<div>I see that char** for opaque data has changed to void**. This choice essentially always requires an explicit cast in user code (because they have something real behind the pointer).</div><div><br></div><div>There is a precedent for using plain void* in cases like this. For example, see the MPI Forum's change of</div>
<div><br></div><div>int MPI_Attr_get(MPI_Comm comm, int keyval,void **attribute_val, int *flag ) /* MPI-1.0 */</div><div><br></div><div>to</div><div><br></div><div><div>int MPI_Attr_get(MPI_Comm comm, int keyval,void *attribute_val, int *flag ) /* MPI-1.1 and later */</div>
</div><div><br></div><div>Note that this change is backward-compatible. I would encourage the ITAPS community to make the same choice.</div><div><br></div><div><br></div><div>I don't know where these discussions take place. In principle I'm subscribed to the itaps-users list, but I have never seen anything posted there. It would be nice if there was at least a notification mechanism for API changes, preferably with an open discussion period. A much-simplified version of the MPI Forum's process would be great.</div>