Runtime error querying a dataset using multiple graph variables

[jaw111]

I run into RuntimeError: Set changed size during iteration when iterating over results from a query and was able to reduce the problem down to the following example code that reproduces the error:

import rdflib

# create a dataset
dataset = rdflib.Dataset()

# create a graph
g = dataset.graph(identifier="urn:example:graph")

sparql_query = """
SELECT *
WHERE {
  GRAPH ?g_1 { }
  GRAPH ?g_2 { }
}"""

results = dataset.query(sparql_query)

print(len(results))

The traceback is:

Traceback (most recent call last):
  File "C:\Users\XXXX\query_test.py", line 19, in <module>
    print(len(results))
          ^^^^^^^^^^^^
  File "C:\Python312\Lib\site-packages\rdflib\query.py", line 347, in __len__
    return len(self.bindings)
               ^^^^^^^^^^^^^
  File "C:\Python312\Lib\site-packages\rdflib\query.py", line 242, in bindings
    self._bindings += list(self._genbindings)
                      ^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Python312\Lib\site-packages\rdflib\plugins\sparql\evaluate.py", line 562, in <genexpr>
    return (row.project(project.PV) for row in res)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Python312\Lib\site-packages\rdflib\plugins\sparql\evaluate.py", line 142, in evalLazyJoin
    for a in evalPart(ctx, join.p1):
  File "C:\Python312\Lib\site-packages\rdflib\plugins\sparql\evaluate.py", line 225, in evalGraph
    for graph in ctx.dataset.contexts():
  File "C:\Python312\Lib\site-packages\rdflib\graph.py", line 2529, in contexts
    for c in super(Dataset, self).contexts(triple):
  File "C:\Python312\Lib\site-packages\rdflib\graph.py", line 2154, in contexts
    for context in self.store.contexts(triple):
  File "C:\Python312\Lib\site-packages\rdflib\plugins\stores\memory.py", line 565, in <genexpr>
    return (context for context in self.__all_contexts)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Set changed size during iteration

[jaw111]

This example shows the same error occurs when the dataset contains statements and the query has triple patterns:

import rdflib

# create a dataset
dataset = rdflib.Dataset()

# create a graph
g = dataset.graph(identifier="urn:example:graph")
g.parse(
    data="""
        <urn:example:subject> <urn:example:predicate> <urn:example:object> .
    """,
    format="turtle"
)

sparql_query = """
SELECT *
WHERE {
  GRAPH ?g_1 { ?s ?p ?o }
  GRAPH ?g_2 { ?s ?p ?o }
}"""

results = dataset.query(sparql_query)

print(len(results))

[gjhiggins]

fwiw, I can confirm your experience with the code examples you so usefully provided (thanks!!).

I'm still pondering the full semantics of the change but if you'd like a local workaround until the issue is formally addressed and rectified, the following change to Memory::contexts fixes the issue without causing any test failures so it's probably safe to use:

diff --git a/rdflib/plugins/stores/memory.py b/rdflib/plugins/stores/memory.py
index 8a9a2da3f..b9f2d9c25 100644
--- a/rdflib/plugins/stores/memory.py
+++ b/rdflib/plugins/stores/memory.py
@@ -549,7 +549,7 @@ class Memory(Store):
         self, triple: _TripleType | None = None
     ) -> Generator[_ContextType, None, None]:
         if triple is None or triple == (None, None, None):
-            return (context for context in self.__all_contexts)
+            return list(self.__all_contexts)
 
         subj, pred, obj = triple
         try:

For interest, here's the test I used (taken directly from your code)

def test_b():
    # create a dataset
    dataset = rdflib.Dataset()

    # create a graph
    g = dataset.graph(identifier="urn:example:graph")
    g.parse(
        data="""
            <urn:example:subject> <urn:example:predicate> <urn:example:object> .
        """,
        format="turtle"
    )

    sparql_query = """
    SELECT *
    WHERE {
      GRAPH ?g_1 { ?s ?p ?o }
      GRAPH ?g_2 { ?s ?p ?o }
    }"""

    results = dataset.query(sparql_query)
    jres = json.loads(results.serialize(format="json"))
    print(json.dumps(jres, indent=4))

and here is the result in prettyprinted JSON - exactly as expected from the inputs:

 {
    "results": {
        "bindings": [
            {
                "s": {
                    "type": "uri",
                    "value": "urn:example:subject"
                },
                "p": {
                    "type": "uri",
                    "value": "urn:example:predicate"
                },
                "o": {
                    "type": "uri",
                    "value": "urn:example:object"
                },
                "g_1": {
                    "type": "uri",
                    "value": "urn:example:graph"
                },
                "g_2": {
                    "type": "uri",
                    "value": "urn:example:graph"
                }
            }
        ]
    },
    "head": {
        "vars": [
            "g_2",
            "s",
            "p",
            "o",
            "g_1"
        ]
    }
}