NextGraph
/
oxigraph
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Activity

Implements SPARQL property path evaluation

Browse Source
pull/10/head
Tpt 5 years ago
parent ad4563135a
commit 1cce2d26a5
5 changed files with 484 additions and 90 deletions
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Show Stats Download Patch File Download Diff File
  1. 372
      lib/src/sparql/eval.rs
  2. 1
      lib/src/sparql/mod.rs
  3. 37
      lib/src/sparql/plan.rs
  4. 43
      lib/src/sparql/plan_builder.rs
  5. 5
      lib/tests/sparql_test_cases.rs

372
lib/src/sparql/eval.rs
Unescape View File

@ -8,6 +8,7 @@ use crate::store::StoreConnection;
use crate::Result;
use chrono::prelude::*;
use digest::Digest;
use failure::format_err;
use md5::Md5;
use num_traits::identities::Zero;
use num_traits::FromPrimitive;
@ -26,6 +27,7 @@ use std::collections::BTreeMap;
use std::collections::HashSet;
use std::convert::TryInto;
use std::fmt::Write;
use std::hash::Hash;
use std::iter::Iterator;
use std::iter::{empty, once};
use std::ops::Deref;
@ -121,10 +123,7 @@ impl<'a, S: StoreConnection + 'a> SimpleEvaluator<S> {
predicate,
object,
graph_name,
} => Box::new(
self.eval_plan(&*child, from)
.flat_map(move |tuple| match tuple {
Ok(tuple) => {
} => Box::new(self.eval_plan(&*child, from).flat_map_ok(move |tuple| {
let mut iter = self.dataset.quads_for_pattern(
get_pattern_value(&subject, &tuple),
get_pattern_value(&predicate, &tuple),
@ -183,10 +182,67 @@ impl<'a, S: StoreConnection + 'a> SimpleEvaluator<S> {
Ok(new_tuple)
}));
iter
})),
PlanNode::PathPatternJoin {
child,
subject,
path,
object,
graph_name,
} => Box::new(self.eval_plan(&*child, from).flat_map_ok(move |tuple| {
let input_subject = get_pattern_value(&subject, &tuple);
let input_object = get_pattern_value(&object, &tuple);
let input_graph_name =
if let Some(graph_name) = get_pattern_value(&graph_name, &tuple) {
graph_name
} else {
return Box::new(once(Err(format_err!(
"Unknown graph name is not allowed when evaluating property path"
)))) as EncodedTuplesIterator<'_>;
};
match (input_subject, input_object) {
(Some(input_subject), Some(input_object)) => Box::new(
self.eval_path_from(path, input_subject, input_graph_name)
.filter_map(move |o| match o {
Ok(o) => {
if o == input_object {
Some(Ok(tuple.clone()))
} else {
None
}
}
Err(error) => Box::new(once(Err(error))),
Err(error) => Some(Err(error)),
}),
)
as EncodedTuplesIterator<'_>,
(Some(input_subject), None) => Box::new(
self.eval_path_from(path, input_subject, input_graph_name)
.map(move |o| {
let mut new_tuple = tuple.clone();
put_pattern_value(&object, o?, &mut new_tuple);
Ok(new_tuple)
}),
),
(None, Some(input_object)) => Box::new(
self.eval_path_to(path, input_object, input_graph_name)
.map(move |s| {
let mut new_tuple = tuple.clone();
put_pattern_value(&subject, s?, &mut new_tuple);
Ok(new_tuple)
}),
),
(None, None) => {
Box::new(self.eval_open_path(path, input_graph_name).map(move |so| {
let mut new_tuple = tuple.clone();
so.map(move |(s, o)| {
put_pattern_value(&subject, s, &mut new_tuple);
put_pattern_value(&object, o, &mut new_tuple);
new_tuple
})
}))
}
}
})),
PlanNode::Join { left, right } => {
//TODO: very dumb implementation
let left_iter = self.eval_plan(&*left, from.clone());
@ -311,9 +367,7 @@ impl<'a, S: StoreConnection + 'a> SimpleEvaluator<S> {
Box::new(errors.into_iter().chain(values.into_iter().map(Ok)))
}
PlanNode::HashDeduplicate { child } => {
let iter = self.eval_plan(&*child, from);
let already_seen = HashSet::with_capacity(iter.size_hint().0);
Box::new(HashDeduplicateIterator { iter, already_seen })
Box::new(hash_deduplicate(self.eval_plan(&*child, from)))
}
PlanNode::Skip { child, count } => Box::new(self.eval_plan(&*child, from).skip(*count)),
PlanNode::Limit { child, count } => {
@ -332,6 +386,171 @@ impl<'a, S: StoreConnection + 'a> SimpleEvaluator<S> {
}
}
fn eval_path_from<'b>(
&'b self,
path: &'b PlanPropertyPath,
start: EncodedTerm,
graph_name: EncodedTerm,
) -> Box<dyn Iterator<Item = Result<EncodedTerm>> + 'b>
where
'a: 'b,
{
match path {
PlanPropertyPath::PredicatePath(p) => Box::new(
self.dataset
.quads_for_pattern(Some(start), Some(*p), None, Some(graph_name))
.map(|t| Ok(t?.object)),
),
PlanPropertyPath::InversePath(p) => self.eval_path_to(&p, start, graph_name),
PlanPropertyPath::SequencePath(a, b) => Box::new(
self.eval_path_from(&a, start, graph_name)
.flat_map_ok(move |middle| self.eval_path_from(&b, middle, graph_name)),
),
PlanPropertyPath::AlternativePath(a, b) => Box::new(
self.eval_path_from(&a, start, graph_name)
.chain(self.eval_path_from(&b, start, graph_name)),
),
PlanPropertyPath::ZeroOrMorePath(p) => {
Box::new(transitive_closure(Some(Ok(start)), move |e| {
self.eval_path_from(p, e, graph_name)
}))
}
PlanPropertyPath::OneOrMorePath(p) => Box::new(transitive_closure(
self.eval_path_from(p, start, graph_name),
move |e| self.eval_path_from(p, e, graph_name),
)),
PlanPropertyPath::ZeroOrOnePath(p) => Box::new(hash_deduplicate(
once(Ok(start)).chain(self.eval_path_from(&p, start, graph_name)),
)),
PlanPropertyPath::NegatedPropertySet(ps) => Box::new(
self.dataset
.quads_for_pattern(Some(start), None, None, Some(graph_name))
.filter(move |t| match t {
Ok(t) => !ps.contains(&t.predicate),
Err(_) => true,
})
.map(|t| Ok(t?.object)),
),
}
}
fn eval_path_to<'b>(
&'b self,
path: &'b PlanPropertyPath,
end: EncodedTerm,
graph_name: EncodedTerm,
) -> Box<dyn Iterator<Item = Result<EncodedTerm>> + 'b>
where
'a: 'b,
{
match path {
PlanPropertyPath::PredicatePath(p) => Box::new(
self.dataset
.quads_for_pattern(None, Some(*p), Some(end), Some(graph_name))
.map(|t| Ok(t?.subject)),
),
PlanPropertyPath::InversePath(p) => self.eval_path_from(&p, end, graph_name),
PlanPropertyPath::SequencePath(a, b) => Box::new(
self.eval_path_to(&b, end, graph_name)
.flat_map_ok(move |middle| self.eval_path_to(&a, middle, graph_name)),
),
PlanPropertyPath::AlternativePath(a, b) => Box::new(
self.eval_path_to(&a, end, graph_name)
.chain(self.eval_path_to(&b, end, graph_name)),
),
PlanPropertyPath::ZeroOrMorePath(p) => {
Box::new(transitive_closure(Some(Ok(end)), move |e| {
self.eval_path_to(p, e, graph_name)
}))
}
PlanPropertyPath::OneOrMorePath(p) => Box::new(transitive_closure(
self.eval_path_to(p, end, graph_name),
move |e| self.eval_path_to(p, e, graph_name),
)),
PlanPropertyPath::ZeroOrOnePath(p) => Box::new(hash_deduplicate(
once(Ok(end)).chain(self.eval_path_to(&p, end, graph_name)),
)),
PlanPropertyPath::NegatedPropertySet(ps) => Box::new(
self.dataset
.quads_for_pattern(None, None, Some(end), Some(graph_name))
.filter(move |t| match t {
Ok(t) => !ps.contains(&t.predicate),
Err(_) => true,
})
.map(|t| Ok(t?.subject)),
),
}
}
fn eval_open_path<'b>(
&'b self,
path: &'b PlanPropertyPath,
graph_name: EncodedTerm,
) -> Box<dyn Iterator<Item = Result<(EncodedTerm, EncodedTerm)>> + 'b>
where
'a: 'b,
{
match path {
PlanPropertyPath::PredicatePath(p) => Box::new(
self.dataset
.quads_for_pattern(None, Some(*p), None, Some(graph_name))
.map(|t| t.map(|t| (t.subject, t.object))),
),
PlanPropertyPath::InversePath(p) => Box::new(
self.eval_open_path(&p, graph_name)
.map(|t| t.map(|(s, o)| (o, s))),
),
PlanPropertyPath::SequencePath(a, b) => Box::new(
self.eval_open_path(&a, graph_name)
.flat_map_ok(move |(start, middle)| {
self.eval_path_from(&b, middle, graph_name)
.map(move |end| Ok((start, end?)))
}),
),
PlanPropertyPath::AlternativePath(a, b) => Box::new(
self.eval_open_path(&a, graph_name)
.chain(self.eval_open_path(&b, graph_name)),
),
PlanPropertyPath::ZeroOrMorePath(p) => Box::new(transitive_closure(
self.get_subject_or_object_identity_pairs(graph_name), //TODO: avoid to inject everything
move |(start, middle)| {
self.eval_path_from(p, middle, graph_name)
.map(move |end| Ok((start, end?)))
},
)),
PlanPropertyPath::OneOrMorePath(p) => Box::new(transitive_closure(
self.eval_open_path(p, graph_name),
move |(start, middle)| {
self.eval_path_from(p, middle, graph_name)
.map(move |end| Ok((start, end?)))
},
)),
PlanPropertyPath::ZeroOrOnePath(p) => Box::new(hash_deduplicate(
self.get_subject_or_object_identity_pairs(graph_name)
.chain(self.eval_open_path(&p, graph_name)),
)),
PlanPropertyPath::NegatedPropertySet(ps) => Box::new(
self.dataset
.quads_for_pattern(None, None, None, Some(graph_name))
.filter(move |t| match t {
Ok(t) => !ps.contains(&t.predicate),
Err(_) => true,
})
.map(|t| t.map(|t| (t.subject, t.object))),
),
}
}
fn get_subject_or_object_identity_pairs<'b>(
&'b self,
graph_name: EncodedTerm,
) -> impl Iterator<Item = Result<(EncodedTerm, EncodedTerm)>> + 'b {
self.dataset
.quads_for_pattern(None, None, None, Some(graph_name))
.flat_map_ok(|t| once(Ok(t.subject)).chain(once(Ok(t.object))))
.map(|e| e.map(|e| (e, e)))
}
fn eval_expression(
&self,
expression: &PlanExpression,
@ -1895,28 +2114,6 @@ impl<'a, S: StoreConnection> Iterator for UnionIterator<'a, S> {
}
}
struct HashDeduplicateIterator<'a> {
iter: EncodedTuplesIterator<'a>,
already_seen: HashSet<EncodedTuple>,
}
impl<'a> Iterator for HashDeduplicateIterator<'a> {
type Item = Result<EncodedTuple>;
fn next(&mut self) -> Option<Result<EncodedTuple>> {
loop {
match self.iter.next()? {
Ok(tuple) => {
if self.already_seen.insert(tuple.clone()) {
return Some(Ok(tuple));
}
}
Err(error) => return Some(Err(error)),
}
}
}
}
struct ConstructIterator<'a, S: StoreConnection> {
eval: &'a SimpleEvaluator<S>,
iter: EncodedTuplesIterator<'a>,
@ -2049,3 +2246,118 @@ impl<T1, T2, I1: Iterator<Item = T1>, I2: Iterator<Item = T2>> Iterator
}
}
}
fn transitive_closure<'a, T: 'a + Copy + Eq + Hash, NI: Iterator<Item = Result<T>> + 'a>(
start: impl IntoIterator<Item = Result<T>>,
next: impl Fn(T) -> NI,
) -> impl Iterator<Item = Result<T>> + 'a {
//TODO: optimize
let mut all = HashSet::<T>::default();
let mut errors = Vec::default();
let mut current = start
.into_iter()
.filter_map(|e| match e {
Ok(e) => {
all.insert(e);
Some(e)
}
Err(error) => {
errors.push(error);
None
}
})
.collect::<Vec<_>>();
while !current.is_empty() {
current = current
.into_iter()
.flat_map(|e| next(e))
.filter_map(|e| match e {
Ok(e) => {
if all.contains(&e) {
None
} else {
all.insert(e);
Some(e)
}
}
Err(error) => {
errors.push(error);
None
}
})
.collect();
}
errors.into_iter().map(Err).chain(all.into_iter().map(Ok))
}
fn hash_deduplicate<T: Eq + Hash + Clone>(
iter: impl Iterator<Item = Result<T>>,
) -> impl Iterator<Item = Result<T>> {
let mut already_seen = HashSet::with_capacity(iter.size_hint().0);
iter.filter(move |e| {
if let Ok(e) = e {
if already_seen.contains(e) {
false
} else {
already_seen.insert(e.clone());
true
}
} else {
true
}
})
}
trait ResultIterator<T>: Iterator<Item = Result<T>> + Sized {
fn flat_map_ok<O, F: FnMut(T) -> U, U: IntoIterator<Item = Result<O>>>(
self,
f: F,
) -> FlatMapOk<T, O, Self, F, U>;
}
impl<T, I: Iterator<Item = Result<T>> + Sized> ResultIterator<T> for I {
fn flat_map_ok<O, F: FnMut(T) -> U, U: IntoIterator<Item = Result<O>>>(
self,
f: F,
) -> FlatMapOk<T, O, Self, F, U> {
FlatMapOk {
inner: self,
f,
current: None,
}
}
}
struct FlatMapOk<
T,
O,
I: Iterator<Item = Result<T>>,
F: FnMut(T) -> U,
U: IntoIterator<Item = Result<O>>,
> {
inner: I,
f: F,
current: Option<U::IntoIter>,
}
impl<T, O, I: Iterator<Item = Result<T>>, F: FnMut(T) -> U, U: IntoIterator<Item = Result<O>>>
Iterator for FlatMapOk<T, O, I, F, U>
{
type Item = Result<O>;
fn next(&mut self) -> Option<Result<O>> {
loop {
if let Some(current) = &mut self.current {
if let Some(next) = current.next() {
return Some(next);
}
}
self.current = None;
match self.inner.next()? {
Ok(e) => self.current = Some((self.f)(e).into_iter()),
Err(error) => return Some(Err(error)),
}
}
}
}

1
lib/src/sparql/mod.rs
Unescape View File

@ -56,6 +56,7 @@ enum SimplePreparedQueryOptions<S: StoreConnection> {
impl<S: StoreConnection> SimplePreparedQuery<S> {
pub(crate) fn new(connection: S, query: &str, base_iri: Option<&str>) -> Result<Self> {
//TODO avoid inserting terms in the Repository StringStore
Ok(Self(match read_sparql_query(query, base_iri)? {
QueryVariants::Select {
algebra,

37
lib/src/sparql/plan.rs
Unescape View File

@ -16,6 +16,13 @@ pub enum PlanNode {
object: PatternValue,
graph_name: PatternValue,
},
PathPatternJoin {
child: Box<PlanNode>,
subject: PatternValue,
path: PlanPropertyPath,
object: PatternValue,
graph_name: PatternValue,
},
Join {
left: Box<PlanNode>,
right: Box<PlanNode>,
@ -103,6 +110,24 @@ impl PlanNode {
}
child.add_variables(set);
}
PlanNode::PathPatternJoin {
child,
subject,
object,
graph_name,
..
} => {
if let PatternValue::Variable(var) = subject {
set.insert(*var);
}
if let PatternValue::Variable(var) = object {
set.insert(*var);
}
if let PatternValue::Variable(var) = graph_name {
set.insert(*var);
}
child.add_variables(set);
}
PlanNode::Filter { child, expression } => {
child.add_variables(set);
expression.add_variables(set);
@ -364,6 +389,18 @@ impl PlanExpression {
}
}
#[derive(Eq, PartialEq, Debug, Clone, Hash)]
pub enum PlanPropertyPath {
PredicatePath(EncodedTerm),
InversePath(Box<PlanPropertyPath>),
SequencePath(Box<PlanPropertyPath>, Box<PlanPropertyPath>),
AlternativePath(Box<PlanPropertyPath>, Box<PlanPropertyPath>),
ZeroOrMorePath(Box<PlanPropertyPath>),
OneOrMorePath(Box<PlanPropertyPath>),
ZeroOrOnePath(Box<PlanPropertyPath>),
NegatedPropertySet(Vec<EncodedTerm>),
}
#[derive(Eq, PartialEq, Debug, Clone, Hash)]
pub enum Comparator {
Asc(PlanExpression),

43
lib/src/sparql/plan_builder.rs
Unescape View File

@ -2,6 +2,7 @@ use crate::model::vocab::xsd;
use crate::model::Literal;
use crate::sparql::algebra::*;
use crate::sparql::model::*;
use crate::sparql::plan::PlanPropertyPath;
use crate::sparql::plan::*;
use crate::store::numeric_encoder::ENCODED_DEFAULT_GRAPH;
use crate::store::StoreConnection;
@ -200,12 +201,52 @@ impl<'a, S: StoreConnection> PlanBuilder<'a, S> {
object: self.pattern_value_from_term_or_variable(&pattern.object, variables)?,
graph_name,
},
TripleOrPathPattern::Path(_pattern) => unimplemented!(),
TripleOrPathPattern::Path(pattern) => PlanNode::PathPatternJoin {
child: Box::new(plan),
subject: self
.pattern_value_from_term_or_variable(&pattern.subject, variables)?,
path: self.build_for_path(&pattern.path)?,
object: self.pattern_value_from_term_or_variable(&pattern.object, variables)?,
graph_name,
},
}
}
Ok(plan)
}
fn build_for_path(&self, path: &PropertyPath) -> Result<PlanPropertyPath> {
Ok(match path {
PropertyPath::PredicatePath(p) => {
PlanPropertyPath::PredicatePath(self.store.encoder().encode_named_node(p)?)
}
PropertyPath::InversePath(p) => {
PlanPropertyPath::InversePath(Box::new(self.build_for_path(p)?))
}
PropertyPath::AlternativePath(a, b) => PlanPropertyPath::AlternativePath(
Box::new(self.build_for_path(a)?),
Box::new(self.build_for_path(b)?),
),
PropertyPath::SequencePath(a, b) => PlanPropertyPath::SequencePath(
Box::new(self.build_for_path(a)?),
Box::new(self.build_for_path(b)?),
),
PropertyPath::ZeroOrMorePath(p) => {
PlanPropertyPath::ZeroOrMorePath(Box::new(self.build_for_path(p)?))
}
PropertyPath::OneOrMorePath(p) => {
PlanPropertyPath::OneOrMorePath(Box::new(self.build_for_path(p)?))
}
PropertyPath::ZeroOrOnePath(p) => {
PlanPropertyPath::ZeroOrOnePath(Box::new(self.build_for_path(p)?))
}
PropertyPath::NegatedPropertySet(p) => PlanPropertyPath::NegatedPropertySet(
p.iter()
.map(|p| self.store.encoder().encode_named_node(p))
.collect::<Result<Vec<_>>>()?,
),
})
}
fn build_for_expression(
&self,
expression: &Expression,

5
lib/tests/sparql_test_cases.rs
Unescape View File

@ -92,6 +92,7 @@ fn sparql_w3c_query_evaluation_testsuite() -> Result<()> {
"http://www.w3.org/2009/sparql/docs/tests/data-sparql11/exists/manifest.ttl",
"http://www.w3.org/2009/sparql/docs/tests/data-sparql11/functions/manifest.ttl",
"http://www.w3.org/2009/sparql/docs/tests/data-sparql11/negation/manifest.ttl",
"http://www.w3.org/2009/sparql/docs/tests/data-sparql11/property-path/manifest.ttl",
];
let test_blacklist = vec![
@ -123,6 +124,8 @@ fn sparql_w3c_query_evaluation_testsuite() -> Result<()> {
NamedNode::parse("http://www.w3.org/2009/sparql/docs/tests/data-sparql11/functions/manifest#bnode01").unwrap(),
//Decimal precision problem
NamedNode::parse("http://www.w3.org/2009/sparql/docs/tests/data-sparql11/functions/manifest#coalesce01").unwrap(),
//Property path with unbound graph name are not supported yet
NamedNode::parse("http://www.w3.org/2009/sparql/docs/tests/data-sparql11/property-path/manifest#pp35").unwrap()
];
let tests: Result<Vec<_>> = manifest_10_urls
@ -165,7 +168,7 @@ fn sparql_w3c_query_evaluation_testsuite() -> Result<()> {
.triples_for_predicate(&rs::INDEX)
.next()
.is_some();
let actual_graph = to_graph(result, with_order)?;
let actual_graph = to_graph(result, with_order).map_err(|e| format_err!("Error constructing result graph for {}: {}", test, e))?;
if actual_graph.is_isomorphic(&expected_graph) {
Ok(())
} else {

Write Preview
Loading…
Cancel
Save