Re: [gomp] Parser for OpenMP directives

[Richard Henderson <rth at redhat dot com>]

On Fri, Mar 25, 2005 at 11:34:53AM +0300, Dmitry Kurochkin wrote:
> +	  if (strncmp ("omp", p, 3))
> +	    break;
> +	  p += 3;
> +	  if (!ISSPACE (*p))
> +	    break;
> +	  ++p;
> +	  while (*p != '\n' && ISSPACE (*p))
> +	    ++p;
> +	  /* This is an OpenMP pragma. Getting it's kind now. */

This is going to have to be done differently.  Note that section 2.1
specifies that preprocessing tokens following "#pragma omp" are 
subject to macro replacement.  Thus

#define foo barrier
#pragma omp foo
  
is legal.  AFAIK, the interface currently provided by libcpp contains
the raw text line.

> @@ -553,6 +701,10 @@
>        if (c_dialect_objc ())
>  	return true;
>        return false;
> +    case CPP_PRAGMA:
> +      if (token->omp_kind == PRAGMA_OMP_THREADPRIVATE)
> +	return true;
> +      return false;

First, please use the -p option to (cvs) diff so that its easier to
see the context of the patch.

Second, as far as I can see, the threadprivate directive should be
handled exactly like we currently handle pragma weak.  I don't see
what this pragma has to do with starting a declspec.  Similarly wrt
c_parser_declaration_or_fndef.

> +	    case PRAGMA_OMP_CLAUSE_IF:
> +	      c_parser_pragma_omp_clause_if (the_parser);
> +	      break;

You'll need to diagnose clases that aren't allowed to be repeated.
Perhaps that's easier to do once you have an actual data structure
that you can record things in...


r~