Merge branch 'knepley/remove-petscsyshh'

* knepley/remove-petscsyshh:
  Sys: Added missing header - The C++ Saga continues
  Sys: God let this be the last checkin
  Sys: Wherein Matt gives up on C++, and eradicates it from error handling
  Sys: Removed ref to PETSc::Exception
  Sys: Fixed missing include
  Sys: Removed petscsys.hh

include/petscsys.h

@@ -2230,11 +2230,6 @@ PETSC_STATIC_INLINE PetscErrorCode PetscMPIIntCast(PetscInt a,PetscMPIInt *b)
 #  define PETSC_MAX_PATH_LEN     4096
 #endif
 
-/* Special support for C++ */
-#if defined(PETSC_CLANGUAGE_CXX) && defined(__cplusplus)
-#include <petscsys.hh>
-#endif
-
 /*MC
 
     UsingFortran - Fortran can be used with PETSc in four distinct approaches

src/sys/error/err.c

@@ -297,7 +297,13 @@ PetscErrorCode  PetscErrorMessage(int errnum,const char *text[],char **specific)
  *
  * in "C++ Coding Standards" by Sutter and Alexandrescu. (This accounts for part of the ongoing C++ binary interface
  * instability.) Having PETSc raise errors as C++ exceptions was probably misguided and should eventually be removed.
+ *
+ * Here is the problem: You have a C++ function call a PETSc function, and you would like to maintain the error message
+ * and stack information from the PETSc error. You could make everyone write exactly this code in their C++, but that
+ * seems crazy to me.
  */
+#include <sstream>
+#include <stdexcept>
 static void PetscCxxErrorThrow() {
   const char *str;
   if (eh && eh->ctx) {
@@ -306,7 +312,7 @@ static void PetscCxxErrorThrow() {
     str = msg->str().c_str();
   } else str = "Error detected in C PETSc";
 
-  throw PETSc::Exception(str);
+  throw std::runtime_error(str);
 }
 #endif