This PR closes issue #562.

As mentioned in the initial comment of this PR, if both missing_value and default_value are specified, the field will be initialized with default_value. If some other action should occur, I'm happy to update this PR.

@MiCurry This looks good so far. I think the same logic as you've implemented for vars is also needed for var_arrays.

@MiCurry This looks good so far. I think the same logic as you've implemented for vars is also needed for var_arrays.

Ah, I forgot about var_arrays. Thanks @mgduda.

I just pushed an update that contains the same changes for var_arrays. I also moved the change for vars to below the call to get_field_information. This is consistent with the var_array change and seems a bit more organized/logical.

All, I have updated the commit and PR message for this PR to be more explicit as to what occurs when missing_value and default_value are specified. From our telecom on 5/26 I believe this implementation is correct, but please let me know if it isn't!

@MiCurry Is it possible that your last force-push of 6fc82f0 removed the changes for var_arrays?

@MiCurry Is it possible that your last force-push of 6fc82f0 removed the changes for var_arrays?

@mgduda You are correct. The last force push just re-added the var array changes!

As part of testing this PR, I tried to assign a missing value to one var in a var_array:

                <var name="testMissingVar" type="real" dimensions="nVertLevels nCells Time" units="unitless" missing_value="9.969209968386869e36"
                        description="A variable that is zero where valid and missing_value where not"
                />
                <var_array name="testMissingArray" type="real" dimensions="nVertLevels nCells Time">
                        <var name="testMissingArrayVar" array_group="testMissingGroup" units="unitless" missing_value="9.969209968386869e36"
                                description="A variable that is one where valid and missing_value where not"
                        />
                </var_array>

This didn't work as expected -- when I look in inc/structs_and_variables.inc, only testMissingVar has the missing_value attribute. I realize this goes beyond this PR but @mgduda and @MiCurry, is that the expected behavior? Are default_value and missing_value only available at the var_array level, not at the individual var level for a var_array?

Thanks for this great work, @MiCurry! I tested using a branch (which I will keep around only until this PR gets merged into ocean/develop): https://github.com/xylar/MPAS-Model/pull/new/ocean/test_missing_value

I found that everything worked as expected if I assigned a real to missing_value. But If I assigned FILLVAL, I got a compile-time error:

../inc/structs_and_variables.inc:8479:39:

       r3Ptr(1) % defaultValue = FILLVAL ! defaultValue taking specified missing_value
                                       1
Error: Symbol 'fillval' at (1) has no IMPLICIT type

This is because FILLVAL isn't getting detected and replaced with the appropriate MPAS_*_FILLVAL in default_value in the same way as it is in missing_value. See:
https://github.com/MiCurry/MPAS-Model/blob/registry/missing_value/src/tools/registry/gen_inc.c#L451-L472
To fix this, these lines:
https://github.com/MiCurry/MPAS-Model/blob/registry/missing_value/src/tools/registry/gen_inc.c#L428-L449
would need to be something like:

	if (strcmp(vartype, "real") == 0){
		(*type) = REAL;
		if(!varval){
			snprintf(default_value, 1024, "0.0");
		} else if(strcmp(varmissval, "FILLVAL") == 0) {
			snprintf(missing_value, 1024, "MPAS_REAL_FILLVAL");
		} else {
			snprintf(default_value, 1024, "%s", varval);
		}
	} else if (strcmp(vartype, "integer") == 0){
		(*type) = INTEGER;
		if(!varval){
			snprintf(default_value, 1024, "0");
		} else if(strcmp(varmissval, "FILLVAL") == 0) {
			snprintf(missing_value, 1024, "MPAS_INT_FILLVAL");
		} else {
			snprintf(default_value, 1024, "%s", varval);
		}
	} else if (strcmp(vartype, "text") == 0){
		(*type) = CHARACTER;
		if(!varval){
			snprintf(default_value, 1024, "''");
		} else if(strcmp(varmissval, "FILLVAL") == 0) {
			snprintf(missing_value, 1024, "MPAS_CHAR_FILLVAL");
		} else {
			snprintf(default_value, 1024, "%s", varval);
		}
	}

(Boy, would this code be more elegant and readable in python...)

I realize this goes beyond this PR but @mgduda and @MiCurry, is that the expected behavior? Are default_value and missing_value only available at the var_array level, not at the individual var level for a var_array?

That was the expected behavior of my implementation and I think the current behavior of master. default_value and missing_value are only available at the var_array level. Here is a snippet of the atmosphere's core scalar var_array (with a few members removed) from core_atmosphere/inc/structs_and_variables.inc where the defaultValue and missingValue is set for a var_array.

      r3Ptr(1) % dimNames(1) = 'num_scalars'
      r3Ptr(1) % dimNames(2) = 'nVertLevels'
      r3Ptr(1) % dimNames(3) = 'nCells'

      r3Ptr(1) % defaultValue = 0.0
      allocate(r3Ptr(1) % attLists(size(r3Ptr(1) % constituentNames, dim=1)))
      do index_counter = 1, size(r3Ptr(1) % constituentNames, dim=1)
         allocate(r3Ptr(1) % attLists(index_counter) % attList)
      end do
      if (associated(newSubPool)) then
         call mpas_pool_get_dimension(newSubPool, 'index_qv', const_index)
      end if
      if (const_index > 0) then
         call mpas_add_att(r3Ptr(1) % attLists(const_index) % attList, 'long_name', 'Water vapor mixing ratio')
         call mpas_add_att(r3Ptr(1) % attLists(const_index) % attList, 'units', 'kg kg^{-1}')
         r3Ptr(1) % missingValue = MPAS_REAL_FILLVAL
         r3Ptr(1) % constituentNames(const_index) = 'qv'
      end if
      if (associated(newSubPool)) then
         call mpas_pool_get_dimension(newSubPool, 'index_qc', const_index)
      end if
      if (const_index > 0) then
         call mpas_add_att(r3Ptr(1) % attLists(const_index) % attList, 'long_name', 'Cloud water mixing ratio')
         call mpas_add_att(r3Ptr(1) % attLists(const_index) % attList, 'units', 'kg kg^{-1}')
         r3Ptr(1) % missingValue = MPAS_REAL_FILLVAL
         r3Ptr(1) % constituentNames(const_index) = 'qc'
      end if
      if (associated(newSubPool)) then
      r3Ptr(1) % block => block

I believe r3Ptr(1) % defaultValue and r3Ptr(1) % missingValue effect all members of the array. In the code snippet above there is a single assignment for r3ptr(1) % defaultValue, but one for each member for the r3Ptr(1) % missingValue assignment. I think the assignments for missingValue are all identical statements, so even if the values assigned to them would be different, the entire array would get the last one that was executed (I think).

I found that everything worked as expected if I assigned a real to missing_value. But If I assigned FILLVAL

@xylar, ah I did not realize one could use FILLVAL to set missing_value and default_value. Thanks for pointing that out, and thanks for giving the code to fix it. I'll get to adding that!

thanks for giving the code to fix it.

You're welcome, with the caveat that I didn't test that code at all so it could be completely useless.

I think the fix may be as simple as changing vararrmissingval to missing_value and similarly for the code in parse_var in the current PR:

@@ -1066,7 +1066,7 @@ int parse_var_array(FILE *fd, ezxml_t registry, ezxml_t superStruct, ezxml_t var
        // If a default_value is not specified, but a missing_value is, then set the
        // default_value (defaultValue) to missing_value
        if(!vararrdefaultval && vararrmissingval) {
-               snprintf(default_value, 1024, "%s ! defaultValue taking specified missing_value", vararrmissingval);
+               snprintf(default_value, 1024, "%s ! defaultValue taking specified missing_value", missing_value);
        }

I think the fix may be as simple as changing vararrmissingval to missing_value and similarly for the code in parse_var in the current PR

@mgduda, ah, you're exactly right! Thanks for catching this.

My suggestion is wrong because default_value would be the expected thing coming out of get_field_information() but then it would get replaced in the new lines @MiCurry has added.

Thanks @MiCurry. I'll test this again later tonight.

Okay @mgduda, @xylar and everyone else, I have just pushed changes that changes vararrmissingval to missing_value. So now adding FILLVAL to missing_value now successfully compiles.

@matthewhoffman is my ocean test sufficient for approval, or would you like to test as well?

@MiCurry thanks for your work on this. Please merge, or I can if you prefer.

@mark-petersen I don't think @MiCurry is able to push to the develop branch, so I'll merge this today.