nc_delete_mp() uses NC3_DATA macro, but is not a dispatch table function

[edhartnett]

Here's something I ran into while doing the HDF5/libsrc4 conversion. I also have the easy fix.

In libsrc/nc3internal.c we have:

/**************************************************/
int
nc_delete_mp(const char * path, int basepe)
{
	NC *nc;
	NC3_INFO* nc3;
	int status;
	int ncid;
	size_t chunk = 512;

	status = nc_open(path,NC_NOWRITE,&ncid);
        if(status) return status;

	status = NC_check_id(ncid,&nc);
        if(status) return status;
	nc3 = NC3_DATA(nc);

	nc3->chunk = chunk;

#if defined(LOCKNUMREC) /* && _CRAYMPP */
	if (status = NC_init_pe(nc3, basepe)) {
		return status;
	}
#else
	/*
	 * !_CRAYMPP, only pe 0 is valid
	 */
	if(basepe != 0)
		return NC_EINVAL;
#endif

	(void) nc_close(ncid);
	if(unlink(path) == -1) {
	    return NC_EIO;	/* No more specific error code is appropriate */
	}
	return NC_NOERR;
}

Note that we are applying NC3_DATA macro, and then modifying the "chunk" field within it.

Problem is, this is not a dispatch function. ;-) So it is also called by nc_test when testing netCDF-4 files. The setting of nc3->chunk changes some random part of the netcdf-4 structure, which happens to have no effect (or didn't, until I did the HDF5/libsrc4 conversion, and encountered this error).

The fix is simple - there is no reason to be messing with nc3->chunk, as the file is about to be deleted. I simply removed the two lines of code.

	nc3 = NC3_DATA(nc);
	nc3->chunk = chunk;

The fix for this will be part of my PR for #856. I wanted to high-light this change since it is in the classic code, outside the libsrc4 directory where all the rest of the #856 changes will be found.

[edhartnett]

This fix made it into one of the aggregate PRs and is now in the code.