+static enum ctables_domain_type
+ctables_function_domain (enum ctables_summary_function function)
+{
+ switch (function)
+ {
+ case CTSF_COUNT:
+ case CTSF_ECOUNT:
+ case CTSF_MISSING:
+ case CSTF_TOTALN:
+ case CTSF_ETOTALN:
+ case CTSF_VALIDN:
+ case CTSF_EVALIDN:
+ case CTSF_MAXIMUM:
+ case CTSF_MINIMUM:
+ case CTSF_RANGE:
+ case CTSF_MEAN:
+ case CTSF_SEMEAN:
+ case CTSF_STDDEV:
+ case CTSF_SUM:
+ case CTSF_VARIANCE:
+ case CTSF_MEDIAN:
+ case CTSF_PTILE:
+ case CTSF_MODE:
+ case CTSF_RESPONSES:
+ NOT_REACHED ();
+
+ case CTSF_COLPCT_COUNT:
+ case CTSF_COLPCT_COUNT_RESPONSES:
+ case CTSF_COLPCT_RESPONSES:
+ case CTSF_COLPCT_RESPONSES_COUNT:
+ case CTSF_COLPCT_SUM:
+ case CTSF_COLPCT_TOTALN:
+ case CTSF_COLPCT_VALIDN:
+ return CTDT_COL;
+
+ case CTSF_LAYERCOLPCT_COUNT:
+ case CTSF_LAYERCOLPCT_COUNT_RESPONSES:
+ case CTSF_LAYERCOLPCT_RESPONSES:
+ case CTSF_LAYERCOLPCT_RESPONSES_COUNT:
+ case CTSF_LAYERCOLPCT_SUM:
+ case CTSF_LAYERCOLPCT_TOTALN:
+ case CTSF_LAYERCOLPCT_VALIDN:
+ return CTDT_LAYERCOL;
+
+ case CTSF_LAYERPCT_COUNT:
+ case CTSF_LAYERPCT_COUNT_RESPONSES:
+ case CTSF_LAYERPCT_RESPONSES:
+ case CTSF_LAYERPCT_RESPONSES_COUNT:
+ case CTSF_LAYERPCT_SUM:
+ case CTSF_LAYERPCT_TOTALN:
+ case CTSF_LAYERPCT_VALIDN:
+ return CTDT_LAYER;
+
+ case CTSF_LAYERROWPCT_COUNT:
+ case CTSF_LAYERROWPCT_COUNT_RESPONSES:
+ case CTSF_LAYERROWPCT_RESPONSES:
+ case CTSF_LAYERROWPCT_RESPONSES_COUNT:
+ case CTSF_LAYERROWPCT_SUM:
+ case CTSF_LAYERROWPCT_TOTALN:
+ case CTSF_LAYERROWPCT_VALIDN:
+ return CTDT_LAYERROW;
+
+ case CTSF_ROWPCT_COUNT:
+ case CTSF_ROWPCT_COUNT_RESPONSES:
+ case CTSF_ROWPCT_RESPONSES:
+ case CTSF_ROWPCT_RESPONSES_COUNT:
+ case CTSF_ROWPCT_SUM:
+ case CTSF_ROWPCT_TOTALN:
+ case CTSF_ROWPCT_VALIDN:
+ return CTDT_ROW;
+
+ case CTSF_SUBTABLEPCT_COUNT:
+ case CTSF_SUBTABLEPCT_COUNT_RESPONSES:
+ case CTSF_SUBTABLEPCT_RESPONSES:
+ case CTSF_SUBTABLEPCT_RESPONSES_COUNT:
+ case CTSF_SUBTABLEPCT_SUM:
+ case CTSF_SUBTABLEPCT_TOTALN:
+ case CTSF_SUBTABLEPCT_VALIDN:
+ return CTDT_SUBTABLE;
+
+ case CTSF_TABLEPCT_COUNT:
+ case CTSF_TABLEPCT_COUNT_RESPONSES:
+ case CTSF_TABLEPCT_RESPONSES:
+ case CTSF_TABLEPCT_RESPONSES_COUNT:
+ case CTSF_TABLEPCT_SUM:
+ case CTSF_TABLEPCT_TOTALN:
+ case CTSF_TABLEPCT_VALIDN:
+ return CTDT_TABLE;
+ }
+
+ NOT_REACHED ();
+}
+
+static double
+ctables_summary_value (const struct ctables_cell *cell,
+ union ctables_summary *s,
+ const struct ctables_summary_spec *ss)
+{
+ switch (ss->function)
+ {
+ case CTSF_COUNT:
+ case CTSF_ECOUNT:
+ return s->valid;
+
+ case CTSF_ROWPCT_COUNT:
+ case CTSF_COLPCT_COUNT:
+ case CTSF_TABLEPCT_COUNT:
+ case CTSF_SUBTABLEPCT_COUNT:
+ case CTSF_LAYERPCT_COUNT:
+ case CTSF_LAYERROWPCT_COUNT:
+ case CTSF_LAYERCOLPCT_COUNT:
+ {
+ enum ctables_domain_type d = ctables_function_domain (ss->function);
+ return (cell->domains[d]->e_valid
+ ? s->valid / cell->domains[d]->e_valid * 100
+ : SYSMIS);
+ }
+
+ case CTSF_ROWPCT_VALIDN:
+ case CTSF_COLPCT_VALIDN:
+ case CTSF_TABLEPCT_VALIDN:
+ case CTSF_SUBTABLEPCT_VALIDN:
+ case CTSF_LAYERPCT_VALIDN:
+ case CTSF_LAYERROWPCT_VALIDN:
+ case CTSF_LAYERCOLPCT_VALIDN:
+ case CTSF_ROWPCT_TOTALN:
+ case CTSF_COLPCT_TOTALN:
+ case CTSF_TABLEPCT_TOTALN:
+ case CTSF_SUBTABLEPCT_TOTALN:
+ case CTSF_LAYERPCT_TOTALN:
+ case CTSF_LAYERROWPCT_TOTALN:
+ case CTSF_LAYERCOLPCT_TOTALN:
+ NOT_REACHED ();
+
+ case CTSF_MISSING:
+ return s->missing;
+
+ case CSTF_TOTALN:
+ case CTSF_ETOTALN:
+ return s->valid + s->missing;
+
+ case CTSF_VALIDN:
+ case CTSF_EVALIDN:
+ return s->valid;
+
+ case CTSF_MAXIMUM:
+ return s->max;
+
+ case CTSF_MINIMUM:
+ return s->min;
+
+ case CTSF_RANGE:
+ return s->max != SYSMIS && s->min != SYSMIS ? s->max - s->min : SYSMIS;
+
+ case CTSF_MEAN:
+ {
+ double mean;
+ moments1_calculate (s->moments, NULL, &mean, NULL, NULL, NULL);
+ return mean;
+ }
+
+ case CTSF_SEMEAN:
+ {
+ double weight, variance;
+ moments1_calculate (s->moments, &weight, NULL, &variance, NULL, NULL);
+ return calc_semean (variance, weight);
+ }
+
+ case CTSF_STDDEV:
+ {
+ double variance;
+ moments1_calculate (s->moments, NULL, NULL, &variance, NULL, NULL);
+ return variance != SYSMIS ? sqrt (variance) : SYSMIS;
+ }
+
+ case CTSF_SUM:
+ {
+ double weight, mean;
+ moments1_calculate (s->moments, &weight, &mean, NULL, NULL, NULL);
+ return weight != SYSMIS && mean != SYSMIS ? weight * mean : SYSMIS;
+ }
+
+ case CTSF_VARIANCE:
+ {
+ double variance;
+ moments1_calculate (s->moments, NULL, NULL, &variance, NULL, NULL);
+ return variance;
+ }
+
+ case CTSF_ROWPCT_SUM:
+ case CTSF_COLPCT_SUM:
+ case CTSF_TABLEPCT_SUM:
+ case CTSF_SUBTABLEPCT_SUM:
+ case CTSF_LAYERPCT_SUM:
+ case CTSF_LAYERROWPCT_SUM:
+ case CTSF_LAYERCOLPCT_SUM:
+ NOT_REACHED ();
+
+ case CTSF_MEDIAN:
+ case CTSF_PTILE:
+ if (s->writer)
+ {
+ struct casereader *reader = casewriter_make_reader (s->writer);
+ s->writer = NULL;
+
+ struct percentile *ptile = percentile_create (
+ ss->function == CTSF_PTILE ? ss->percentile : 0.5, s->ovalid);
+ struct order_stats *os = &ptile->parent;
+ order_stats_accumulate_idx (&os, 1, reader, 1, 0);
+ s->ovalue = percentile_calculate (ptile, PC_HAVERAGE);
+ statistic_destroy (&ptile->parent.parent);
+ }
+ return s->ovalue;
+
+ case CTSF_MODE:
+ if (s->writer)
+ {
+ struct casereader *reader = casewriter_make_reader (s->writer);
+ s->writer = NULL;
+
+ struct mode *mode = mode_create ();
+ struct order_stats *os = &mode->parent;
+ order_stats_accumulate_idx (&os, 1, reader, 1, 0);
+ s->ovalue = mode->mode;
+ statistic_destroy (&mode->parent.parent);
+ }
+ return s->ovalue;
+
+ case CTSF_RESPONSES:
+ case CTSF_ROWPCT_RESPONSES:
+ case CTSF_COLPCT_RESPONSES:
+ case CTSF_TABLEPCT_RESPONSES:
+ case CTSF_SUBTABLEPCT_RESPONSES:
+ case CTSF_LAYERPCT_RESPONSES:
+ case CTSF_LAYERROWPCT_RESPONSES:
+ case CTSF_LAYERCOLPCT_RESPONSES:
+ case CTSF_ROWPCT_RESPONSES_COUNT:
+ case CTSF_COLPCT_RESPONSES_COUNT:
+ case CTSF_TABLEPCT_RESPONSES_COUNT:
+ case CTSF_SUBTABLEPCT_RESPONSES_COUNT:
+ case CTSF_LAYERPCT_RESPONSES_COUNT:
+ case CTSF_LAYERROWPCT_RESPONSES_COUNT:
+ case CTSF_LAYERCOLPCT_RESPONSES_COUNT:
+ case CTSF_ROWPCT_COUNT_RESPONSES:
+ case CTSF_COLPCT_COUNT_RESPONSES:
+ case CTSF_TABLEPCT_COUNT_RESPONSES:
+ case CTSF_SUBTABLEPCT_COUNT_RESPONSES:
+ case CTSF_LAYERPCT_COUNT_RESPONSES:
+ case CTSF_LAYERROWPCT_COUNT_RESPONSES:
+ case CTSF_LAYERCOLPCT_COUNT_RESPONSES:
+ NOT_REACHED ();
+ }
+
+ NOT_REACHED ();
+}
+
+struct ctables_cell_sort_aux
+ {
+ const struct ctables_nest *nest;
+ enum pivot_axis_type a;
+ };
+
+static int
+ctables_cell_compare_3way (const void *a_, const void *b_, const void *aux_)
+{
+ const struct ctables_cell_sort_aux *aux = aux_;
+ struct ctables_cell *const *ap = a_;
+ struct ctables_cell *const *bp = b_;
+ const struct ctables_cell *a = *ap;
+ const struct ctables_cell *b = *bp;
+
+ const struct ctables_nest *nest = aux->nest;
+ for (size_t i = 0; i < nest->n; i++)
+ if (i != nest->scale_idx)
+ {
+ const struct variable *var = nest->vars[i];
+ const struct ctables_cell_value *a_cv = &a->axes[aux->a].cvs[i];
+ const struct ctables_cell_value *b_cv = &b->axes[aux->a].cvs[i];
+ if (a_cv->category != b_cv->category)
+ return a_cv->category > b_cv->category ? 1 : -1;
+
+ const union value *a_val = &a_cv->value;
+ const union value *b_val = &b_cv->value;
+ switch (a_cv->category->type)
+ {
+ case CCT_NUMBER:
+ case CCT_STRING:
+ case CCT_SUBTOTAL:
+ case CCT_HSUBTOTAL:
+ case CCT_TOTAL:
+ /* Must be equal. */
+ continue;
+
+ case CCT_RANGE:
+ case CCT_MISSING:
+ case CCT_OTHERNM:
+ {
+ int cmp = value_compare_3way (a_val, b_val, var_get_width (var));
+ if (cmp)
+ return cmp;
+ }
+ break;
+
+ case CCT_VALUE:
+ {
+ int cmp = value_compare_3way (a_val, b_val, var_get_width (var));
+ if (cmp)
+ return a_cv->category->sort_ascending ? cmp : -cmp;
+ }
+ break;
+
+ case CCT_LABEL:
+ {
+ const char *a_label = var_lookup_value_label (var, a_val);
+ const char *b_label = var_lookup_value_label (var, b_val);
+ int cmp = (a_label
+ ? (b_label ? strcmp (a_label, b_label) : 1)
+ : (b_label ? -1 : value_compare_3way (
+ a_val, b_val, var_get_width (var))));
+ if (cmp)
+ return a_cv->category->sort_ascending ? cmp : -cmp;
+ }
+ break;
+
+ case CCT_FUNCTION:
+ NOT_REACHED ();
+ }
+ }
+ return 0;
+}
+
+/* Algorithm:
+
+ For each row:
+ For each ctables_table:
+ For each combination of row vars:
+ For each combination of column vars:
+ For each combination of layer vars:
+ Add entry
+ Make a table of row values:
+ Sort entries by row values
+ Assign a 0-based index to each actual value
+ Construct a dimension
+ Make a table of column values
+ Make a table of layer values
+ For each entry:
+ Fill the table entry using the indexes from before.
+ */
+
+static struct ctables_domain *
+ctables_domain_insert (struct ctables_section *s, struct ctables_cell *cell,
+ enum ctables_domain_type domain)
+{
+ size_t hash = 0;
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ {
+ const struct ctables_nest *nest = s->nests[a];
+ for (size_t i = 0; i < nest->n_domains[domain]; i++)
+ {
+ size_t v_idx = nest->domains[domain][i];
+ hash = value_hash (&cell->axes[a].cvs[v_idx].value,
+ var_get_width (nest->vars[v_idx]), hash);
+ }
+ }
+
+ struct ctables_domain *d;
+ HMAP_FOR_EACH_WITH_HASH (d, struct ctables_domain, node, hash, &s->domains[domain])
+ {
+ const struct ctables_cell *df = d->example;
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ {
+ const struct ctables_nest *nest = s->nests[a];
+ for (size_t i = 0; i < nest->n_domains[domain]; i++)
+ {
+ size_t v_idx = nest->domains[domain][i];
+ if (!value_equal (&df->axes[a].cvs[v_idx].value,
+ &cell->axes[a].cvs[v_idx].value,
+ var_get_width (nest->vars[v_idx])))
+ goto not_equal;
+ }
+ }
+ return d;
+
+ not_equal: ;
+ }
+
+ d = xmalloc (sizeof *d);
+ *d = (struct ctables_domain) { .example = cell };
+ hmap_insert (&s->domains[domain], &d->node, hash);
+ return d;
+}
+
+static const struct ctables_category *
+ctables_categories_match (const struct ctables_categories *c,
+ const union value *v, const struct variable *var)
+{
+ const struct ctables_category *othernm = NULL;
+ for (size_t i = c->n_cats; i-- > 0; )
+ {
+ const struct ctables_category *cat = &c->cats[i];
+ switch (cat->type)
+ {
+ case CCT_NUMBER:
+ if (cat->number == v->f)
+ return cat;
+ break;
+
+ case CCT_STRING:
+ NOT_REACHED ();
+
+ case CCT_RANGE:
+ if ((cat->range[0] == -DBL_MAX || v->f >= cat->range[0])
+ && (cat->range[1] == DBL_MAX || v->f <= cat->range[1]))
+ return cat;
+ break;
+
+ case CCT_MISSING:
+ if (var_is_value_missing (var, v))
+ return cat;
+ break;
+
+ case CCT_OTHERNM:
+ if (!othernm)
+ othernm = cat;
+ break;
+
+ case CCT_SUBTOTAL:
+ case CCT_HSUBTOTAL:
+ case CCT_TOTAL:
+ break;
+
+ case CCT_VALUE:
+ case CCT_LABEL:
+ case CCT_FUNCTION:
+ return (cat->include_missing || !var_is_value_missing (var, v) ? cat
+ : NULL);
+ }
+ }
+
+ return var_is_value_missing (var, v) ? NULL : othernm;
+}
+
+static const struct ctables_category *
+ctables_categories_total (const struct ctables_categories *c)
+{
+ const struct ctables_category *first = &c->cats[0];
+ const struct ctables_category *last = &c->cats[c->n_cats - 1];
+ return (first->type == CCT_TOTAL ? first
+ : last->type == CCT_TOTAL ? last
+ : NULL);
+}
+
+static struct ctables_cell *
+ctables_cell_insert__ (struct ctables_section *s, const struct ccase *c,
+ const struct ctables_category *cats[PIVOT_N_AXES][10])
+{
+ const struct ctables_nest *ss = s->nests[s->table->summary_axis];
+
+ size_t hash = 0;
+ enum ctables_summary_variant sv = CSV_CELL;
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ {
+ const struct ctables_nest *nest = s->nests[a];
+ for (size_t i = 0; i < nest->n; i++)
+ if (i != nest->scale_idx)
+ {
+ hash = hash_pointer (cats[a][i], hash);
+ if (cats[a][i]->type != CCT_TOTAL
+ && cats[a][i]->type != CCT_SUBTOTAL
+ && cats[a][i]->type != CCT_HSUBTOTAL)
+ hash = value_hash (case_data (c, nest->vars[i]),
+ var_get_width (nest->vars[i]), hash);
+ else
+ sv = CSV_TOTAL;
+ }
+ }
+
+ struct ctables_cell *cell;
+ HMAP_FOR_EACH_WITH_HASH (cell, struct ctables_cell, node, hash, &s->cells)
+ {
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ {
+ const struct ctables_nest *nest = s->nests[a];
+ for (size_t i = 0; i < nest->n; i++)
+ if (i != nest->scale_idx
+ && (cats[a][i] != cell->axes[a].cvs[i].category
+ || (cats[a][i]->type != CCT_TOTAL
+ && cats[a][i]->type != CCT_SUBTOTAL
+ && cats[a][i]->type != CCT_HSUBTOTAL
+ && !value_equal (case_data (c, nest->vars[i]),
+ &cell->axes[a].cvs[i].value,
+ var_get_width (nest->vars[i])))))
+ goto not_equal;
+ }
+
+ return cell;
+
+ not_equal: ;
+ }
+
+ cell = xmalloc (sizeof *cell);
+ cell->hide = false;
+ cell->sv = sv;
+ cell->contributes_to_domains = true;
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ {
+ const struct ctables_nest *nest = s->nests[a];
+ cell->axes[a].cvs = (nest->n
+ ? xnmalloc (nest->n, sizeof *cell->axes[a].cvs)
+ : NULL);
+ for (size_t i = 0; i < nest->n; i++)
+ {
+ const struct ctables_category *cat = cats[a][i];
+
+ if (i != nest->scale_idx)
+ {
+ const struct ctables_category *subtotal = cat->subtotal;
+ if (subtotal && subtotal->type == CCT_HSUBTOTAL)
+ cell->hide = true;
+
+ if (cat->type == CCT_TOTAL || cat->type == CCT_SUBTOTAL || cat->type == CCT_HSUBTOTAL)
+ cell->contributes_to_domains = false;
+ }
+
+ cell->axes[a].cvs[i].category = cat;
+ value_clone (&cell->axes[a].cvs[i].value, case_data (c, nest->vars[i]),
+ var_get_width (nest->vars[i]));
+ }
+ }
+
+ const struct ctables_summary_spec_set *specs = &ss->specs[cell->sv];
+ cell->summaries = xmalloc (specs->n * sizeof *cell->summaries);
+ for (size_t i = 0; i < specs->n; i++)
+ ctables_summary_init (&cell->summaries[i], &specs->specs[i]);
+ for (enum ctables_domain_type dt = 0; dt < N_CTDTS; dt++)
+ cell->domains[dt] = ctables_domain_insert (s, cell, dt);
+ hmap_insert (&s->cells, &cell->node, hash);
+ return cell;
+}
+
+static void
+ctables_cell_add__ (struct ctables_section *s, const struct ccase *c,
+ const struct ctables_category *cats[PIVOT_N_AXES][10],
+ double d_weight, double e_weight)
+{
+ struct ctables_cell *cell = ctables_cell_insert__ (s, c, cats);
+ const struct ctables_nest *ss = s->nests[s->table->summary_axis];
+
+ const struct ctables_summary_spec_set *specs = &ss->specs[cell->sv];
+ for (size_t i = 0; i < specs->n; i++)
+ ctables_summary_add (&cell->summaries[i], &specs->specs[i], specs->var,
+ case_data (c, specs->var), d_weight, e_weight);
+ if (cell->contributes_to_domains)
+ {
+ for (enum ctables_domain_type dt = 0; dt < N_CTDTS; dt++)
+ {
+ cell->domains[dt]->d_valid += d_weight;
+ cell->domains[dt]->e_valid += e_weight;
+ }
+ }
+}
+
+static void
+recurse_totals (struct ctables_section *s, const struct ccase *c,
+ const struct ctables_category *cats[PIVOT_N_AXES][10],
+ double d_weight, double e_weight,
+ enum pivot_axis_type start_axis, size_t start_nest)
+{
+ for (enum pivot_axis_type a = start_axis; a < PIVOT_N_AXES; a++)
+ {
+ const struct ctables_nest *nest = s->nests[a];
+ for (size_t i = start_nest; i < nest->n; i++)
+ {
+ if (i == nest->scale_idx)
+ continue;
+
+ const struct variable *var = nest->vars[i];
+
+ const struct ctables_category *total = ctables_categories_total (
+ s->table->categories[var_get_dict_index (var)]);
+ if (total)
+ {
+ const struct ctables_category *save = cats[a][i];
+ cats[a][i] = total;
+ ctables_cell_add__ (s, c, cats, d_weight, e_weight);
+ recurse_totals (s, c, cats, d_weight, e_weight, a, i + 1);
+ cats[a][i] = save;
+ }
+ }
+ start_nest = 0;
+ }
+}
+
+static void
+recurse_subtotals (struct ctables_section *s, const struct ccase *c,
+ const struct ctables_category *cats[PIVOT_N_AXES][10],
+ double d_weight, double e_weight,
+ enum pivot_axis_type start_axis, size_t start_nest)
+{
+ for (enum pivot_axis_type a = start_axis; a < PIVOT_N_AXES; a++)
+ {
+ const struct ctables_nest *nest = s->nests[a];
+ for (size_t i = start_nest; i < nest->n; i++)
+ {
+ if (i == nest->scale_idx)
+ continue;
+
+ const struct ctables_category *save = cats[a][i];
+ if (save->subtotal)
+ {
+ cats[a][i] = save->subtotal;
+ ctables_cell_add__ (s, c, cats, d_weight, e_weight);
+ recurse_subtotals (s, c, cats, d_weight, e_weight, a, i + 1);
+ cats[a][i] = save;
+ }
+ }
+ start_nest = 0;
+ }
+}
+
+static void
+ctables_add_occurrence (const struct variable *var,
+ const union value *value,
+ struct hmap *occurrences)
+{
+ int width = var_get_width (var);
+ unsigned int hash = value_hash (value, width, 0);
+
+ struct ctables_section_value *sv;
+ HMAP_FOR_EACH_WITH_HASH (sv, struct ctables_section_value, node, hash,
+ occurrences)
+ if (value_equal (value, &sv->value, width))
+ return;
+
+ sv = xmalloc (sizeof *sv);
+ value_clone (&sv->value, value, width);
+ hmap_insert (occurrences, &sv->node, hash);
+}
+
+static void
+ctables_cell_insert (struct ctables_section *s,
+ const struct ccase *c,
+ double d_weight, double e_weight)
+{
+ const struct ctables_category *cats[PIVOT_N_AXES][10]; /* XXX */
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ {
+ const struct ctables_nest *nest = s->nests[a];
+ for (size_t i = 0; i < nest->n; i++)
+ {
+ if (i == nest->scale_idx)
+ continue;
+
+ const struct variable *var = nest->vars[i];
+ const union value *value = case_data (c, var);
+
+ if (var_is_numeric (var) && value->f == SYSMIS)
+ return;
+
+ cats[a][i] = ctables_categories_match (
+ s->table->categories[var_get_dict_index (var)], value, var);
+ if (!cats[a][i])
+ return;
+ }
+ }
+
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ {
+ const struct ctables_nest *nest = s->nests[a];
+ for (size_t i = 0; i < nest->n; i++)
+ if (i != nest->scale_idx)
+ {
+ const struct variable *var = nest->vars[i];
+ const union value *value = case_data (c, var);
+ ctables_add_occurrence (var, value, &s->occurrences[a][i]);
+ }
+ }
+
+ ctables_cell_add__ (s, c, cats, d_weight, e_weight);
+
+ recurse_totals (s, c, cats, d_weight, e_weight, 0, 0);
+ recurse_subtotals (s, c, cats, d_weight, e_weight, 0, 0);
+}
+
+struct merge_item
+ {
+ const struct ctables_summary_spec_set *set;
+ size_t ofs;
+ };
+
+static int
+merge_item_compare_3way (const struct merge_item *a, const struct merge_item *b)
+{
+ const struct ctables_summary_spec *as = &a->set->specs[a->ofs];
+ const struct ctables_summary_spec *bs = &b->set->specs[b->ofs];
+ if (as->function != bs->function)
+ return as->function > bs->function ? 1 : -1;
+ else if (as->percentile != bs->percentile)
+ return as->percentile < bs->percentile ? 1 : -1;
+ return strcmp (as->label, bs->label);
+}
+
+static struct pivot_value *
+ctables_category_create_label (const struct ctables_category *cat,
+ const struct variable *var,
+ const union value *value)
+{
+ return (cat->type == CCT_TOTAL || cat->type == CCT_SUBTOTAL || cat->type == CCT_HSUBTOTAL
+ ? pivot_value_new_user_text (cat->total_label, SIZE_MAX)
+ : pivot_value_new_var_value (var, value));
+}
+
+static struct ctables_value *
+ctables_value_find__ (struct ctables_table *t, const union value *value,
+ int width, unsigned int hash)
+{
+ struct ctables_value *clv;
+ HMAP_FOR_EACH_WITH_HASH (clv, struct ctables_value, node,
+ hash, &t->clabels_values_map)
+ if (value_equal (value, &clv->value, width))
+ return clv;
+ return NULL;
+}
+
+static struct ctables_value *
+ctables_value_find (struct ctables_table *t,
+ const union value *value, int width)
+{
+ return ctables_value_find__ (t, value, width,
+ value_hash (value, width, 0));
+}
+
+static void
+ctables_table_add_section (struct ctables_table *t, enum pivot_axis_type a,
+ size_t ix[PIVOT_N_AXES])
+{
+ if (a < PIVOT_N_AXES)
+ {
+ size_t limit = MAX (t->stacks[a].n, 1);
+ for (ix[a] = 0; ix[a] < limit; ix[a]++)
+ ctables_table_add_section (t, a + 1, ix);
+ }
+ else
+ {
+ struct ctables_section *s = &t->sections[t->n_sections++];
+ *s = (struct ctables_section) {
+ .table = t,
+ .cells = HMAP_INITIALIZER (s->cells),
+ };
+ for (a = 0; a < PIVOT_N_AXES; a++)
+ if (t->stacks[a].n)
+ {
+ struct ctables_nest *nest = &t->stacks[a].nests[ix[a]];
+ s->nests[a] = nest;
+ s->occurrences[a] = xnmalloc (nest->n, sizeof *s->occurrences[a]);
+ for (size_t i = 0; i < nest->n; i++)
+ hmap_init (&s->occurrences[a][i]);
+ }
+ for (size_t i = 0; i < N_CTDTS; i++)
+ hmap_init (&s->domains[i]);
+ }
+}
+
+static void
+ctables_table_output (struct ctables *ct, struct ctables_table *t)
+{
+ struct pivot_table *pt = pivot_table_create__ (
+ (t->title
+ ? pivot_value_new_user_text (t->title, SIZE_MAX)
+ : pivot_value_new_text (N_("Custom Tables"))),
+ "Custom Tables");
+ if (t->caption)
+ pivot_table_set_caption (
+ pt, pivot_value_new_user_text (t->caption, SIZE_MAX));
+ if (t->corner)
+ pivot_table_set_caption (
+ pt, pivot_value_new_user_text (t->corner, SIZE_MAX));
+
+ bool summary_dimension = (t->summary_axis != t->slabels_axis
+ || (!t->slabels_visible
+ && t->summary_specs.n > 1));
+ if (summary_dimension)
+ {
+ struct pivot_dimension *d = pivot_dimension_create (
+ pt, t->slabels_axis, N_("Statistics"));
+ const struct ctables_summary_spec_set *specs = &t->summary_specs;
+ if (!t->slabels_visible)
+ d->hide_all_labels = true;
+ for (size_t i = 0; i < specs->n; i++)
+ pivot_category_create_leaf (
+ d->root, pivot_value_new_text (specs->specs[i].label));
+ }
+
+ bool categories_dimension = t->clabels_example != NULL;
+ if (categories_dimension)
+ {
+ struct pivot_dimension *d = pivot_dimension_create (
+ pt, t->label_axis[t->clabels_from_axis],
+ t->clabels_from_axis == PIVOT_AXIS_ROW
+ ? N_("Row Categories")
+ : N_("Column Categories"));
+ const struct variable *var = t->clabels_example;
+ const struct ctables_categories *c = t->categories[var_get_dict_index (var)];
+ for (size_t i = 0; i < t->n_clabels_values; i++)
+ {
+ const struct ctables_value *value = t->clabels_values[i];
+ const struct ctables_category *cat = ctables_categories_match (c, &value->value, var);
+ assert (cat != NULL);
+ pivot_category_create_leaf (d->root, ctables_category_create_label (
+ cat, t->clabels_example, &value->value));
+ }
+ }
+
+ pivot_table_set_look (pt, ct->look);
+ struct pivot_dimension *d[PIVOT_N_AXES];
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ {
+ static const char *names[] = {
+ [PIVOT_AXIS_ROW] = N_("Rows"),
+ [PIVOT_AXIS_COLUMN] = N_("Columns"),
+ [PIVOT_AXIS_LAYER] = N_("Layers"),
+ };
+ d[a] = (t->axes[a] || a == t->summary_axis
+ ? pivot_dimension_create (pt, a, names[a])
+ : NULL);
+ if (!d[a])
+ continue;
+
+ assert (t->axes[a]);
+
+ for (size_t i = 0; i < t->stacks[a].n; i++)
+ {
+ struct ctables_nest *nest = &t->stacks[a].nests[i];
+ struct ctables_section **sections = xnmalloc (t->n_sections,
+ sizeof *sections);
+ size_t n_sections = 0;
+
+ size_t n_total_cells = 0;
+ size_t max_depth = 0;
+ for (size_t j = 0; j < t->n_sections; j++)
+ if (t->sections[j].nests[a] == nest)
+ {
+ struct ctables_section *s = &t->sections[j];
+ sections[n_sections++] = s;
+ n_total_cells += s->cells.count;
+
+ size_t depth = s->nests[a]->n;
+ max_depth = MAX (depth, max_depth);
+ }
+
+ struct ctables_cell **sorted = xnmalloc (n_total_cells,
+ sizeof *sorted);
+ size_t n_sorted = 0;
+
+ for (size_t j = 0; j < n_sections; j++)
+ {
+ struct ctables_section *s = sections[j];
+
+ struct ctables_cell *cell;
+ HMAP_FOR_EACH (cell, struct ctables_cell, node, &s->cells)
+ if (!cell->hide)
+ sorted[n_sorted++] = cell;
+ assert (n_sorted <= n_total_cells);
+ }
+
+ struct ctables_cell_sort_aux aux = { .nest = nest, .a = a };
+ sort (sorted, n_sorted, sizeof *sorted, ctables_cell_compare_3way, &aux);
+
+ struct ctables_level
+ {
+ enum ctables_level_type
+ {
+ CTL_VAR, /* Variable label for nest->vars[var_idx]. */
+ CTL_CATEGORY, /* Category for nest->vars[var_idx]. */
+ CTL_SUMMARY, /* Summary functions. */
+ }
+ type;
+
+ size_t var_idx;
+ };
+ struct ctables_level *levels = xnmalloc (1 + 2 * max_depth, sizeof *levels);
+ size_t n_levels = 0;
+ for (size_t k = 0; k < nest->n; k++)
+ {
+ enum ctables_vlabel vlabel = ct->vlabels[var_get_dict_index (nest->vars[k])];
+ if (vlabel != CTVL_NONE)
+ {
+ levels[n_levels++] = (struct ctables_level) {
+ .type = CTL_VAR,
+ .var_idx = k,
+ };
+ }
+
+ if (nest->scale_idx != k
+ && (k != nest->n - 1 || t->label_axis[a] == a))
+ {
+ levels[n_levels++] = (struct ctables_level) {
+ .type = CTL_CATEGORY,
+ .var_idx = k,
+ };
+ }
+ }
+
+ if (!summary_dimension && a == t->slabels_axis)
+ {
+ levels[n_levels++] = (struct ctables_level) {
+ .type = CTL_SUMMARY,
+ .var_idx = SIZE_MAX,
+ };
+ }
+
+ /* Pivot categories:
+
+ - variable label for nest->vars[0], if vlabel != CTVL_NONE
+ - category for nest->vars[0], if nest->scale_idx != 0
+ - variable label for nest->vars[1], if vlabel != CTVL_NONE
+ - category for nest->vars[1], if nest->scale_idx != 1
+ ...
+ - variable label for nest->vars[n - 1], if vlabel != CTVL_NONE
+ - category for nest->vars[n - 1], if t->label_axis[a] == a && nest->scale_idx != n - 1.
+ - summary function, if 'a == t->slabels_axis && a ==
+ t->summary_axis'.
+
+ Additional dimensions:
+
+ - If 'a == t->slabels_axis && a != t->summary_axis', add a summary
+ dimension.
+ - If 't->label_axis[b] == a' for some 'b != a', add a category
+ dimension to 'a'.
+ */
+
+
+ struct pivot_category **groups = xnmalloc (1 + 2 * max_depth, sizeof *groups);
+ int prev_leaf = 0;
+ for (size_t j = 0; j < n_sorted; j++)
+ {
+ struct ctables_cell *cell = sorted[j];
+ struct ctables_cell *prev = j > 0 ? sorted[j - 1] : NULL;
+
+ size_t n_common = 0;
+ if (j > 0)
+ {
+ for (; n_common < n_levels; n_common++)
+ {
+ const struct ctables_level *level = &levels[n_common];
+ if (level->type == CTL_CATEGORY)
+ {
+ size_t var_idx = level->var_idx;
+ const struct ctables_category *c = cell->axes[a].cvs[var_idx].category;
+ if (prev->axes[a].cvs[var_idx].category != c)
+ break;
+ else if (c->type != CCT_SUBTOTAL
+ && c->type != CCT_HSUBTOTAL
+ && c->type != CCT_TOTAL
+ && !value_equal (&prev->axes[a].cvs[var_idx].value,
+ &cell->axes[a].cvs[var_idx].value,
+ var_get_type (nest->vars[var_idx])))
+ break;
+ }
+ }
+ }
+
+ for (size_t k = n_common; k < n_levels; k++)
+ {
+ const struct ctables_level *level = &levels[k];
+ struct pivot_category *parent = k ? groups[k - 1] : d[a]->root;
+ if (level->type == CTL_SUMMARY)
+ {
+ assert (k == n_levels - 1);
+
+ const struct ctables_summary_spec_set *specs = &t->summary_specs;
+ for (size_t m = 0; m < specs->n; m++)
+ {
+ int leaf = pivot_category_create_leaf (
+ parent, pivot_value_new_text (specs->specs[m].label));
+ if (!m)
+ prev_leaf = leaf;
+ }
+ }
+ else
+ {
+ const struct variable *var = nest->vars[level->var_idx];
+ struct pivot_value *label;
+ if (level->type == CTL_VAR)
+ label = pivot_value_new_variable (var);
+ else if (level->type == CTL_CATEGORY)
+ {
+ const struct ctables_cell_value *cv = &cell->axes[a].cvs[level->var_idx];
+ label = ctables_category_create_label (cv->category,
+ var, &cv->value);
+ }
+ else
+ NOT_REACHED ();
+
+ if (k == n_levels - 1)
+ prev_leaf = pivot_category_create_leaf (parent, label);
+ else
+ groups[k] = pivot_category_create_group__ (parent, label);
+ }
+ }
+
+ cell->axes[a].leaf = prev_leaf;
+ }
+ free (sorted);
+ free (groups);
+ }
+ }
+
+ for (size_t i = 0; i < t->n_sections; i++)
+ {
+ struct ctables_section *s = &t->sections[i];
+
+ struct ctables_cell *cell;
+ HMAP_FOR_EACH (cell, struct ctables_cell, node, &s->cells)
+ {
+ if (cell->hide)
+ continue;
+
+ const struct ctables_nest *specs_nest = s->nests[t->summary_axis];
+ const struct ctables_summary_spec_set *specs = &specs_nest->specs[cell->sv];
+ for (size_t j = 0; j < specs->n; j++)
+ {
+ size_t dindexes[5];
+ size_t n_dindexes = 0;
+
+ if (summary_dimension)
+ dindexes[n_dindexes++] = specs->specs[j].axis_idx;
+
+ if (categories_dimension)
+ {
+ const struct ctables_nest *clabels_nest = s->nests[t->clabels_from_axis];
+ const struct variable *var = clabels_nest->vars[clabels_nest->n - 1];
+ const union value *value = &cell->axes[t->clabels_from_axis].cvs[clabels_nest->n - 1].value;
+ const struct ctables_value *ctv = ctables_value_find (t, value, var_get_width (var));
+ assert (ctv != NULL);
+ dindexes[n_dindexes++] = ctv->leaf;
+ }
+
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ if (d[a])
+ {
+ int leaf = cell->axes[a].leaf;
+ if (a == t->summary_axis && !summary_dimension)
+ leaf += j;
+ dindexes[n_dindexes++] = leaf;
+ }
+
+ double d = ctables_summary_value (cell, &cell->summaries[j], &specs->specs[j]);
+ struct pivot_value *value = pivot_value_new_number (d);
+ value->numeric.format = specs->specs[j].format;
+ pivot_table_put (pt, dindexes, n_dindexes, value);
+ }
+ }
+ }
+
+ pivot_table_submit (pt);
+}
+
+static bool
+ctables_check_label_position (struct ctables_table *t, enum pivot_axis_type a)
+{
+ enum pivot_axis_type label_pos = t->label_axis[a];
+ if (label_pos == a)
+ return true;
+
+ t->clabels_from_axis = a;
+
+ const char *subcommand_name = a == PIVOT_AXIS_ROW ? "ROWLABELS" : "COLLABELS";
+ const char *pos_name = label_pos == PIVOT_AXIS_LAYER ? "LAYER" : "OPPOSITE";
+
+ const struct ctables_stack *stack = &t->stacks[a];
+ if (!stack->n)
+ return true;
+
+ const struct ctables_nest *n0 = &stack->nests[0];
+ assert (n0->n > 0);
+ const struct variable *v0 = n0->vars[n0->n - 1];
+ struct ctables_categories *c0 = t->categories[var_get_dict_index (v0)];
+ t->clabels_example = v0;
+
+ for (size_t i = 0; i < c0->n_cats; i++)
+ if (c0->cats[i].type == CCT_FUNCTION)
+ {
+ msg (SE, _("%s=%s is not allowed with sorting based "
+ "on a summary function."),
+ subcommand_name, pos_name);
+ return false;
+ }
+ if (n0->n - 1 == n0->scale_idx)
+ {
+ msg (SE, _("%s=%s requires the variables to be moved to be categorical, "
+ "but %s is a scale variable."),
+ subcommand_name, pos_name, var_get_name (v0));
+ return false;
+ }
+
+ for (size_t i = 1; i < stack->n; i++)
+ {
+ const struct ctables_nest *ni = &stack->nests[i];
+ assert (ni->n > 0);
+ const struct variable *vi = ni->vars[ni->n - 1];
+ struct ctables_categories *ci = t->categories[var_get_dict_index (vi)];
+
+ if (ni->n - 1 == ni->scale_idx)
+ {
+ msg (SE, _("%s=%s requires the variables to be moved to be "
+ "categorical, but %s is a scale variable."),
+ subcommand_name, pos_name, var_get_name (vi));
+ return false;
+ }
+ if (var_get_width (v0) != var_get_width (vi))
+ {
+ msg (SE, _("%s=%s requires the variables to be "
+ "moved to have the same width, but %s has "
+ "width %d and %s has width %d."),
+ subcommand_name, pos_name,
+ var_get_name (v0), var_get_width (v0),
+ var_get_name (vi), var_get_width (vi));
+ return false;
+ }
+ if (!val_labs_equal (var_get_value_labels (v0),
+ var_get_value_labels (vi)))
+ {
+ msg (SE, _("%s=%s requires the variables to be "
+ "moved to have the same value labels, but %s "
+ "and %s have different value labels."),
+ subcommand_name, pos_name,
+ var_get_name (v0), var_get_name (vi));
+ return false;
+ }
+ if (!ctables_categories_equal (c0, ci))
+ {
+ msg (SE, _("%s=%s requires the variables to be "
+ "moved to have the same category "
+ "specifications, but %s and %s have different "
+ "category specifications."),
+ subcommand_name, pos_name,
+ var_get_name (v0), var_get_name (vi));
+ return false;
+ }
+ }
+
+ return true;
+}
+
+static bool
+ctables_prepare_table (struct ctables_table *t)
+{
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ if (t->axes[a])
+ {
+ t->stacks[a] = enumerate_fts (a, t->axes[a]);
+
+ for (size_t j = 0; j < t->stacks[a].n; j++)
+ {
+ struct ctables_nest *nest = &t->stacks[a].nests[j];
+ for (enum ctables_domain_type dt = 0; dt < N_CTDTS; dt++)
+ {
+ nest->domains[dt] = xmalloc (nest->n * sizeof *nest->domains[dt]);
+ nest->n_domains[dt] = 0;
+
+ for (size_t k = 0; k < nest->n; k++)
+ {
+ if (k == nest->scale_idx)
+ continue;
+
+ switch (dt)
+ {
+ case CTDT_TABLE:
+ continue;
+
+ case CTDT_LAYER:
+ if (a != PIVOT_AXIS_LAYER)
+ continue;
+ break;
+
+ case CTDT_SUBTABLE:
+ case CTDT_ROW:
+ case CTDT_COL:
+ if (dt == CTDT_SUBTABLE ? a != PIVOT_AXIS_LAYER
+ : dt == CTDT_ROW ? a == PIVOT_AXIS_COLUMN
+ : a == PIVOT_AXIS_ROW)
+ {
+ if (k == nest->n - 1
+ || (nest->scale_idx == nest->n - 1
+ && k == nest->n - 2))
+ continue;
+ }
+ break;
+
+ case CTDT_LAYERROW:
+ if (a == PIVOT_AXIS_COLUMN)
+ continue;
+ break;
+
+ case CTDT_LAYERCOL:
+ if (a == PIVOT_AXIS_ROW)
+ continue;
+ break;
+ }
+
+ nest->domains[dt][nest->n_domains[dt]++] = k;
+ }
+ }
+ }
+ }
+ else
+ {
+ struct ctables_nest *nest = xmalloc (sizeof *nest);
+ *nest = (struct ctables_nest) { .n = 0 };
+ t->stacks[a] = (struct ctables_stack) { .nests = nest, .n = 1 };
+ }
+
+ struct ctables_stack *stack = &t->stacks[t->summary_axis];
+ for (size_t i = 0; i < stack->n; i++)
+ {
+ struct ctables_nest *nest = &stack->nests[i];
+ if (!nest->specs[CSV_CELL].n)
+ {
+ struct ctables_summary_spec_set *specs = &nest->specs[CSV_CELL];
+ specs->specs = xmalloc (sizeof *specs->specs);
+ specs->n = 1;
+
+ enum ctables_summary_function function
+ = specs->var ? CTSF_MEAN : CTSF_COUNT;
+ struct ctables_var var = { .is_mrset = false, .var = specs->var };
+
+ *specs->specs = (struct ctables_summary_spec) {
+ .function = function,
+ .format = ctables_summary_default_format (function, &var),
+ .label = ctables_summary_default_label (function, 0),
+ };
+ if (!specs->var)
+ specs->var = nest->vars[0];
+
+ ctables_summary_spec_set_clone (&nest->specs[CSV_TOTAL],
+ &nest->specs[CSV_CELL]);
+ }
+ else if (!nest->specs[CSV_TOTAL].n)
+ ctables_summary_spec_set_clone (&nest->specs[CSV_TOTAL],
+ &nest->specs[CSV_CELL]);
+ }
+
+ struct ctables_summary_spec_set *merged = &t->summary_specs;
+ struct merge_item *items = xnmalloc (2 * stack->n, sizeof *items);
+ size_t n_left = 0;
+ for (size_t j = 0; j < stack->n; j++)
+ {
+ const struct ctables_nest *nest = &stack->nests[j];
+ if (nest->n)
+ for (enum ctables_summary_variant sv = 0; sv < N_CSVS; sv++)
+ items[n_left++] = (struct merge_item) { .set = &nest->specs[sv] };
+ }
+
+ while (n_left > 0)
+ {
+ struct merge_item min = items[0];
+ for (size_t j = 1; j < n_left; j++)
+ if (merge_item_compare_3way (&items[j], &min) < 0)
+ min = items[j];
+
+ if (merged->n >= merged->allocated)
+ merged->specs = x2nrealloc (merged->specs, &merged->allocated,
+ sizeof *merged->specs);
+ merged->specs[merged->n++] = min.set->specs[min.ofs];
+
+ for (size_t j = 0; j < n_left; )
+ {
+ if (merge_item_compare_3way (&items[j], &min) == 0)
+ {
+ struct merge_item *item = &items[j];
+ item->set->specs[item->ofs].axis_idx = merged->n - 1;
+ if (++item->ofs >= item->set->n)
+ {
+ items[j] = items[--n_left];
+ continue;
+ }
+ }
+ j++;
+ }
+ }
+
+#if 0
+ for (size_t j = 0; j < merged->n; j++)
+ printf ("%s\n", ctables_summary_function_name (merged->specs[j].function));
+
+ for (size_t j = 0; j < stack->n; j++)
+ {
+ const struct ctables_nest *nest = &stack->nests[j];
+ for (enum ctables_summary_variant sv = 0; sv < N_CSVS; sv++)
+ {
+ const struct ctables_summary_spec_set *specs = &nest->specs[sv];
+ for (size_t k = 0; k < specs->n; k++)
+ printf ("(%s, %zu) ", ctables_summary_function_name (specs->specs[k].function),
+ specs->specs[k].axis_idx);
+ printf ("\n");
+ }
+ }
+#endif
+
+ return (ctables_check_label_position (t, PIVOT_AXIS_ROW)
+ && ctables_check_label_position (t, PIVOT_AXIS_COLUMN));
+}
+
+static void
+ctables_insert_clabels_values (struct ctables_table *t, const struct ccase *c,
+ enum pivot_axis_type a)
+{
+ struct ctables_stack *stack = &t->stacks[a];
+ for (size_t i = 0; i < stack->n; i++)
+ {
+ const struct ctables_nest *nest = &stack->nests[i];
+ const struct variable *var = nest->vars[nest->n - 1];
+ int width = var_get_width (var);
+ const union value *value = case_data (c, var);
+
+ if (var_is_numeric (var) && value->f == SYSMIS)
+ continue;
+
+ if (!ctables_categories_match (t->categories [var_get_dict_index (var)],
+ value, var))
+ continue;
+
+ unsigned int hash = value_hash (value, width, 0);
+
+ struct ctables_value *clv = ctables_value_find__ (t, value, width, hash);
+ if (!clv)
+ {
+ clv = xmalloc (sizeof *clv);
+ value_clone (&clv->value, value, width);
+ hmap_insert (&t->clabels_values_map, &clv->node, hash);
+ }
+ }
+}
+
+static int
+compare_clabels_values_3way (const void *a_, const void *b_, const void *width_)
+{
+ const struct ctables_value *const *ap = a_;
+ const struct ctables_value *const *bp = b_;
+ const struct ctables_value *a = *ap;
+ const struct ctables_value *b = *bp;
+ const int *width = width_;
+ return value_compare_3way (&a->value, &b->value, *width);
+}
+
+static void
+ctables_sort_clabels_values (struct ctables_table *t)
+{
+ int width = var_get_width (t->clabels_example);
+
+ size_t n = hmap_count (&t->clabels_values_map);
+ t->clabels_values = xnmalloc (n, sizeof *t->clabels_values);
+
+ struct ctables_value *clv;
+ size_t i = 0;
+ HMAP_FOR_EACH (clv, struct ctables_value, node, &t->clabels_values_map)
+ t->clabels_values[i++] = clv;
+ t->n_clabels_values = n;
+ assert (i == n);
+
+ sort (t->clabels_values, n, sizeof *t->clabels_values,
+ compare_clabels_values_3way, &width);
+
+ for (size_t i = 0; i < n; i++)
+ t->clabels_values[i]->leaf = i;
+}
+
+static void
+ctables_add_category_occurrences (const struct variable *var,
+ struct hmap *occurrences,
+ const struct ctables_categories *cats)
+{
+ const struct val_labs *val_labs = var_get_value_labels (var);
+
+ for (size_t i = 0; i < cats->n_cats; i++)
+ {
+ const struct ctables_category *c = &cats->cats[i];
+ switch (c->type)
+ {
+ case CCT_NUMBER:
+ ctables_add_occurrence (var, &(const union value) { .f = c->number },
+ occurrences);
+ break;
+
+ case CCT_STRING:
+ abort (); /* XXX */
+
+ case CCT_RANGE:
+ assert (var_is_numeric (var));
+ for (const struct val_lab *vl = val_labs_first (val_labs); vl;
+ vl = val_labs_next (val_labs, vl))
+ if (vl->value.f >= c->range[0] && vl->value.f <= c->range[1])
+ ctables_add_occurrence (var, &vl->value, occurrences);
+ break;
+
+ case CCT_MISSING:
+ for (const struct val_lab *vl = val_labs_first (val_labs); vl;
+ vl = val_labs_next (val_labs, vl))
+ if (var_is_value_missing (var, &vl->value))
+ ctables_add_occurrence (var, &vl->value, occurrences);
+ break;
+
+ case CCT_OTHERNM:
+ for (const struct val_lab *vl = val_labs_first (val_labs); vl;
+ vl = val_labs_next (val_labs, vl))
+ ctables_add_occurrence (var, &vl->value, occurrences);
+ break;
+
+ case CCT_SUBTOTAL:
+ case CCT_HSUBTOTAL:
+ case CCT_TOTAL:
+ break;
+
+ case CCT_VALUE:
+ case CCT_LABEL:
+ case CCT_FUNCTION:
+ for (const struct val_lab *vl = val_labs_first (val_labs); vl;
+ vl = val_labs_next (val_labs, vl))
+ if (c->include_missing || !var_is_value_missing (var, &vl->value))
+ ctables_add_occurrence (var, &vl->value, occurrences);
+ break;
+ }
+ }
+}
+
+static void
+ctables_section_recurse_add_empty_categories (
+ struct ctables_section *s,
+ const struct ctables_category *cats[PIVOT_N_AXES][10], struct ccase *c,
+ enum pivot_axis_type a, size_t a_idx)
+{
+ if (a >= PIVOT_N_AXES)
+ ctables_cell_insert__ (s, c, cats);
+ else if (!s->nests[a] || a_idx >= s->nests[a]->n)
+ ctables_section_recurse_add_empty_categories (s, cats, c, a + 1, 0);
+ else
+ {
+ const struct variable *var = s->nests[a]->vars[a_idx];
+ int width = var_get_width (var);
+ const struct hmap *occurrences = &s->occurrences[a][a_idx];
+ const struct ctables_section_value *sv;
+ HMAP_FOR_EACH (sv, struct ctables_section_value, node, occurrences)
+ {
+ union value *value = case_data_rw (c, var);
+ value_destroy (value, width);
+ value_clone (value, &sv->value, width);
+ cats[a][a_idx] = ctables_categories_match (
+ s->table->categories[var_get_dict_index (var)], value, var);
+ assert (cats[a][a_idx] != NULL);
+ ctables_section_recurse_add_empty_categories (s, cats, c, a, a_idx + 1);
+ }
+ }
+}
+
+static void
+ctables_section_add_empty_categories (struct ctables_section *s)
+{
+ bool show_empty = false;
+ for (size_t a = 0; a < PIVOT_N_AXES; a++)
+ if (s->nests[a])
+ for (size_t k = 0; k < s->nests[a]->n; k++)
+ if (k != s->nests[a]->scale_idx)
+ {
+ const struct variable *var = s->nests[a]->vars[k];
+ const struct ctables_categories *cats = s->table->categories[
+ var_get_dict_index (var)];
+ if (cats->show_empty)
+ {
+ show_empty = true;
+ ctables_add_category_occurrences (var, &s->occurrences[a][k], cats);
+ }
+ }
+ if (!show_empty)
+ return;
+
+ const struct ctables_category *cats[PIVOT_N_AXES][10]; /* XXX */
+ struct ccase *c = case_create (dict_get_proto (s->table->ctables->dict));
+ ctables_section_recurse_add_empty_categories (s, cats, c, 0, 0);
+ case_unref (c);
+}
+
+static bool
+ctables_execute (struct dataset *ds, struct ctables *ct)
+{
+ for (size_t i = 0; i < ct->n_tables; i++)
+ {
+ struct ctables_table *t = ct->tables[i];
+ t->sections = xnmalloc (MAX (1, t->stacks[PIVOT_AXIS_ROW].n) *
+ MAX (1, t->stacks[PIVOT_AXIS_COLUMN].n) *
+ MAX (1, t->stacks[PIVOT_AXIS_LAYER].n),
+ sizeof *t->sections);
+ size_t ix[PIVOT_N_AXES];
+ ctables_table_add_section (t, 0, ix);
+ }
+
+ struct casereader *input = proc_open (ds);
+ bool warn_on_invalid = true;
+ for (struct ccase *c = casereader_read (input); c;
+ case_unref (c), c = casereader_read (input))
+ {
+ double d_weight = dict_get_case_weight (dataset_dict (ds), c,
+ &warn_on_invalid);
+ double e_weight = (ct->e_weight
+ ? var_force_valid_weight (ct->e_weight,
+ case_num (c, ct->e_weight),
+ &warn_on_invalid)
+ : d_weight);
+
+ for (size_t i = 0; i < ct->n_tables; i++)
+ {
+ struct ctables_table *t = ct->tables[i];
+
+ for (size_t j = 0; j < t->n_sections; j++)
+ ctables_cell_insert (&t->sections[j], c, d_weight, e_weight);
+
+ for (enum pivot_axis_type a = 0; a < PIVOT_N_AXES; a++)
+ if (t->label_axis[a] != a)
+ ctables_insert_clabels_values (t, c, a);
+ }
+ }
+ casereader_destroy (input);
+
+ for (size_t i = 0; i < ct->n_tables; i++)
+ {
+ struct ctables_table *t = ct->tables[i];
+
+ if (t->clabels_example)
+ ctables_sort_clabels_values (t);
+
+ for (size_t j = 0; j < t->n_sections; j++)
+ ctables_section_add_empty_categories (&t->sections[j]);
+
+ ctables_table_output (ct, ct->tables[i]);
+ }
+ return proc_commit (ds);
+}
+\f
+/* Postcomputes. */
+
+typedef struct ctables_pcexpr *parse_recursively_func (struct lexer *);
+
+static void
+ctables_pcexpr_destroy (struct ctables_pcexpr *e)
+{
+ if (e)
+ {
+ switch (e->op)
+ {
+ case CTPO_CAT_STRING:
+ free (e->string);
+ break;
+
+ case CTPO_ADD:
+ case CTPO_SUB:
+ case CTPO_MUL:
+ case CTPO_DIV:
+ case CTPO_POW:
+ case CTPO_NEG:
+ for (size_t i = 0; i < 2; i++)
+ ctables_pcexpr_destroy (e->subs[i]);
+ break;
+
+ case CTPO_CONSTANT:
+ case CTPO_CAT_NUMBER:
+ case CTPO_CAT_RANGE:
+ case CTPO_CAT_MISSING:
+ case CTPO_CAT_OTHERNM:
+ case CTPO_CAT_SUBTOTAL:
+ case CTPO_CAT_TOTAL:
+ break;
+ }
+
+ msg_location_destroy (e->location);
+ free (e);
+ }
+}
+
+static struct ctables_pcexpr *
+ctables_pcexpr_allocate_binary (enum ctables_postcompute_op op,
+ struct ctables_pcexpr *sub0,
+ struct ctables_pcexpr *sub1)
+{
+ struct ctables_pcexpr *e = xmalloc (sizeof *e);
+ *e = (struct ctables_pcexpr) {
+ .op = op,
+ .subs = { sub0, sub1 },
+ .ofs = { sub0->ofs[0], sub1->ofs[1] }
+ };
+ return e;
+}
+
+static struct msg_location *
+ctables_pcexpr_location (struct lexer *lexer, const struct ctables_pcexpr *e_)
+{
+ if (!e_->location)
+ {
+ struct ctables_pcexpr *e = CONST_CAST (struct ctables_pcexpr *, e_);
+ e->location = lex_ofs_location (lexer, e->ofs[0], e->ofs[1]);
+ }
+ return e_->location;
+}
+
+/* How to parse an operator. */
+struct operator
+ {
+ enum token_type token;
+ enum ctables_postcompute_op op;
+ };
+
+static const struct operator *
+match_operator (struct lexer *lexer, const struct operator ops[], size_t n_ops)
+{
+ for (const struct operator *op = ops; op < ops + n_ops; op++)
+ if (lex_token (lexer) == op->token)
+ {
+ if (op->token != T_NEG_NUM)
+ lex_get (lexer);
+
+ return op;
+ }
+
+ return NULL;
+}
+
+static struct ctables_pcexpr *
+parse_binary_operators__ (struct lexer *lexer,
+ const struct operator ops[], size_t n_ops,
+ parse_recursively_func *parse_next_level,
+ const char *chain_warning,
+ struct ctables_pcexpr *lhs)
+{
+ for (int op_count = 0; ; op_count++)
+ {
+ const struct operator *op = match_operator (lexer, ops, n_ops);
+ if (!op)
+ {
+ if (op_count > 1 && chain_warning)
+ msg_at (SW, ctables_pcexpr_location (lexer, lhs),
+ "%s", chain_warning);
+
+ return lhs;
+ }
+
+ struct ctables_pcexpr *rhs = parse_next_level (lexer);
+ if (!rhs)
+ {
+ ctables_pcexpr_destroy (lhs);
+ return NULL;
+ }
+
+ lhs = ctables_pcexpr_allocate_binary (op->op, lhs, rhs);
+ }
+}
+
+static struct ctables_pcexpr *
+parse_binary_operators (struct lexer *lexer,
+ const struct operator ops[], size_t n_ops,
+ parse_recursively_func *parse_next_level,
+ const char *chain_warning)
+{
+ struct ctables_pcexpr *lhs = parse_next_level (lexer);
+ if (!lhs)
+ return NULL;
+
+ return parse_binary_operators__ (lexer, ops, n_ops, parse_next_level,
+ chain_warning, lhs);
+}
+
+static struct ctables_pcexpr *parse_add (struct lexer *);
+
+static struct ctables_pcexpr
+ctpo_cat_range (double low, double high)
+{
+ return (struct ctables_pcexpr) {
+ .op = CTPO_CAT_RANGE,
+ .range = { low, high },
+ };
+}
+
+static struct ctables_pcexpr *
+parse_primary (struct lexer *lexer)
+{
+ int start_ofs = lex_ofs (lexer);
+ struct ctables_pcexpr e;
+ if (lex_is_number (lexer))
+ {
+ e = (struct ctables_pcexpr) { .op = CTPO_CONSTANT,
+ .number = lex_number (lexer) };
+ lex_get (lexer);
+ }
+ else if (lex_match_id (lexer, "MISSING"))
+ e = (struct ctables_pcexpr) { .op = CTPO_CAT_MISSING };
+ else if (lex_match_id (lexer, "OTHERNM"))
+ e = (struct ctables_pcexpr) { .op = CTPO_CAT_OTHERNM };
+ else if (lex_match_id (lexer, "TOTAL"))
+ e = (struct ctables_pcexpr) { .op = CTPO_CAT_TOTAL };
+ else if (lex_match_id (lexer, "SUBTOTAL"))
+ {
+ size_t subtotal_index = 0;
+ if (lex_match (lexer, T_LBRACK))
+ {
+ if (!lex_force_int_range (lexer, "SUBTOTAL", 1, LONG_MAX))
+ return NULL;
+ subtotal_index = lex_integer (lexer);
+ lex_get (lexer);
+ if (!lex_force_match (lexer, T_RBRACK))
+ return NULL;
+ }
+ e = (struct ctables_pcexpr) { .op = CTPO_CAT_SUBTOTAL,
+ .subtotal_index = subtotal_index };
+ }
+ else if (lex_match (lexer, T_LBRACK))
+ {
+ if (lex_match_id (lexer, "LO"))
+ {
+ if (!lex_force_match_id (lexer, "THRU") || lex_force_num (lexer))
+ return false;
+ e = ctpo_cat_range (-DBL_MAX, lex_number (lexer));
+ lex_get (lexer);
+ }
+ else if (lex_is_number (lexer))
+ {
+ double number = lex_number (lexer);
+ lex_get (lexer);
+ if (lex_match_id (lexer, "THRU"))
+ {
+ if (lex_match_id (lexer, "HI"))
+ e = ctpo_cat_range (number, DBL_MAX);
+ else
+ {
+ if (!lex_force_num (lexer))
+ return false;
+ e = ctpo_cat_range (number, lex_number (lexer));
+ lex_get (lexer);
+ }
+ }
+ else
+ e = (struct ctables_pcexpr) { .op = CTPO_CAT_NUMBER,
+ .number = number };
+ }
+ else if (lex_is_string (lexer))
+ {
+ e = (struct ctables_pcexpr) {
+ .op = CTPO_CAT_STRING,
+ .string = ss_xstrdup (lex_tokss (lexer)),
+ };
+ lex_get (lexer);
+ }
+ else
+ {
+ lex_error (lexer, NULL);
+ return NULL;
+ }
+
+ if (!lex_force_match (lexer, T_RBRACK))
+ {
+ if (e.op == CTPO_CAT_STRING)
+ free (e.string);
+ return NULL;
+ }
+ }
+ else if (lex_match (lexer, T_LPAREN))
+ {
+ struct ctables_pcexpr *ep = parse_add (lexer);
+ if (!ep)
+ return NULL;
+ if (!lex_force_match (lexer, T_RPAREN))
+ {
+ ctables_pcexpr_destroy (ep);
+ return NULL;
+ }
+ return ep;
+ }
+ else
+ {
+ lex_error (lexer, NULL);
+ return NULL;
+ }
+
+ e.ofs[0] = start_ofs;
+ e.ofs[1] = lex_ofs (lexer) - 1;
+ return xmemdup (&e, sizeof e);
+}
+
+static struct ctables_pcexpr *
+ctables_pcexpr_allocate_neg (struct ctables_pcexpr *sub,
+ struct lexer *lexer, int start_ofs)
+{
+ struct ctables_pcexpr *e = xmalloc (sizeof *e);
+ *e = (struct ctables_pcexpr) {
+ .op = CTPO_NEG,
+ .subs = { sub },
+ .ofs = { start_ofs, lex_ofs (lexer) - 1 },
+ };
+ return e;
+}
+
+static struct ctables_pcexpr *
+parse_exp (struct lexer *lexer)
+{
+ static const struct operator op = { T_EXP, CTPO_POW };
+
+ const char *chain_warning =
+ _("The exponentiation operator (`**') is left-associative: "
+ "`a**b**c' equals `(a**b)**c', not `a**(b**c)'. "
+ "To disable this warning, insert parentheses.");
+
+ if (lex_token (lexer) != T_NEG_NUM || lex_next_token (lexer, 1) != T_EXP)
+ return parse_binary_operators (lexer, &op, 1,
+ parse_primary, chain_warning);
+
+ /* Special case for situations like "-5**6", which must be parsed as
+ -(5**6). */
+
+ int start_ofs = lex_ofs (lexer);
+ struct ctables_pcexpr *lhs = xmalloc (sizeof *lhs);
+ *lhs = (struct ctables_pcexpr) {
+ .op = CTPO_CONSTANT,
+ .number = -lex_tokval (lexer),
+ .ofs = { start_ofs, lex_ofs (lexer) },
+ };
+ lex_get (lexer);
+
+ struct ctables_pcexpr *node = parse_binary_operators__ (
+ lexer, &op, 1, parse_primary, chain_warning, lhs);
+ if (!node)
+ return NULL;
+
+ return ctables_pcexpr_allocate_neg (node, lexer, start_ofs);
+}
+
+/* Parses the unary minus level. */
+static struct ctables_pcexpr *
+parse_neg (struct lexer *lexer)
+{
+ int start_ofs = lex_ofs (lexer);
+ if (!lex_match (lexer, T_DASH))
+ return parse_exp (lexer);
+
+ struct ctables_pcexpr *inner = parse_neg (lexer);
+ if (!inner)
+ return NULL;
+
+ return ctables_pcexpr_allocate_neg (inner, lexer, start_ofs);
+}
+
+/* Parses the multiplication and division level. */
+static struct ctables_pcexpr *
+parse_mul (struct lexer *lexer)
+{
+ static const struct operator ops[] =
+ {
+ { T_ASTERISK, CTPO_MUL },
+ { T_SLASH, CTPO_DIV },
+ };
+
+ return parse_binary_operators (lexer, ops, sizeof ops / sizeof *ops,
+ parse_neg, NULL);
+}
+
+/* Parses the addition and subtraction level. */
+static struct ctables_pcexpr *
+parse_add (struct lexer *lexer)
+{
+ static const struct operator ops[] =
+ {
+ { T_PLUS, CTPO_ADD },
+ { T_DASH, CTPO_SUB },
+ { T_NEG_NUM, CTPO_ADD },
+ };
+
+ return parse_binary_operators (lexer, ops, sizeof ops / sizeof *ops,
+ parse_mul, NULL);
+}
+
+static struct ctables_postcompute *
+ctables_find_postcompute (struct ctables *ct, const char *name)
+{
+ struct ctables_postcompute *pc;
+ HMAP_FOR_EACH_WITH_HASH (pc, struct ctables_postcompute, hmap_node,
+ utf8_hash_case_string (name, 0), &ct->postcomputes)
+ if (!utf8_strcasecmp (pc->name, name))
+ return pc;
+ return NULL;
+}
+
+static bool
+ctables_parse_pcompute (struct lexer *lexer, struct ctables *ct)
+{
+ int start_ofs = lex_ofs (lexer) - 1;
+
+ if (!lex_force_match (lexer, T_AND) || !lex_force_id (lexer))
+ return false;
+
+ char *name = ss_xstrdup (lex_tokss (lexer));
+
+ lex_get (lexer);
+ if (!lex_force_match (lexer, T_EQUALS)
+ || !lex_force_match_id (lexer, "EXPR")
+ || !lex_force_match (lexer, T_LPAREN))
+ {
+ free (name);
+ return false;
+ }
+
+ struct ctables_pcexpr *expr = parse_add (lexer);
+ if (!expr || !lex_force_match (lexer, T_RPAREN))
+ {
+ free (name);
+ return false;
+ }
+
+ struct msg_location *location = lex_ofs_location (lexer, start_ofs,
+ lex_ofs (lexer) - 1);
+
+ struct ctables_postcompute *pc = ctables_find_postcompute (ct, name);
+ if (pc)
+ {
+ msg_at (SW, location, _("New definition of &%s will override the "
+ "previous definition."),
+ pc->name);
+ msg_at (SN, pc->location, _("This is the previous definition."));
+
+ ctables_pcexpr_destroy (pc->expr);
+ msg_location_destroy (pc->location);
+ free (name);
+ }
+ else
+ {
+ pc = xmalloc (sizeof *pc);
+ *pc = (struct ctables_postcompute) { .name = name };
+ hmap_insert (&ct->postcomputes, &pc->hmap_node,
+ utf8_hash_case_string (pc->name, 0));
+ }
+ pc->expr = expr;
+ pc->location = location;
+ return true;
+}
+
+static bool
+ctables_parse_pproperties_format (struct lexer *lexer,
+ struct ctables_summary_spec_set *sss)
+{
+ *sss = (struct ctables_summary_spec_set) { .n = 0 };
+
+ while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH
+ && !(lex_token (lexer) == T_ID
+ && (lex_id_match (ss_cstr ("LABEL"), lex_tokss (lexer))
+ || lex_id_match (ss_cstr ("HIDESOURCECATS"),
+ lex_tokss (lexer)))))
+ {
+ /* Parse function. */
+ enum ctables_summary_function function;
+ if (!parse_ctables_summary_function (lexer, &function))
+ goto error;
+
+ /* Parse percentile. */
+ double percentile = 0;
+ if (function == CTSF_PTILE)
+ {
+ if (!lex_force_num_range_closed (lexer, "PTILE", 0, 100))
+ goto error;
+ percentile = lex_number (lexer);
+ lex_get (lexer);
+ }
+
+ /* Parse format. */
+ struct fmt_spec format;
+ if (!parse_format_specifier (lexer, &format)
+ || !fmt_check_output (&format)
+ || !fmt_check_type_compat (&format, VAL_NUMERIC))
+ goto error;
+
+ if (sss->n >= sss->allocated)
+ sss->specs = x2nrealloc (sss->specs, &sss->allocated,
+ sizeof *sss->specs);
+ sss->specs[sss->n++] = (struct ctables_summary_spec) {
+ .function = function,
+ .percentile = percentile,
+ .format = format,
+ };
+ }
+ return true;
+
+error:
+ ctables_summary_spec_set_uninit (sss);
+ return false;
+}
+
+static bool
+ctables_parse_pproperties (struct lexer *lexer, struct ctables *ct)
+{
+ struct ctables_postcompute **pcs = NULL;
+ size_t n_pcs = 0;
+ size_t allocated_pcs = 0;
+
+ while (lex_match (lexer, T_AND))
+ {
+ if (!lex_force_id (lexer))
+ goto error;
+ struct ctables_postcompute *pc
+ = ctables_find_postcompute (ct, lex_tokcstr (lexer));
+ if (!pc)
+ {
+ msg (SE, _("Unknown computed category &%s."), lex_tokcstr (lexer));
+ goto error;
+ }
+ lex_get (lexer);
+
+ if (n_pcs >= allocated_pcs)
+ pcs = x2nrealloc (pcs, &allocated_pcs, sizeof *pcs);
+ pcs[n_pcs++] = pc;
+ }
+
+ while (lex_token (lexer) != T_SLASH && lex_token (lexer) != T_ENDCMD)
+ {
+ if (lex_match_id (lexer, "LABEL"))
+ {
+ lex_match (lexer, T_EQUALS);
+ if (!lex_force_string (lexer))
+ goto error;
+
+ for (size_t i = 0; i < n_pcs; i++)
+ {
+ free (pcs[i]->label);
+ pcs[i]->label = ss_xstrdup (lex_tokss (lexer));
+ }
+
+ lex_get (lexer);
+ }
+ else if (lex_match_id (lexer, "FORMAT"))
+ {
+ lex_match (lexer, T_EQUALS);
+
+ struct ctables_summary_spec_set sss;
+ if (!ctables_parse_pproperties_format (lexer, &sss))
+ goto error;
+
+ for (size_t i = 0; i < n_pcs; i++)
+ {
+ if (pcs[i]->specs)
+ ctables_summary_spec_set_uninit (pcs[i]->specs);
+ else
+ pcs[i]->specs = xmalloc (sizeof *pcs[i]->specs);
+ ctables_summary_spec_set_clone (pcs[i]->specs, &sss);
+ }
+ ctables_summary_spec_set_uninit (&sss);
+ }
+ else if (lex_match_id (lexer, "HIDESOURCECATS"))
+ {
+ lex_match (lexer, T_EQUALS);
+ bool hide_source_cats;
+ if (!parse_bool (lexer, &hide_source_cats))
+ goto error;
+ for (size_t i = 0; i < n_pcs; i++)
+ pcs[i]->hide_source_cats = hide_source_cats;
+ }
+ else
+ {
+ lex_error_expecting (lexer, "LABEL", "FORMAT", "HIDESOURCECATS");
+ goto error;
+ }
+ }
+ free (pcs);
+ return true;
+
+error:
+ free (pcs);
+ return false;
+}
+
+int