MATRIX DATA: Handle multiple splits properly

diff --git a/src/language/data-io/matrix-data.c b/src/language/data-io/matrix-data.c
index 7855caf4646d37d71091f9461e6d6971bb0ef155..fb214fd644a399c517652e854ad2c2ba31be804b 100644 (file)
--- a/src/language/data-io/matrix-data.c
+++ b/src/language/data-io/matrix-data.c
@@ -72,6 +72,8 @@ struct matrix_format
   const struct variable *rowtype;
   const struct variable *varname;
   int n_continuous_vars;
+  struct variable **split_vars;
+  size_t n_split_vars;
 };
 
 /*
@@ -114,16 +116,41 @@ preprocess (struct casereader *casereader0, const struct dictionary *dict, void
   struct casewriter *writer;
   writer = autopaging_writer_create (proto);
 
-  double *temp_matrix =
-    xcalloc (sizeof (*temp_matrix),
-            mformat->n_continuous_vars * mformat->n_continuous_vars);
+  double **matrices = NULL;
+  size_t n_splits = 0;
+
+  const size_t sizeof_matrix =
+    sizeof (double) * mformat->n_continuous_vars * mformat->n_continuous_vars;
+
 
   /* Make an initial pass to populate our temporary matrix */
   struct casereader *pass0 = casereader_clone (casereader0);
   struct ccase *c;
+  unsigned int prev_split_hash = 1;
   int row = (mformat->triangle == LOWER && mformat->diagonal == NO_DIAGONAL) ? 1 : 0;
   for (; (c = casereader_read (pass0)) != NULL; case_unref (c))
     {
+      int s;
+      unsigned int split_hash = 0;
+      for (s = 0; s < mformat->n_split_vars; ++s)
+       {
+         const struct variable *svar = mformat->split_vars[s];
+         const union value *sv = case_data (c, svar);
+         split_hash = value_hash (sv, var_get_width (svar), split_hash);
+       }
+
+      if (matrices == NULL || prev_split_hash != split_hash)
+       {
+         row = (mformat->triangle == LOWER && mformat->diagonal == NO_DIAGONAL) ?
+           1 : 0;
+
+         n_splits++;
+         matrices = xrealloc (matrices, sizeof (double*)  * n_splits);
+         matrices[n_splits - 1] = xmalloc (sizeof_matrix);
+       }
+
+      prev_split_hash = split_hash;
+
       int c_offset = (mformat->triangle == UPPER) ? row : 0;
       if (mformat->triangle == UPPER && mformat->diagonal == NO_DIAGONAL)
        c_offset++;
@@ -137,13 +164,16 @@ preprocess (struct casereader *casereader0, const struct dictionary *dict, void
            {
              const struct variable *var =
                dict_get_var (dict,
-                             1 + col - c_offset + var_get_dict_index (mformat->varname));
+                             1 + col - c_offset +
+                             var_get_dict_index (mformat->varname));
 
              double e = case_data (c, var)->f;
              if (e == SYSMIS)
                continue;
-             temp_matrix [col + mformat->n_continuous_vars * row] = e;
-             temp_matrix [row + mformat->n_continuous_vars * col] = e;
+
+
+             (matrices[n_splits-1])[col + mformat->n_continuous_vars * row] = e;
+             (matrices[n_splits-1]) [row + mformat->n_continuous_vars * col] = e;
            }
          row++;
        }
@@ -155,8 +185,26 @@ preprocess (struct casereader *casereader0, const struct dictionary *dict, void
   const int idx = var_get_dict_index (mformat->varname);
   row = 0;
   struct ccase *prev_case = NULL;
+  prev_split_hash = 1;
+  n_splits = 0;
   for (; (c = casereader_read (casereader0)) != NULL; prev_case = c)
     {
+      int s;
+      unsigned int split_hash = 0;
+      for (s = 0; s < mformat->n_split_vars; ++s)
+       {
+         const struct variable *svar = mformat->split_vars[s];
+         const union value *sv = case_data (c, svar);
+         split_hash = value_hash (sv, var_get_width (svar), split_hash);
+       }
+      if (prev_split_hash != split_hash)
+       {
+         n_splits++;
+         row = 0;
+       }
+
+      prev_split_hash = split_hash;
+
       case_unref (prev_case);
       struct ccase *outcase = case_create (proto);
       case_copy (outcase, 0, c, 0, caseproto_get_n_widths (proto));
@@ -175,7 +223,7 @@ preprocess (struct casereader *casereader0, const struct dictionary *dict, void
              union value *dest_val =
                case_data_rw_idx (outcase,
                                  1 + col + var_get_dict_index (mformat->varname));
-             dest_val->f = temp_matrix [col + mformat->n_continuous_vars * row];
+             dest_val->f = (matrices[n_splits - 1])[col + mformat->n_continuous_vars * row];
              if (col == row && mformat->diagonal == NO_DIAGONAL)
                dest_val->f = 1.0;
            }
@@ -219,7 +267,7 @@ preprocess (struct casereader *casereader0, const struct dictionary *dict, void
          union value *dest_val =
            case_data_rw_idx (outcase, 1 + col +
                              var_get_dict_index (mformat->varname));
-         dest_val->f = temp_matrix [col + mformat->n_continuous_vars * row];
+         dest_val->f = (matrices[n_splits - 1]) [col + mformat->n_continuous_vars * row];
          if (col == row && mformat->diagonal == NO_DIAGONAL)
            dest_val->f = 1.0;
        }
@@ -230,7 +278,10 @@ preprocess (struct casereader *casereader0, const struct dictionary *dict, void
   if (prev_case)
     case_unref (prev_case);
 
-  free (temp_matrix);
+  int i;
+  for (i = 0 ; i < n_splits; ++i)
+    free (matrices[i]);
+  free (matrices);
   struct casereader *reader1 = casewriter_make_reader (writer);
   casereader_destroy (casereader0);
   return reader1;
@@ -266,6 +317,7 @@ cmd_matrix (struct lexer *lexer, struct dataset *ds)
   mformat.varname = dict_create_var (dict, "VARNAME_", 8);
 
   mformat.n_continuous_vars = 0;
+  mformat.n_split_vars = 0;
 
   if (! lex_force_match_id (lexer, "VARIABLES"))
     goto error;
@@ -364,22 +416,19 @@ cmd_matrix (struct lexer *lexer, struct dataset *ds)
       else if (lex_match_id (lexer, "SPLIT"))
        {
          lex_match (lexer, T_EQUALS);
-         struct variable **split_vars = NULL;
-         size_t n_split_vars;
-         if (! parse_variables (lexer, dict, &split_vars, &n_split_vars, 0))
+         if (! parse_variables (lexer, dict, &mformat.split_vars, &mformat.n_split_vars, 0))
            {
-             free (split_vars);
+             free (mformat.split_vars);
              goto error;
            }
          int i;
-         for (i = 0; i < n_split_vars; ++i)
+         for (i = 0; i < mformat.n_split_vars; ++i)
            {
              const struct fmt_spec fmt = fmt_for_input (FMT_F, 4, 0);
-             var_set_both_formats (split_vars[i], &fmt);
+             var_set_both_formats (mformat.split_vars[i], &fmt);
            }
-         dict_reorder_vars (dict, split_vars, n_split_vars);
-         mformat.n_continuous_vars -= n_split_vars;
-         free (split_vars);
+         dict_reorder_vars (dict, mformat.split_vars, mformat.n_split_vars);
+         mformat.n_continuous_vars -= mformat.n_split_vars;
        }
       else
        {

diff --git a/tests/language/data-io/matrix-data.at b/tests/language/data-io/matrix-data.at
index cba500ff104100282e9f66388f5288ca59b44897..3047b1f57f55e9180a1d87ee7da0a0f4e1f27713 100644 (file)
--- a/tests/language/data-io/matrix-data.at
+++ b/tests/language/data-io/matrix-data.at
@@ -9,7 +9,7 @@ matrix data
     /format = lower diagonal
     /file = 'matrix.dat'
     .
-    
+
 list.
 ])
 
@@ -186,22 +186,27 @@ corr    ,var04,7.0000,5.0000,4.0000,1.0000
 AT_CLEANUP
 
 
-AT_SETUP([Matrix data split])
+
+AT_SETUP([Matrix data split data])
 
 AT_DATA([matrix-data.pspp], [dnl
 matrix data
-    variables = s1 s2 rowtype_  var01 TO var04
-    /split = s1 s2
-    /format = full diagonal.
+    variables = s1 s2 rowtype_  var01 var02 var03
+    /split=s1 s2.
 
 begin data
-1 1 mean 34 35 36 37
-1 1 sd   22 11 55 66
-1 0 n    100 101 102 103
-1 0 corr 1 9 8 7
-0 1 corr 9 1 6 5
-0 1 corr 8 6 1 4
-0 0 corr 7 5 4 1
+8 0   mean     21.4  5.0  72.9
+8 0   sd       6.5   1.6  22.8
+8 0   n        106   106  106
+8 0   corr     1
+8 0   corr    .41  1
+8 0   corr    -.16  -.22  1
+8 1   mean     11.4  1.0  52.9
+8 1   sd       9.5   8.6  12.8
+8 1   n        10   11  12
+8 1   corr     1
+8 1   corr    .51  1
+8 1   corr    .36  -.41  1
 end data.
 
 display dictionary.
@@ -209,6 +214,7 @@ display dictionary.
 list.
 ])
 
+
 AT_CHECK([pspp -O format=csv matrix-data.pspp], [0], [dnl
 Variable,Description,Position
 s1,Format: F4.0,1
@@ -218,20 +224,21 @@ VARNAME_,Format: A8,4
 var01,Format: F10.4,5
 var02,Format: F10.4,6
 var03,Format: F10.4,7
-var04,Format: F10.4,8
 
 Table: Data List
-s1,s2,ROWTYPE_,VARNAME_,var01,var02,var03,var04
-1,1,mean    ,,34.0000,35.0000,36.0000,37.0000
-1,1,STDDEV  ,,22.0000,11.0000,55.0000,66.0000
-1,0,n       ,,100.0000,101.0000,102.0000,103.0000
-1,0,corr    ,var01,1.0000,9.0000,8.0000,7.0000
-0,1,corr    ,var02,9.0000,1.0000,6.0000,5.0000
-0,1,corr    ,var03,8.0000,6.0000,1.0000,4.0000
-0,0,corr    ,var04,7.0000,5.0000,4.0000,1.0000
+s1,s2,ROWTYPE_,VARNAME_,var01,var02,var03
+8,0,mean    ,,21.4000,5.0000,72.9000
+8,0,STDDEV  ,,6.5000,1.6000,22.8000
+8,0,n       ,,106.0000,106.0000,106.0000
+8,0,corr    ,var01,1.0000,.4100,-.1600
+8,0,corr    ,var02,.4100,1.0000,-.2200
+8,0,corr    ,var03,-.1600,-.2200,1.0000
+8,1,mean    ,,11.4000,1.0000,52.9000
+8,1,STDDEV  ,,9.5000,8.6000,12.8000
+8,1,n       ,,10.0000,11.0000,12.0000
+8,1,corr    ,var01,1.0000,.5100,.3600
+8,1,corr    ,var02,.5100,1.0000,-.4100
+8,1,corr    ,var03,.3600,-.4100,1.0000
 ])
 
 AT_CLEANUP
-
-
-