Merge pull request #1520 from ampli/mlist-nosort

linkage_equiv_p(): Bugfix string comparisons

link-grammar/linkage/analyze-linkage.c

@@ -27,8 +27,10 @@
  * If the GCD is equal to one of them, a pointer to it is returned.
  * Otherwise a new string for the GCD is put in the string set.
  *
- * Note: The head and dependent indicators (lower-case h and d) are
- * ignored, as the intersection cannot include them.
+ * Notes:
+ * 1. The head and dependent indicators (lower-case h and d) are
+ *    ignored, as the intersection cannot include them.
+ * 2. The returned strings is not always in the same string set.
  */
 const char *intersect_strings(String_set *sset, const Connector *c1,
                                      const Connector *c2)

link-grammar/parse/parse.c

@@ -312,10 +312,12 @@ static int linkage_equiv_p(Linkage lpv, Linkage lnx)
 		Link * plk = &lpv->link_array[li];
 		Link * nlk = &lnx->link_array[li];
 
-		// String set guarantees that if the pointer differs,
-		// then the string does too.
-		if (plk->link_name != nlk->link_name)
-			return strcmp(plk->link_name, nlk->link_name);
+		// Note (see intersect_strings()):
+		// link_name is not always in the same string set, so inequality
+		// test cannot be done here.
+		if (plk->link_name == nlk->link_name) continue;
+		int lncmp = strcmp(plk->link_name, nlk->link_name);
+		if (lncmp) return lncmp;
 	}
 
 	// Compare words. The chosen_disjuncts->word_string is the
@@ -338,8 +340,13 @@ static int linkage_equiv_p(Linkage lpv, Linkage lnx)
 			if (NULL == ndj) continue;
 			return 1;
 		}
-		if (pdj->word_string != ndj->word_string)
-			return strcmp(pdj->word_string, ndj->word_string);
+
+		// Note (see build_word_expressions()):
+		// word_string is not always in the same string set, so inequality
+		// test cannot be done here.
+		if (pdj->word_string == ndj->word_string) continue;
+		int wscmp = strcmp(pdj->word_string, ndj->word_string);
+		if (wscmp) return wscmp;
 	}
 
 	// Compare connector types at the link endpoints. If we are here,

link-grammar/tokenize/lookup-exprs.c

@@ -46,13 +46,16 @@ static Dict_node *dictionary_all_categories(Dictionary dict)
 /**
  * build_word_expressions() -- build list of expressions for a word.
  *
- * Looks up a word in the dictionary, fetching from it matching words and their
- * expressions.  Returns NULL if it's not there.  If there, it builds the list
- * of expressions for the word, and returns a pointer to it.
- * The subword of Gword w is used for this lookup, unless the subword is
- * explicitly given as parameter s. The subword of Gword w is always used as
- * the base word for each expression, and its subscript is the one from the
- * dictionary word of the expression.
+ * Look up the subword of \p w in the dictionary, fetching from it
+ * matching words and their expressions. Return NULL if it's not there.
+ * If there, build an X_node list for the word and return it.
+ *
+ * For each dictionary word and its expression, set the X_node string
+ * and exp fields accordingly. However, if a non-NULL \p s is provided,
+ * use it instead of the dictionary word, and as its subscript, use the
+ * subscript of the dictionary word. Note that this means that the
+ * X_node strings (which are later assigned to the disjuncts that are
+ * derived from them) may be from two different string sets.
  */
 static X_node * build_word_expressions(Sentence sent, const Gword *w,
                                        const char *s, Parse_Options opts)