viewvc/llvm-project?rev=167363&view=rev
> Log:
> Support interleaving of other pragmas with FP_CONTRACT at the beginning of a
> compound statement.
>
> Modified:
>     cfe/trunk/include/clang/Parse/Parser.h
>     cfe/trunk/lib/Parse/ParseStmt.cpp

Test case please!

>
> Modified: cfe/trunk/include/clang/Parse/Parser.h
> URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Parse/Parser.h?rev=167363&r1=167362&r2=167363&view=diff
> ==============================================================================
> --- cfe/trunk/include/clang/Parse/Parser.h (original)
> +++ cfe/trunk/include/clang/Parse/Parser.h Sat Nov  3 17:29:05 2012
> @@ -1498,6 +1498,7 @@
>    StmtResult ParseCompoundStatement(bool isStmtExpr = false);
>    StmtResult ParseCompoundStatement(bool isStmtExpr,
>                                      unsigned ScopeFlags);
> +  void ParseCompoundStatementLeadingPragmas();
>    StmtResult ParseCompoundStatementBody(bool isStmtExpr = false);
>    bool ParseParenExprOrCondition(ExprResult &ExprResult,
>                                   Decl *&DeclResult,
>
> Modified: cfe/trunk/lib/Parse/ParseStmt.cpp
> URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Parse/ParseStmt.cpp?rev=167363&r1=167362&r2=167363&view=diff
> ==============================================================================
> --- cfe/trunk/lib/Parse/ParseStmt.cpp (original)
> +++ cfe/trunk/lib/Parse/ParseStmt.cpp Sat Nov  3 17:29:05 2012
> @@ -706,6 +706,48 @@
>    return ParseCompoundStatementBody(isStmtExpr);
>  }
>
> +/// Parse any pragmas at the start of the compound expression. We handle these
> +/// separately since some pragmas (FP_CONTRACT) must appear before any C
> +/// statement in the compound, but may be intermingled with other pragmas.
> +void Parser::ParseCompoundStatementLeadingPragmas() {
> +  bool checkForPragmas = true;
> +  while (checkForPragmas) {
> +    switch (Tok.getKind()) {
> +    case tok::annot_pragma_vis:
> +      HandlePragmaVisibility();
> +      break;
> +    case tok::annot_pragma_pack:
> +      HandlePragmaPack();
> +      break;
> +    case tok::annot_pragma_msstruct:
> +      HandlePragmaMSStruct();
> +      break;
> +    case tok::annot_pragma_align:
> +      HandlePragmaAlign();
> +      break;
> +    case tok::annot_pragma_weak:
> +      HandlePragmaWeak();
> +      break;
> +    case tok::annot_pragma_weakalias:
> +      HandlePragmaWeakAlias();
> +      break;
> +    case tok::annot_pragma_redefine_extname:
> +      HandlePragmaRedefineExtname();
> +      break;
> +    case tok::annot_pragma_opencl_extension:
> +      HandlePragmaOpenCLExtension();
> +      break;
> +    case tok::annot_pragma_fp_contract:
> +      HandlePragmaFPContract();
> +      break;
> +    default:
> +      checkForPragmas = false;
> +      break;
> +    }
> +  }

This seems rather brittle. Is there anything that will prevent it from
becoming a bug every time someone adds a new pragma? Is there any way
to factor things so that we get a compiler error or something here
that indicates we might need to handle every pragma?

What I'm thinking is a function that converts the pragma tokens to a
pragma enum that we have some confidence people will update. Then we
can have a covering enum switch here (and else where).

-Chandler