Copilot commented on code in PR #12278:
URL: https://github.com/apache/gluten/pull/12278#discussion_r3415663202
##########
cpp/velox/substrait/SubstraitToVeloxExpr.cc:
##########
@@ -217,13 +217,29 @@ std::shared_ptr<const core::FieldAccessTypedExpr>
SubstraitVeloxExprConverter::t
auto inputColumnType = inputType;
for (;;) {
auto idx = tmp->field();
- fieldAccess = makeFieldAccessExpr(inputColumnType->nameOf(idx),
inputColumnType->childAt(idx), fieldAccess);
+ VELOX_USER_CHECK(
+ idx >= 0 && static_cast<uint32_t>(idx) < inputColumnType->size(),
+ "Field reference index {} is out of range for the {}-field row
type.",
+ idx,
+ inputColumnType->size());
Review Comment:
The new bounds-check error message is missing key debugging context (e.g.,
the row type name/layout and the current field path). Consider including
`inputColumnType->toString()` (or equivalent) and, if available in this scope,
the accumulated dereference path so users can identify which reference is
invalid.
##########
cpp/velox/substrait/SubstraitToVeloxExpr.cc:
##########
@@ -217,13 +217,29 @@ std::shared_ptr<const core::FieldAccessTypedExpr>
SubstraitVeloxExprConverter::t
auto inputColumnType = inputType;
for (;;) {
auto idx = tmp->field();
- fieldAccess = makeFieldAccessExpr(inputColumnType->nameOf(idx),
inputColumnType->childAt(idx), fieldAccess);
+ VELOX_USER_CHECK(
+ idx >= 0 && static_cast<uint32_t>(idx) < inputColumnType->size(),
+ "Field reference index {} is out of range for the {}-field row
type.",
+ idx,
+ inputColumnType->size());
+ const TypePtr childType = inputColumnType->childAt(idx);
+ fieldAccess = makeFieldAccessExpr(inputColumnType->nameOf(idx),
childType, fieldAccess);
if (!tmp->has_child()) {
break;
}
- inputColumnType = asRowType(inputColumnType->childAt(idx));
+ // Descending into a nested field is only valid when the current child
is
+ // itself a struct/row. For array/map/primitive children (e.g. a field
+ // nested under an array, as in Delta's "updating array type" case)
+ // asRowType() returns null; previously the next loop iteration
+ // dereferenced that null RowType and crashed the process with a
SIGSEGV.
+ // Throw a user error instead so plan validation fails cleanly and the
+ // query falls back to vanilla execution.
+ inputColumnType = asRowType(childType);
+ VELOX_USER_CHECK_NOT_NULL(
+ inputColumnType,
+ "Nested field reference into a non-struct type (e.g. an array or
map element) is not supported.");
Review Comment:
This error is actionable but too generic: it doesn’t report the actual
encountered child type. Including `childType->toString()` (and optionally the
field name/index) would make failures much easier to diagnose, especially for
deeply nested schemas.
##########
.github/workflows/delta_spark_ut.yml:
##########
@@ -0,0 +1,656 @@
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements. See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Runs Delta Lake's `spark` sbt module unit tests against a Gluten Velox bundle
+# that is built from the source in this repository. The pipeline:
+#
+# 1. Builds the Velox/Gluten native libraries (centos-7 + vcpkg, x86_64).
+# 2. Builds the Gluten Java/Scala jars and assembles the
+# `gluten-velox-bundle-spark<spark>_<scala>-linux_amd64-<version>.jar`
+# fat jar for Spark 4.1 + Scala 2.13 + Java 17 with the Delta profile.
+# 3. Clones delta-io/delta at the requested release tag (default `v4.2.0`),
+# drops the bundle jar into `spark-unified/lib/` only (NOT `spark/lib/`
+# -- see setup-delta.sh for the unmanagedJars scoping rationale),
+# patches Delta's `DeltaSQLCommandTest` to register the Gluten plugin,
+# and runs `sbt spark/test` sharded across the matrix.
+#
+# Limited to Velox + x86 to keep the matrix simple, per the pipeline's purpose
+# of validating Gluten changes against the latest Delta release.
+
+name: Delta Spark UT (Gluten)
+
+on:
+ workflow_dispatch:
+ inputs:
+ delta_ref:
+ description: 'delta-io/delta git ref (tag/branch/SHA) to test against'
+ required: true
+ default: 'v4.2.0'
+ spark_version:
+ description: 'Delta `-DsparkVersion` value (must match the Gluten -P
profile below)'
+ required: true
+ default: '4.1'
+ test_parallelism:
+ description: 'Forked test JVMs per shard (TEST_PARALLELISM_COUNT)'
+ required: true
+ default: '1'
+ update_baseline:
+ description: 'Seed/refresh the known-failures baseline instead of
enforcing it'
+ type: boolean
+ required: false
+ default: false
+ fail_on_fixed:
+ description: 'Fail when a baseline test now passes (keeps the baseline
honest)'
+ type: boolean
+ required: false
+ default: true
+ pull_request:
+ paths:
+ - '.github/workflows/delta_spark_ut.yml'
+ - '.github/workflows/util/delta-spark-ut/**'
+ - 'gluten-delta/**'
+ - 'backends-velox/src-delta40/**/DeltaSQLCommandTest.scala'
+
+env:
+ ACTIONS_ALLOW_USE_UNSECURE_NODE_VERSION: true
+ MVN_CMD: 'build/mvn -ntp'
Review Comment:
Setting `ACTIONS_ALLOW_USE_UNSECURE_NODE_VERSION: true` weakens the
workflow’s security posture and should be avoided unless strictly required. If
it’s necessary for a specific legacy action, scope it to only the job/step that
needs it, or remove it if no longer required by the selected `actions/*@v4`
versions.
##########
.github/workflows/util/delta-spark-ut/setup-delta.sh:
##########
@@ -0,0 +1,141 @@
+#!/usr/bin/env bash
+
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements. See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+#
+# Prepares a delta-io/delta clone for running its `spark` module tests with the
+# Gluten (Velox) bundle jar on the classpath.
+#
+# Usage:
+# setup-delta.sh <delta_ref> <delta_dir> <gluten_bundle_jar>
<gluten_repo_root>
+#
+# Arguments:
+# delta_ref - git ref (tag/branch/sha) to check out (e.g. v4.2.0)
+# delta_dir - destination directory for the Delta clone
+# gluten_bundle_jar - path to the gluten-velox-bundle fat jar
+# gluten_repo_root - path to the Gluten repository root (used to locate
+#
backends-velox/src-delta40/.../DeltaSQLCommandTest.scala)
+#
+
+set -euo pipefail
+
+if [ "$#" -ne 4 ]; then
+ echo "Usage: $0 <delta_ref> <delta_dir> <gluten_bundle_jar>
<gluten_repo_root>" >&2
+ exit 1
+fi
+
+DELTA_REF="$1"
+DELTA_DIR="$2"
+GLUTEN_BUNDLE_JAR="$3"
+GLUTEN_ROOT="$4"
+
+if [ ! -f "$GLUTEN_BUNDLE_JAR" ]; then
+ echo "Gluten bundle jar not found: $GLUTEN_BUNDLE_JAR" >&2
+ exit 1
+fi
+
+# Reuse the existing DeltaSQLCommandTest from Gluten's backends-velox module
+# rather than maintaining a separate copy. This file is compiled as part of the
+# unified `spark` project's Test scope, which has the Gluten bundle on its
+# classpath (via spark-unified/lib/), so the typed GlutenConfig /
VeloxDeltaConfig
+# imports resolve correctly.
+PATCH_SOURCE="$GLUTEN_ROOT/backends-velox/src-delta40/test/scala/org/apache/spark/sql/delta/test/DeltaSQLCommandTest.scala"
+if [ ! -f "$PATCH_SOURCE" ]; then
+ echo "Gluten DeltaSQLCommandTest not found: $PATCH_SOURCE" >&2
+ exit 1
+fi
+
+echo "::group::Cloning delta-io/delta @ ${DELTA_REF}"
+# Shallow clone the requested tag/branch. Fall back to full clone when the ref
is a SHA.
+if ! git clone --depth 1 --branch "$DELTA_REF"
https://github.com/delta-io/delta.git "$DELTA_DIR"; then
+ echo "Shallow clone of ref '${DELTA_REF}' failed, falling back to full
clone."
+ rm -rf "$DELTA_DIR"
+ git clone https://github.com/delta-io/delta.git "$DELTA_DIR"
+ git -C "$DELTA_DIR" checkout "$DELTA_REF"
+fi
Review Comment:
`git clone ... \"$DELTA_DIR\"` will fail if `$DELTA_DIR` already exists
(common when re-running locally or after a partially failed run). Consider
adding an explicit preflight check to either (a) fail with a clear message when
`$DELTA_DIR` exists, or (b) remove it before cloning (consistent with the
fallback path’s `rm -rf`).
##########
.github/workflows/util/delta-spark-ut/compare-test-results.py:
##########
@@ -0,0 +1,467 @@
+#!/usr/bin/env python3
+#
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements. See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Gate / seed / aggregate the Delta-on-Gluten unit test results.
+
+Running delta-io/delta's ScalaTest suite against the Gluten Velox bundle
+produces many *expected* failures (Gluten does not yet support every Delta
+code path). To keep the red/green signal meaningful while we fix those
+failures incrementally, we maintain a committed baseline of known failing
+tests (``known-failures.txt``) and compare each CI run against it.
+
+This script has three modes:
+
+``enforce`` (default, per shard)
+ Parse the JUnit XML produced by ``sbt spark/test`` (ScalaTest ``-u``
+ reporter) and compare against the baseline:
+
+ * regression -- a test that FAILED but is NOT in the baseline. These
+ fail the build: a previously-passing test just started failing.
+ * expected -- a test that failed and IS in the baseline. Ignored.
+ * fixed -- a baseline test that now PASSES. By default these also
+ fail the build (``--fail-on-fixed true``) so the baseline stays honest
+ and contributors remove entries as they fix them.
+
+ If the baseline is empty (not yet bootstrapped) the mode automatically
+ degrades to ``seed`` so the first run is never spuriously red.
+
+``seed`` (bootstrap / ``update_baseline``)
+ Never fails. Just writes the current shard's failing tests so the baseline
+ can be (re)generated from a real run.
+
+``aggregate`` (final job)
+ Merge every shard's ``--failures-out`` / ``--ran-out`` file into a single,
+ sorted, ready-to-commit ``known-failures.txt`` and report stale baseline
+ entries (tests no longer present in any shard).
+
+Baseline file format (``known-failures.txt``)::
+
+ # comment lines start with '#'
+ <fully.qualified.SuiteName>#<test display name>
+
+The suite is always a JVM class name (dot-separated, never starts with '#'),
+so a line whose first non-space character is '#' is unambiguously a comment,
+and the FIRST '#' after the suite separates suite from the (possibly
+'#'-containing) test name.
+
+Only the Python standard library is used so the script runs in the bare
+centos image used by the Delta UT pipeline with no ``pip install``.
+"""
+
+import argparse
+import glob
+import os
+import sys
+import xml.etree.ElementTree as ET
+
+# Synthetic "test name" recorded when a whole suite aborts (e.g. beforeAll
+# throws) so that the JUnit XML reports a suite-level error with no per-test
+# <testcase>. Without this, a suite that used to pass but now aborts entirely
+# would record zero failing testcases and the regression would be missed.
+SUITE_ABORTED = "<suite aborted>"
+
+SEP = "#"
+
+
+def eprint(*args, **kwargs):
+ print(*args, file=sys.stderr, **kwargs)
+
+
+# --------------------------------------------------------------------------- #
+# Baseline (known-failures.txt) parsing / formatting
+# --------------------------------------------------------------------------- #
+def format_entry(suite, test):
+ return "{}{}{}".format(suite, SEP, test)
+
+
+def parse_entry(line):
+ """Parse a 'suite#test' line into (suite, test) or return None for
blanks/comments."""
+ stripped = line.strip()
+ if not stripped or stripped.startswith("#"):
+ return None
+ idx = stripped.find(SEP)
+ if idx < 0:
+ # No separator: treat the whole line as a suite-level entry.
+ return (stripped, SUITE_ABORTED)
+ return (stripped[:idx], stripped[idx + len(SEP) :])
+
+
+def load_entries(path):
+ """Load a set of (suite, test) tuples from a baseline/shard-list file."""
+ entries = set()
+ if not path or not os.path.exists(path):
+ return entries
+ with open(path, "r", encoding="utf-8") as fh:
+ for line in fh:
+ parsed = parse_entry(line)
+ if parsed is not None:
+ entries.add(parsed)
+ return entries
+
+
+def write_entries(path, entries, header=None):
+ """Write a sorted set of (suite, test) tuples to a file."""
+ os.makedirs(os.path.dirname(os.path.abspath(path)) or ".", exist_ok=True)
+ with open(path, "w", encoding="utf-8") as fh:
+ if header:
+ for hl in header.splitlines():
+ fh.write(hl.rstrip() + "\n")
+ for suite, test in sorted(entries):
+ # Defensive: collapse any stray newlines so each entry stays on
one line.
+ safe_test = test.replace("\r", " ").replace("\n", " ")
+ fh.write(format_entry(suite, safe_test) + "\n")
+
+
+# --------------------------------------------------------------------------- #
+# JUnit XML parsing
+# --------------------------------------------------------------------------- #
+def _iter_testsuites(root):
+ """Yield every <testsuite> element regardless of whether the file root is
+ <testsuites> (wrapper) or a single <testsuite>."""
+ tag = root.tag.split("}")[-1] # strip any namespace
+ if tag == "testsuites":
+ for child in root:
+ if child.tag.split("}")[-1] == "testsuite":
+ yield child
+ elif tag == "testsuite":
+ yield root
+
+
+def _child_local_tags(elem):
+ return {c.tag.split("}")[-1] for c in elem}
+
+
+def parse_reports(reports_dir):
+ """Walk reports_dir for JUnit XML and classify every test.
+
+ Returns (passed, failed, skipped) sets of (suite, test) tuples. A test is
+ 'failed' if its <testcase> has a <failure> or <error> child, 'skipped' if
+ it has a <skipped> child, otherwise 'passed'. Suite-level aborts (a
+ <testsuite> reporting errors/failures with no failing <testcase>) are
+ recorded as a synthetic (suite, SUITE_ABORTED) failure.
+ """
+ passed, failed, skipped = set(), set(), set()
+
+ xml_files = []
+ # ScalaTest's -u reporter and Maven surefire both write `TEST-<suite>.xml`
+ # under a `target/.../*-reports/` dir. Restrict the secondary glob to
+ # `target/` so we never parse Delta's own XML *test resources* (which live
+ # under src/test/resources and are not reports). The <testsuite>-root guard
+ # below is a final safety net.
+ for pattern in ("**/TEST-*.xml", "**/target/**/*.xml"):
+ xml_files.extend(glob.glob(os.path.join(reports_dir, pattern),
recursive=True))
+ xml_files = sorted(set(xml_files))
Review Comment:
Globbing and parsing `**/target/**/*.xml` can become very expensive in large
workspaces and may pick up many irrelevant XML files (even if filtered later by
root tag). To reduce runtime and flakiness, restrict patterns to known report
paths (e.g. `**/target/test-reports/TEST-*.xml`,
`**/target/surefire-reports/TEST-*.xml`) and keep the broader scan only as an
opt-in fallback.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
