Regex in replace_query_select is buggy

[lu-pl]

The regex used in rdfproxy.utils.sparql_utils.replace_query_select_clause is completely inappropriate for the task at hand. E.g. it is currently not possible for rdfproxy to deal with query strings containing newlines like

query = """
select ?parent \n ?child ?name
where {
    values (?parent ?child ?name) {
        ('x' 'c' 'foo')
        ('y' 'd' UNDEF)
        ('y' 'e' UNDEF)
        ('z' UNDEF UNDEF)
    }
}
"""

In that case, the current regex select\s.* matches until \n and creates

select (count(*) as ?cnt) ?child ?name
where {
    values (?parent ?child ?name) {
        ('x' 'c' 'foo')
        ('y' 'd' UNDEF)
        ('y' 'e' UNDEF)
        ('z' UNDEF UNDEF)
    }
}

which is invalid SPARQL and leads to the org.openrdf.query.MalformedQueryException: Bad aggregate [...] Java when targeting Blazegraph/Wikidata (or any other Jena-based triplestore).

Solution proposal

A much more apt regex for selecting the entire SELECT clause would be (select\s+[\s\S]*?)(?=\s+where).
This non-greedily matches the SELECT clause with all variable references and performs a lookahead for WHERE.

[lu-pl]

For the above mentioned Blazegraph exception see the SPARQL Specification 'Aggregate Projection Restrictions'

[lu-pl]

Note: It is currently not possible to recognize the above query as invalid in a query sanity check using RDFLib, see #2960

[lu-pl]

Note: A better version of the pattern would be \A(select\s+[\s\S]*?)(?=\s+where) with an absolute anchor.

Since this is used with re.sub which substitutes the "leftmost non-overlapping occurrences of [the] pattern", this doesn't really matter though.