Incorrect handling of NULL count for nc_get/put_vars*() for netCDF-4 files

[edhartnett]

In nc4_get_vars() (in libhdf5/nc4hdf5.c) we have:

 /* Open this dataset if necessary, also checking for a weird case:
   * a non-coordinate (and non-scalar) variable that has the same
   * name as a dimension. */
  if (var->hdf5_name && strlen(var->hdf5_name) >= strlen(NON_COORD_PREPEND) &&
      strncmp(var->hdf5_name, NON_COORD_PREPEND, strlen(NON_COORD_PREPEND)) == 0 &&
      var->ndims)
     name_to_use = var->hdf5_name;
  else
     name_to_use = var->hdr.name;

Test coverage shows we are not testing the first half of this if statement. We need to test the case of a dataset which has the same name as a dimension, but is not a coordinate variable. I will add this test.

[edhartnett]

After some investigation, it turns out this code was never executed anyway. In all cases var->hdf4_datasetid was already set by the time the if statement is reached. I have replaced it with an assert.

However, this lead to another discovery. We also had this code in both put and get vars:

      start[i] = (startp == NULL ? 0 : startp[i]);
      count[i] = (countp == NULL ? 1 : countp[i]);
      stride[i] = (stridep == NULL ? 1 : stridep[i]);

This is correct for the start and stride, but not the count. If countp is NULL, then count[i] needs to be set to the full extent of the dim, not to 1.

This was also untested by current tests. I have fixed the code and added a test. I will put a PR up in a while with the fixes.