[petsc-dev] CHKERRXX() to propagate errors from C

[Barry Smith]

On Oct 19, 2011, at 2:45 PM, Matthew Knepley wrote:

> On Wed, Oct 19, 2011 at 7:39 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> #define CHKERRXX(n)            do {if (PetscUnlikely(n)) {PetscError(PETSC_COMM_SELF,__LINE__,PETSC_FUNCTION_NAME,__FILE__,__SDIR__,n,PETSC_ERROR_IN_CXX,0);}} while(0)
> 
> Inside PetscError():
> 
>   if (p == PETSC_ERROR_IN_CXX) {
>     const char *str;
>     if (eh->ctx) {
>       std::ostringstream *msg; 
>       msg = (std::ostringstream*) eh->ctx;
>       str = msg->str().c_str();
>     } else {
>       str = "Error detected in C PETSc";
>     }
>     throw PETSc::Exception(str);
>   }
> 
> This causes SEGV because eh is NULL when propagating an error from C (the caller checks a return code using CHKERRXX). This seems like a logic error, but I don't know what was intended.
> 
> I changed this to if (eh && eh->ctx) which fixes the SEGV, but I don't know if something else was intended. I only pushed this to petsc-dev, but once we decide on the "right" fix, it should be ported to petsc-3.2.
> 
> This was intended to get PETSc-like error checking from C++-style functions that throw exceptions. This was mostly for code I
> wrote I think. Your fix is fine.
> 
> After several years of this, I am now of the opinion that exceptions suck all the way, and I should rewrite all my C++ to use
> return values. The stack does not get wiped out, you can still catch specific errors, and propagate ones you do not catch.
> Does anyone have any serious defense of exceptions?

    The properly manage the clean up of partially created objects etc don't they, which our C approach cannot handle. Not that I'm advocating for using them or not.

   Barry

> 
> I think we should retain the code for interoperability with C++-heads.
> 
>    Matt
> 
> -- 
> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
> -- Norbert Wiener