perf(gfql): materialize indexed point-query rows directly

.github/workflows/ci.yml

@@ -1598,19 +1598,18 @@ jobs:
         ./bin/test-graphviz.sh
 
   gfql-routes-off:
-    # Non-blocking ledger: replays the GFQL suites with one hot path declined per cell
+    # Required replay: run the GFQL suites with one hot path declined per cell
     # (GFQL_ROUTES_OFF); engagement pins are skipped via the route_engaged marker, so the
     # uploaded lists hold route-vs-general result divergences only.
     needs: [changes, test-gfql-core, generate-lockfiles]
     if: ${{ needs.changes.outputs.gfql == 'true' || github.event_name == 'workflow_dispatch' || github.event_name == 'schedule' }}
     runs-on: ubuntu-latest
-    continue-on-error: true
     timeout-minutes: 45
 
     strategy:
       fail-fast: false
       matrix:
-        mode: [native-fast, polars-seeded, polars-plain, index-hop, indexed-kernel, cypher-fast, all-off]
+        mode: [polars-point-rows, point-rows, native-fast, polars-single-node, polars-seeded, polars-plain, index-hop, indexed-kernel, cypher-fast, all-off]
 
     steps:
 

CHANGELOG.md

@@ -19,6 +19,9 @@ This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 
 ### Fixed
 
+* GFQL: indexed scalar node and directed single-hop queries followed by `rows(source=...)` and optional `select(...)` now produce row tables directly on pandas, cuDF, and Polars. Joined property projections preserve repeated edge matches without constructing full traversal results.
+* GFQL: indexed scalar equality filters no longer retain null-valued rows when the comparison produces a nullable boolean mask.
+
 * GFQL: a native chain whose edge alias is named like the source, destination or edge-ID binding column silently overwrote that binding on pandas and cuDF (the seed's edges vanished, or the edge ids became `True`) and raised a raw polars `SchemaError`; it is now the same typed decline (E108) the node-ID collision already gets, on every engine, before execution — the rule the Cypher route already applied (#2050).
 * GFQL polars: a colliding alias no longer leaves the internal `__gfql_shadow_restore__<alias>__` column on op-list results; the chain keeps the shadowed values under that name only while a compiled Cypher pipeline runs (its row pipeline reads them back), and on `gfql([...])` / `chain([...])` the marker simply shadows the column as on pandas and cuDF. Nothing is stripped from results: user-defined columns and pipelines composed from successive `gfql` calls are untouched.
 

bin/test-polars.sh

@@ -89,7 +89,12 @@ POLARS_TEST_FILES=(
     graphistry/tests/compute/gfql/test_native_seed_skip_refilter.py
     graphistry/tests/compute/gfql/test_polars_native_seed_resolution.py
     graphistry/tests/compute/gfql/lazy/engine/polars/chain_specializations/test_polars_admission.py
+    graphistry/tests/compute/gfql/lazy/engine/polars/test_predicates.py
+    graphistry/tests/compute/gfql/lazy/engine/polars/chain_specializations/test_point_rows.py
+    graphistry/tests/compute/gfql/lazy/engine/polars/chain_specializations/test_hotpaths.py
+    graphistry/tests/compute/gfql/test_polars_indexed_join_helpers.py
     graphistry/tests/compute/chain_specializations/test_native_admission.py
+    graphistry/tests/compute/chain_specializations/test_point_rows.py
     graphistry/tests/compute/gfql/test_undirected_pairs_2026.py
     graphistry/tests/compute/gfql/test_native_seed_lane_explain.py
     # #1882/#1913-f4/#1879 crash-family pins: the polars params (filter helpers on polars
@@ -103,6 +108,7 @@ POLARS_TEST_FILES=(
     graphistry/tests/compute/gfql/test_polars_rows_entity_groupby.py
     graphistry/tests/compute/gfql/test_seeded_typed_hop_fastpath.py
     graphistry/tests/compute/gfql/test_residual_polars_native.py
+    graphistry/tests/compute/gfql/index/test_engine_arrays.py
     graphistry/tests/compute/gfql/index/test_auto_engine_agreement.py
     graphistry/tests/compute/gfql/index/test_degree_consult.py
     graphistry/tests/compute/gfql/test_single_alias_cache_key.py
@@ -127,6 +133,11 @@ POLARS_TEST_FILES=(
     graphistry/tests/compute/test_chain_alias_column_collision.py
     graphistry/tests/compute/test_gfql_op_list_hides_internal_columns.py
     graphistry/tests/compute/gfql/routes/test_route_harness.py
+    graphistry/tests/compute/gfql/routes/test_point_boundaries.py
+    graphistry/tests/compute/gfql/routes/test_has_collision_contract.py
+    graphistry/tests/compute/gfql/routes/test_node_selection_rows.py
+    graphistry/tests/compute/gfql/test_join_backend_contracts.py
+    graphistry/tests/compute/test_chain_validation_execution.py
     graphistry/tests/compute/gfql/test_engine_polars_semi_key_dedup.py
     graphistry/tests/compute/gfql/test_engine_polars_call_modality.py
     graphistry/tests/compute/gfql/test_engine_polars_gpu.py

bin/test-routes-off.sh

@@ -1,17 +1,20 @@
 #!/usr/bin/env bash
 # Replay the GFQL suites with each hot path declined (GFQL_ROUTES_OFF, see graphistry/tests/conftest.py).
 # Engagement pins carry the route_engaged marker and are skipped, so every remaining failure is a
-# route-vs-general result divergence. Non-blocking ledger: always exits 0; per-mode logs + id lists in $OUT.
+# route-vs-general result divergence. Fail on any failed replay; retain per-mode logs and id lists.
 set -uo pipefail
 cd "$(dirname "$0")/.."
-MODES=${MODES:-native-fast polars-seeded polars-plain index-hop indexed-kernel cypher-fast all-off}
+# One registry drives forcing and replay: new routes automatically join all-off.
+ALL_ROUTES=$(python -c 'from graphistry.tests.compute.gfql.routes.switch import ROUTES; print(",".join(ROUTES))') || exit $?
+MODES=${MODES:-${ALL_ROUTES//,/ } all-off}
 SUITES=${SUITES:-graphistry/tests/compute/test_chain.py graphistry/tests/compute/test_hop.py graphistry/tests/compute/test_gfql.py graphistry/tests/compute/gfql}
 OUT=${OUT:-build/routes-off}
 mkdir -p "$OUT"
+status=0
 for mode in $MODES; do
-  if [ "$mode" = all-off ]; then routes=native-fast,polars-seeded,polars-plain,index-hop,indexed-kernel,cypher-fast; else routes=$mode; fi
-  GFQL_ROUTES_OFF=$routes python -m pytest $SUITES -q -p no:cacheprovider -o addopts="" -rfE > "$OUT/$mode.log" 2>&1
+  if [ "$mode" = all-off ]; then routes=$ALL_ROUTES; else routes=$mode; fi
+  GFQL_ROUTES_OFF=$routes python -m pytest $SUITES -q -p no:cacheprovider -o addopts="" -rfE > "$OUT/$mode.log" 2>&1 || status=1
   grep -E "^(FAILED|ERROR) " "$OUT/$mode.log" | sed 's/ - .*//' | sort -u > "$OUT/$mode.divergences"
   echo "$mode: $(wc -l < "$OUT/$mode.divergences") divergence id(s); $(tail -1 "$OUT/$mode.log")"
 done
-exit 0
+exit "$status"

graphistry/Engine.py

@@ -138,6 +138,12 @@ def resolve_engine(
             except ImportError:
                 pass
 
+        type_module = type(g_or_df).__module__
+        if 'dask' in type_module and 'cudf' not in type_module:
+            import dask.dataframe as dd
+            if isinstance(g_or_df, dd.DataFrame):
+                return Engine.PANDAS
+
         if 'cudf.core.dataframe' in str(getmodule(g_or_df)):
             has_cudf_dependancy_, _, _ = lazy_cudf_import()
             if has_cudf_dependancy_:

graphistry/compute/chain.py

@@ -12,6 +12,8 @@
 from .typing import DataFrameT, SeriesT
 from .util import generate_safe_column_name
 from .chain_specializations.hotpaths import _try_chain_fast_path
+from .chain_specializations.point_rows import _try_point_rows
+from .engine_coercion import ensure_local_engine_match
 from graphistry.compute.validate.validate_schema import validate_chain_schema, validate_graph_shape
 from graphistry.compute.gfql.strictness import StrictInput
 from graphistry.compute.gfql.same_path_types import (
@@ -545,6 +547,9 @@ def apply_output_slice(op: ASTObject, op_label: ASTObject, df):
                     out_df[base] = out_df[c_x].where(out_df[c_x].notna(), out_df[c])
                 out_df = out_df.drop(columns=[c, c_x])
 
+    # Empty pandas merges can move the binding column behind aliases or properties.
+    if out_df.columns[0] != id:
+        out_df = out_df[[id, *[column for column in out_df.columns if column != id]]]
     return out_df
 
 
@@ -969,6 +974,8 @@ def _chain_with_strictness(
                     "Install RAPIDS/cudf_polars, or use engine='polars' for native CPU execution."
                 )
     self = _coerce_input_formats(self, engine_concrete_early)
+    if engine_concrete_early == Engine.PANDAS:
+        self = ensure_local_engine_match(self, engine_concrete_early)
 
     if engine_concrete_early in POLARS_ENGINES:
         # Native polars chain lives in a dedicated dispatched module so the
@@ -977,7 +984,8 @@ def _chain_with_strictness(
         # POLARS_GPU = the same lazy engine with the GPU execution target.
         # (Dependency guards for polars / cudf_polars are above, pre-coercion.)
         if validate_schema:
-            Chain(ops if not isinstance(ops, Chain) else ops.chain).validate(collect_all=False)
+            # Construct a fresh validator: the constructor validates children once.
+            Chain(ops if not isinstance(ops, Chain) else ops.chain)
             validate_graph_shape(self, ops, collect_all=False)  # pandas gets this via validate_chain_schema (#1889)
         from graphistry.compute.gfql.lazy.engine.polars.chain import chain_polars
         from graphistry.compute.gfql.lazy import target_mode, ExecutionTarget
@@ -1005,6 +1013,10 @@ def _chain_with_strictness(
         finally:
             call_thread_local.policy = old_policy
     else:
+        point_ops = ops.chain if isinstance(ops, Chain) else ops
+        point_result = _try_point_rows(self, point_ops, engine_concrete_early, start_nodes, validate_schema)
+        if point_result is not None:
+            return point_result
         return _chain_impl(self, ops, engine, validate_schema, policy, context, start_nodes)
 
 
@@ -1042,7 +1054,8 @@ def _chain_impl(
         ops = ops.chain
 
     if validate_schema:
-        Chain(ops).validate(collect_all=False)
+        # Revalidate mutable operations on every execution, including reused Chains.
+        Chain(ops)
 
     from graphistry.compute.ast import ASTCall
 
@@ -1231,6 +1244,27 @@ def _chain_impl(
                 from .gfql.exec_context import clear_row_exec_context
                 g_out = clear_row_exec_context(g_out)
             success = True
+        elif len(ops) == 1 and isinstance(ops[0], ASTNode):
+            # A node selection preserves each source row. Rejoining node IDs in
+            # the traversal combine would multiply duplicate rows and properties.
+            g_out = g_stack[0]
+            alias = ops[0]._name
+            if alias is not None:
+                cols = [c for c in g_out._nodes.columns if c != alias]
+                if g._node in cols:
+                    cols = [g._node, *[c for c in cols if c != g._node]]
+                cols = ([*cols, alias] if alias in g._nodes.columns
+                        else [*cols[:1], alias, *cols[1:]])
+                g_out = g_out.nodes(g_out._nodes[cols].reset_index(drop=True))
+            if synthesized_empty_edges:
+                g_out = self.nodes(g_out._nodes, g._node)
+            elif added_edge_index:
+                g_out = self.nodes(g_out._nodes, g._node).edges(
+                    g_out._edges.drop(columns=[g._edge]), edge=original_edge)
+            elif g._edge is not None:
+                edge_cols = [g._edge, *[c for c in g_out._edges.columns if c != g._edge]]
+                g_out = g_out.edges(g_out._edges[edge_cols])
+            success = True
         else:
             # Phase 2: Backward pass to propagate downstream constraints.
             g_stack_reverse : List[Plottable] = []