Bugfix 1562 develop grid_diag (#1564)

* Per #1562, add the same grid_diag fix for the develop branch.

* Per #1562, removing the poly = CONUS.poly mask from GridDiagConfig_TMP. That settting masked a problem in the handling of missing data. Exercising the mask.poly option is tested in another unit test. This will change the output and break the nightly build, but that's good since we'll do more thorough testing.

met/src/libcode/vx_series_data/series_pdf.cc

@@ -66,8 +66,9 @@ void update_pdf(
 
     for(int i = 0; i < dp.nx(); i++) {
         for(int j = 0; j < dp.ny(); j++) {
-            if(!mp.s_is_on(i, j)) continue;
             double value = dp.get(i, j);
+            if(!mp.s_is_on(i, j) ||
+               is_bad_data(value)) continue;
             int k = floor((value - min) / delta);
             if(k < 0) k = 0;
             if(k >= pdf.size()) k = pdf.size() - 1;
@@ -92,12 +93,14 @@ void update_joint_pdf(
 
     for(int i = 0; i < dp_A.nx(); i++) {
         for(int j = 0; j < dp_A.ny(); j++) {
-            if(!mp.s_is_on(i, j)) continue;
             double value_A = dp_A.get(i, j);
+            double value_B = dp_B.get(i, j);
+            if(!mp.s_is_on(i, j)    ||
+               is_bad_data(value_A) ||
+               is_bad_data(value_B)) continue;
             int k_A = floor((value_A - min_A) / delta_A);
             if(k_A < 0) k_A = 0;
             if(k_A >= n_A) k_A = n_A - 1;
-            double value_B = dp_B.get(i, j);
             int k_B = floor((value_B - min_B) / delta_B);
             if(k_B < 0) k_B = 0;
             if(k_B >= n_B) k_B = n_B - 1;

test/config/GridDiagConfig_TMP

@@ -62,7 +62,7 @@ data = {
 
 mask = {
    grid = "";
-   poly = "MET_BASE/poly/CONUS.poly";
+   poly = "";
 }
 
 ////////////////////////////////////////////////////////////////////////////////