Off-by-one error in dict_put_impl

[MPvHarmelen]

There is an off-by-one error in this if-statement:

virtuoso-opensource/libsrc/Wi/xslt.c

Lines 3589 to 3591 in 4bdeace

	if ((0 < ht->ht_dict_max_entries) &&
	((ht->ht_inserts - ht->ht_deletes) > ht->ht_dict_max_entries) )
	goto skip_insertion; /* see below */

Substituting size for inserts - deletions, this statement reads:

if ((0 < ht->ht_dict_max_entries) && (size > ht->ht_dict_max_entries) )
  goto skip_insertion;  /* see below */

which means that the first time an insertion is skipped, size is strictly larger than ht_dict_max_entries, while size should only be allowed to be at most exactly equal to ht_dict_max_entries.

This causes the following query to return 10001 results, even though ResultSetMaxRows is set to 10000 in virtuoso.ini:

CONSTRUCT {
  ?sub ?pred ?obj .
} WHERE {
  ?sub ?pred ?obj .
} LIMIT 10005

The number of results returned by this query is consistently one larger than the value of ResultSetMaxRows.

In turn, this causes the following code to fail to return the X-SPARQL-MaxRows header in the case of construct queries, because length (rset) is actually one larger than maxrows:

virtuoso-opensource/libsrc/Wi/sparql_io.sql

Lines 3915 to 3916 in 4bdeace

	if (isvector (rset) and length (rset) = maxrows)
	http_header (http_header_get () || sprintf ('X-SPARQL-MaxRows: %d\r\n', maxrows));

[MPvHarmelen]

Unfortunately it turned out not to be that easy. The off-by-one error is a fact, but wasn't causing the x-sparql-maxrows header not being sent. Instead, the results of a construct query were wrapped inside two one-element lists. See b6a89c7.