fix(typing): two real defects that CI's type gate cannot see (#1791) (#1792)

Both were found by running the repo's own mypy config against a real
pandas-stubs install. CI's `python-lint-types` cannot see them: `bin/typecheck.sh`
prefers `uvx mypy`, which runs mypy in an isolated ephemeral env with none of the
locked stubs, so every pandas symbol degrades to `Any` (see #1791 for the full
triage — CI's own pinned pandas-stubs 3.0.3.260530 already reports 140 errors).

1. graphviz: `KeyError: None` on a bound-but-unlabelled frame

   `g_to_pgv` asserts `_nodes`/`_edges` are set but not that the *bindings* are.
   `g.nodes(df)` with no `node=` (or `g.edges(df)` with no source/destination)
   leaves them None, and `layout_graphviz`/`render_graphviz` only call
   `materialize_nodes()` when `_nodes` is None — so those graphs reached
   `row[None]` and died with a bare `KeyError: None` from inside the row loop.
   Now resolved once, up front, into non-Optional locals with an actionable
   ValueError. Validation moved ahead of the optional `pygraphviz` import so a
   caller error is not masked by a missing-backend error (and so it is testable
   without the optional dep).

2. search_any: `pat` was bound to BOTH a module and a str in one scope

   `import pandas.api.types as pat` and, 10 lines later, `pat, case = term, ...`.
   Ordering saves it today (the module use strictly precedes the rebind), but it
   is one statement reorder from `AttributeError: 'str' object has no attribute
   'is_bool_dtype'`. Renamed the module alias to `pd_types`.

Also corrects two `dtype: object` params to the engine-agnostic `DType` alias —
these receive pandas/cuDF/polars dtypes, so `object` was both less accurate and
on the repo's banned-annotation list.

mypy differential under CI's own pin (mypy 2.1.0 + pandas-stubs 3.0.3.260530):
140 -> 128, zero new errors, `search_any.py` and `graphviz.py` both to zero.
Under the container stubs (pandas-stubs 3.0.0.260204 + mypy 1.19.1): 173 -> 161.
ruff clean. `graphistry/tests/compute` with `--gpus all`: 9 failed / 7030 passed,
identical to the master baseline (all 9 pre-existing igraph/cudf/dask).


Claude-Session: https://claude.ai/code/session_015YsqAZQLbqjSDrYSFz2GoB

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

Author: lmeyerov

graphistry/compute/gfql/search_any.py

@@ -10,7 +10,7 @@
 import re
 from typing import List, Optional
 
-from graphistry.compute.typing import DataFrameT, SeriesT
+from graphistry.compute.typing import DataFrameT, DType, SeriesT
 
 # inspector's numeric-term gate (streamgl-viz sortAndFilterRowsByQuery.js)
 _NUMERIC_TERM_RE = re.compile(r"^[0-9.\-]+$")
@@ -20,12 +20,12 @@ def is_numeric_term(term: str) -> bool:
     return bool(_NUMERIC_TERM_RE.match(term))
 
 
-def _is_searchable_string_dtype(dtype: object) -> bool:
+def _is_searchable_string_dtype(dtype: DType) -> bool:
     import pandas.api.types as pat  # cuDF mirrors the pandas dtype API
     return bool(pat.is_string_dtype(dtype)) or bool(pat.is_object_dtype(dtype))
 
 
-def _is_int_dtype(dtype: object) -> bool:
+def _is_int_dtype(dtype: DType) -> bool:
     import pandas.api.types as pat
     return bool(pat.is_integer_dtype(dtype))
 
@@ -93,11 +93,14 @@ def search_any_mask(
         # truncate) and temporal is unverified — decline honestly rather than
         # silently mismatch the pandas oracle (wave-3 W3-1). string/int/bool render
         # identically (bool parity is pinned) and stay native.
-        import pandas.api.types as pat
+        # NB: aliased ``pd_types``, not ``pat`` — ``pat`` is rebound below to the search
+        # PATTERN string, and a module/str double-binding in one scope is one statement
+        # reorder away from ``AttributeError: 'str' object has no attribute ...``.
+        import pandas.api.types as pd_types
         for c in cols:
             dt = df[c].dtype
             if not (_is_searchable_string_dtype(dt) or _is_int_dtype(dt)
-                    or bool(pat.is_bool_dtype(dt))):
+                    or bool(pd_types.is_bool_dtype(dt))):
                 raise NotImplementedError(
                     "cuDF searchAny explicit columns support string/int/bool dtypes "
                     "only (float/temporal stringification diverges from pandas); "

graphistry/plugins/graphviz.py

@@ -27,11 +27,31 @@ def g_to_pgv(
     include_positions: bool = False
 ) -> AGraph:
 
-    import pygraphviz as pgv
-
     assert g._nodes is not None
     assert g._edges is not None
 
+    # The id/endpoint BINDINGS are Optional and are not implied by the frames being set:
+    # ``g.nodes(df)`` (no ``node=``) / ``g.edges(df)`` (no source/destination) leave them
+    # None, and ``layout_graphviz``/``render_graphviz`` only ``materialize_nodes()`` when
+    # ``_nodes`` is None -- so an explicitly-bound-but-unlabelled nodes frame reached
+    # ``row[None]`` below and died with a bare ``KeyError: None`` from deep inside the row
+    # loop. Resolve them once, up front, into non-Optional locals.
+    node_col = g._node
+    src_col = g._source
+    dst_col = g._destination
+    if node_col is None:
+        raise ValueError(
+            "graphviz export requires a bound node id column; "
+            "use g.nodes(df, node='id') or g.materialize_nodes()"
+        )
+    if src_col is None or dst_col is None:
+        raise ValueError(
+            "graphviz export requires bound edge endpoint columns; "
+            "use g.edges(df, source='src', destination='dst')"
+        )
+
+    import pygraphviz as pgv
+
     graph = pgv.AGraph(directed=directed, strict=strict)
 
     node_attr_cols: Set[NodeAttr] = {
@@ -58,7 +78,7 @@ def g_to_pgv(
         if pos_cols is not None:
             attrs['pos'] = f"{row[pos_cols[0]]},{row[pos_cols[1]]}"
         graph.add_node(
-            row[g._node],
+            row[node_col],
             **attrs
         )
 
@@ -75,8 +95,8 @@ def g_to_pgv(
     edges_pdf = ensure_pandas(g._edges)
     for _, row in edges_pdf.iterrows():
         graph.add_edge(
-            row[g._source],
-            row[g._destination],
+            row[src_col],
+            row[dst_col],
             **{c: row[c] for c in edge_attr_cols if row[c] is not None}
         )
 

graphistry/tests/plugins/test_graphviz.py

@@ -156,3 +156,28 @@ def test_render_graphviz_bytes(self, tree_g: Plottable) -> None:
 
         svg_bytes = render_graphviz(tree_g, "dot", format="svg", max_nodes=100, max_edges=200)
         assert b'<svg' in svg_bytes
+
+
+class Test_graphviz_bindings:
+    """#1791: the id/endpoint bindings are Optional and are NOT implied by the frames
+    being set. Before the fix these reached ``row[None]`` and raised a bare
+    ``KeyError: None`` from inside the row loop. No pygraphviz needed: the bindings are
+    validated before the optional backend import."""
+
+    def test_unbound_node_column_raises_value_error(self) -> None:
+        # .nodes(df) with no node= leaves _node None while _nodes is set, so
+        # layout_graphviz/render_graphviz do NOT materialize_nodes() over it
+        g = graphistry.edges(
+            pd.DataFrame({'s': ['a'], 'd': ['b']}), 's', 'd'
+        ).nodes(pd.DataFrame({'id': ['a', 'b']}))
+        assert g._node is None
+        with pytest.raises(ValueError, match='bound node id column'):
+            g_to_pgv(g)
+
+    def test_unbound_edge_endpoints_raise_value_error(self) -> None:
+        g = graphistry.edges(
+            pd.DataFrame({'s': ['a'], 'd': ['b']})
+        ).nodes(pd.DataFrame({'id': ['a', 'b']}), 'id')
+        assert g._source is None and g._destination is None
+        with pytest.raises(ValueError, match='bound edge endpoint columns'):
+            g_to_pgv(g)