From f85e479caa67613a88490c08a99210629056989c Mon Sep 17 00:00:00 2001
From: Erik Welch <erik.n.welch@gmail.com>
Date: Mon, 12 Oct 2026 03:05:58 -0500
Subject: [PATCH] Reject unholdable weights in from_networkx without asking
 scipy

from_networkx builds its coo arrays directly, but for weights a matrix
cannot hold (strings, objects, ints beyond 64 bits, uniform-length
sequences) it still called nx.to_scipy_sparse_array, only so that scipy
would raise. That made the error depend on scipy: ValueError since 1.15,
a TypeError or a NetworkXError blaming the sparse format before (which
abf6f620 restated as ValueError), and, with scipy not installed, a
ModuleNotFoundError from inside networkx. The fallback never returned a
matrix, so from_networkx otherwise had no use for scipy at all.

Raise the ValueError directly, naming the attribute and what it gave:

    edge weights must be bool, int (at most 64 bits), float or complex
    scalars; attribute 'weight' gives dtype <U1

and drop _from_networkx_via_scipy. The docstring gains a Raises section
for this and for the NetworkXError cases it shares with
nx.to_scipy_sparse_array. Only the message changes for callers with
scipy; callers without it get the ValueError instead of an ImportError.

The rejection tests are folded into one that covers each kind of weight
on all four graph classes, with scipy importable and with it blocked
through sys.modules, so they no longer skip without scipy. The
multigraph scipy-free test blocks scipy the same way instead of
monkeypatching the removed fallback.
---
 graphblas/io/_networkx.py  | 53 ++++++++++++++------------------------
 graphblas/tests/test_io.py | 53 ++++++++++++++++++++++++++------------
 2 files changed, 56 insertions(+), 50 deletions(-)

diff --git a/graphblas/io/_networkx.py b/graphblas/io/_networkx.py
index 3e7cfba6e..5ead709ed 100644
--- a/graphblas/io/_networkx.py
+++ b/graphblas/io/_networkx.py
@@ -1,7 +1,6 @@
 import numpy as np
 
 from ..dtypes import lookup_dtype
-from ._scipy import from_scipy_sparse
 
 
 def from_networkx(G, nodelist=None, dtype=None, weight="weight", name=None):
@@ -24,6 +23,14 @@ def from_networkx(G, nodelist=None, dtype=None, weight="weight", name=None):
     -------
     :class:`~graphblas.Matrix`
 
+    Raises
+    ------
+    networkx.NetworkXError
+        If the graph has no nodes, or ``nodelist`` is empty, repeats a node, or
+        names one the graph does not have (as ``nx.to_scipy_sparse_array`` does).
+    ValueError
+        If a weight is not a bool, int (at most 64 bits), float or complex scalar.
+
     """
     import networkx as nx
 
@@ -81,28 +88,19 @@ def from_networkx(G, nodelist=None, dtype=None, weight="weight", name=None):
         # An explicit dtype= is applied above and never reaches this branch.
         values = values.astype(np.int64)
     if values.ndim != 1 or values.dtype.kind not in "biufc":
-        # Defer to scipy so the error matches the previous behavior exactly.
         # Two kinds of weight land here: non-numeric attributes (object arrays,
-        # but also e.g. all-string weights, which infer a <U dtype rather than
-        # object), and sequence-valued attributes of uniform length, which infer
-        # a 2-D numeric array that from_coo would otherwise accept as a UDT.
-        try:
-            return _from_networkx_via_scipy(G, nodelist, dtype, weight, name)
-        except (nx.NetworkXError, TypeError) as err:
-            # Since scipy 1.15 an unsupported dtype is rejected while the coo array
-            # is built, and that ValueError reaches us unchanged. Older scipy builds
-            # the array and fails in the coo -> csr conversion instead: scipy raises
-            # TypeError ("no supported conversion for types"), which networkx 3.4+
-            # wraps in a NetworkXError that blames the sparse format while
-            # networkx <= 3.3 lets the TypeError propagate.
-            # Restate it so every supported stack reports the same error for the
-            # same graph. The graph and nodelist checks above are the ones
-            # nx.to_scipy_sparse_array makes, so a NetworkXError or TypeError from
-            # the fallback can only be that dtype complaint.
-            raise ValueError(
-                f"scipy.sparse does not support dtype {values.dtype}; "
-                "edge weights must be numeric scalars"
-            ) from err
+        # which ints beyond 64 bits also give, and e.g. all-string weights, which
+        # infer a <U dtype rather than object), and sequence-valued attributes of
+        # uniform length, which infer a 2-D numeric array that from_coo would
+        # otherwise accept as a UDT. nx.to_scipy_sparse_array rejects the same
+        # weights, but asking it made the error depend on scipy: ValueError since
+        # 1.15, TypeError or a NetworkXError blaming the sparse format before, and
+        # ImportError without scipy installed.
+        got = f"dtype {values.dtype}" if values.ndim == 1 else f"arrays of shape {values.shape[1:]}"
+        raise ValueError(
+            "edge weights must be bool, int (at most 64 bits), float or complex scalars; "
+            f"attribute {weight!r} gives {got}"
+        )
     if values.size == 0:
         # An empty graph has no data to infer a dtype from; scipy defaults an
         # empty coo array to float64, so match that when dtype is unset.
@@ -138,17 +136,6 @@ def from_networkx(G, nodelist=None, dtype=None, weight="weight", name=None):
     return Matrix.from_coo(rows, cols, values, nrows=nlen, ncols=nlen, dup_op=dup_op, name=name)
 
 
-def _from_networkx_via_scipy(G, nodelist, dtype, weight, name):
-    """Fallback path: convert through a scipy sparse array.
-
-    ``dtype`` is already normalized to a numpy type (or None) by the caller.
-    """
-    import networkx as nx
-
-    A = nx.to_scipy_sparse_array(G, nodelist=nodelist, dtype=dtype, weight=weight)
-    return from_scipy_sparse(A, name=name)
-
-
 # A DiGraph spends about 350 bytes on each node, so a graph with this many nodes would
 # need hundreds of terabytes. Refusing up front beats a long climb to out-of-memory,
 # and it never refuses a graph that could have been built.
diff --git a/graphblas/tests/test_io.py b/graphblas/tests/test_io.py
index 26c87107e..065a5deca 100644
--- a/graphblas/tests/test_io.py
+++ b/graphblas/tests/test_io.py
@@ -1,4 +1,6 @@
 import functools
+import re
+import sys
 from io import BytesIO, StringIO
 
 import numpy as np
@@ -320,19 +322,41 @@ def test_from_networkx_int32_weights_widen_like_from_coo():
     assert gb.Matrix.from_coo([0], [1], [np.int16(2)]).dtype == dtypes.INT16
 
 
-@pytest.mark.skipif("not nx or not ss")
+def _block_scipy(monkeypatch):
+    # ``import scipy`` (and ``import scipy.sparse``) raise ImportError while a
+    # sys.modules entry is None; monkeypatch puts the real modules back after.
+    for modname in ("scipy", "scipy.sparse"):
+        monkeypatch.setitem(sys.modules, modname, None)
+
+
+@pytest.mark.skipif("not nx")
 @pytest.mark.parametrize(
     "graph_cls", [nx.Graph, nx.DiGraph, nx.MultiGraph, nx.MultiDiGraph] if nx else []
 )
-def test_from_networkx_rejects_array_weights(graph_cls):
-    # Sequence weights of uniform length infer a 2-D numeric array, which passes
-    # a dtype-kind test but which from_coo would happily read as a UDT. scipy has
-    # always rejected these, so the direct path must defer rather than quietly
-    # widen what from_networkx accepts.
+@pytest.mark.parametrize(
+    ("weights", "got"),
+    [
+        # Sequences of uniform length infer a 2-D numeric array, which passes a
+        # dtype-kind test but which from_coo would happily read as a UDT.
+        ([[1, 2], [3, 4]], "arrays of shape (2,)"),
+        # Strings infer a 1-D <U array, so a shape test alone would let them in.
+        (["a", "b"], "dtype <U1"),
+        ([2**70, 1], "dtype object"),  # beyond 64 bits
+        ([object(), 1], "dtype object"),
+    ],
+)
+@pytest.mark.parametrize("block_scipy", [False, True])
+def test_from_networkx_rejects_weights_a_matrix_cannot_hold(
+    graph_cls, weights, got, block_scipy, monkeypatch
+):
+    # The same ValueError whether or not scipy is installed, and on any scipy:
+    # deferring to nx.to_scipy_sparse_array gave ImportError without it.
+    if block_scipy:
+        _block_scipy(monkeypatch)
     G = graph_cls()
-    G.add_edge(0, 1, weight=[1, 2])
-    G.add_edge(1, 0, weight=[3, 4])
-    with pytest.raises(ValueError, match="must be 1-D"):
+    G.add_edge(0, 1, weight=weights[0])
+    G.add_edge(1, 2, weight=weights[1])
+    with pytest.raises(ValueError, match=re.escape(f"attribute 'weight' gives {got}")):
         gb.io.from_networkx(G)
 
 
@@ -377,14 +401,9 @@ def test_from_networkx_rejects_bad_nodelist():
 
 @pytest.mark.skipif("not nx")
 def test_from_networkx_multigraph_is_scipy_free(monkeypatch):
-    # A numeric-weight multigraph must ingest via the direct coo path, not the
-    # scipy fallback, so networkx ingest no longer requires scipy for this case.
-    import graphblas.io._networkx as _gnx
-
-    def _boom(*args, **kwargs):  # pragma: no cover (only runs if the direct path regresses)
-        raise AssertionError("scipy fallback should not be used for a numeric multigraph")
-
-    monkeypatch.setattr(_gnx, "_from_networkx_via_scipy", _boom)
+    # A numeric-weight multigraph ingests via the direct coo path, so networkx
+    # ingest does not require scipy for it.
+    _block_scipy(monkeypatch)
     G = nx.MultiDiGraph()
     G.add_weighted_edges_from([(0, 1, 2.0), (0, 1, 3.0), (2, 0, 1.0), (1, 1, 4.0), (1, 1, 0.5)])
     M = gb.io.from_networkx(G)