Re: [petsc-dev] how come rank is an MPInt?

2018-02-13 Thread Smith, Barry F. 

  PetscLayOutFindOwner and PetscLayoutFindOwnerIndex() also have this 
perversity of using the wrong type. But I guess once we let the standards slip 
once we might as well willy nilly mix mix two different types as one anywhere.



> On Feb 13, 2018, at 12:10 PM, Jed Brown  wrote:
> 
> "Smith, Barry F."  writes:
> 
>> typedef struct {
>>  PetscInt rank;/* Rank of owner */
>>  PetscInt index;   /* Index of node on rank */
>> } PetscSFNode;
>> 
>> 
>> Shouldn't rank be a PetscMPIInt? and its various uses be fixed?
> 
> The intent at some point was for this to be addressable from Fortran77
> as a single array.  PetscInt is always large enough to hold an MPI int.
> We have the type in Fortran now, but is there a guarantee that alignment
> matches C alignment?  If so, we could switch.

Re: [petsc-dev] how come rank is an MPInt?

2018-02-13 Thread Jed Brown 
"Smith, Barry F."  writes:

> typedef struct {
>   PetscInt rank;/* Rank of owner */
>   PetscInt index;   /* Index of node on rank */
> } PetscSFNode;
>
>
> Shouldn't rank be a PetscMPIInt? and its various uses be fixed?

The intent at some point was for this to be addressable from Fortran77
as a single array.  PetscInt is always large enough to hold an MPI int.
We have the type in Fortran now, but is there a guarantee that alignment
matches C alignment?  If so, we could switch.