Add a GitHub Actions CI workflow

.github/workflows/ci.yml

@@ -0,0 +1,132 @@
+# pgColumnar continuous integration.
+#
+# This mirrors the cadence the project already runs locally rather than
+# inventing a second one: a build preflight across every supported major, which
+# is what catches an API break, and the suite run on one major, which is what
+# catches a behaviour break. The full five-major suite matrix stays a local gate
+# (test/run_all_versions.sh) because PostgreSQL 19 is beta and not packaged in
+# PGDG stable, so CI cannot reproduce it honestly.
+#
+# Warnings are failures here for the same reason they are in the local matrix.
+
+name: build and test
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ci-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  # Build against every supported major. Fast, and it is what an API change
+  # between majors trips first.
+  build:
+    name: build (PG ${{ matrix.pg }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        pg: ['15', '16', '17', '18']
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install PostgreSQL ${{ matrix.pg }} and codec headers
+        run: |
+          set -euo pipefail
+          sudo install -d /usr/share/postgresql-common/pgdg
+          sudo curl -fsSL -o /usr/share/postgresql-common/pgdg/apt.postgresql.org.asc \
+            https://www.postgresql.org/media/keys/ACCC4CF8.asc
+          echo "deb [signed-by=/usr/share/postgresql-common/pgdg/apt.postgresql.org.asc] \
+            https://apt.postgresql.org/pub/repos/apt $(lsb_release -cs)-pgdg main" \
+            | sudo tee /etc/apt/sources.list.d/pgdg.list >/dev/null
+          sudo apt-get update
+          sudo apt-get install -y --no-install-recommends \
+            postgresql-${{ matrix.pg }} postgresql-server-dev-${{ matrix.pg }} \
+            liblz4-dev libzstd-dev zlib1g-dev
+
+      - name: Build, treating warnings as failures
+        run: |
+          set -euo pipefail
+          PG_CONFIG=/usr/lib/postgresql/${{ matrix.pg }}/bin/pg_config
+          "$PG_CONFIG" --version
+          make PG_CONFIG="$PG_CONFIG" 2> build.err || { cat build.err; exit 1; }
+          if grep -q 'warning:' build.err; then
+            echo "::error::compiler warnings are failures in this project"
+            grep 'warning:' build.err
+            exit 1
+          fi
+          cat build.err || true
+
+      - name: Verify the module's symbols resolve against the server
+        run: |
+          set -euo pipefail
+          BIN=$(/usr/lib/postgresql/${{ matrix.pg }}/bin/pg_config --bindir)
+          missing=$(nm -D --undefined-only pgcolumnar.so \
+                    | awk '{print $2}' | sort -u \
+                    | while read -r s; do
+                        nm -D --defined-only "$BIN/postgres" 2>/dev/null \
+                          | grep -q " $s\$" || echo "$s"
+                      done | grep -vE '^(_|__|GLIBC)' | head -20 || true)
+          # Informational: the local gate does this strictly, and the packaged
+          # server exports a different set than a source build, so a difference
+          # here is reported rather than failed on.
+          [ -z "$missing" ] && echo "all resolve" || { echo "unresolved (informational):"; echo "$missing"; }
+
+  # Run the suites on one major. This is the behaviour gate; the local
+  # five-major matrix remains the release gate.
+  suites:
+    name: suites (PG 17)
+    runs-on: ubuntu-latest
+    needs: build
+    timeout-minutes: 60
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install PostgreSQL 17, codec headers, and pyarrow
+        run: |
+          set -euo pipefail
+          sudo install -d /usr/share/postgresql-common/pgdg
+          sudo curl -fsSL -o /usr/share/postgresql-common/pgdg/apt.postgresql.org.asc \
+            https://www.postgresql.org/media/keys/ACCC4CF8.asc
+          echo "deb [signed-by=/usr/share/postgresql-common/pgdg/apt.postgresql.org.asc] \
+            https://apt.postgresql.org/pub/repos/apt $(lsb_release -cs)-pgdg main" \
+            | sudo tee /etc/apt/sources.list.d/pgdg.list >/dev/null
+          sudo apt-get update
+          sudo apt-get install -y --no-install-recommends \
+            postgresql-17 postgresql-server-dev-17 \
+            liblz4-dev libzstd-dev zlib1g-dev python3-pip
+          # The Arrow and Parquet suites use pyarrow as an independent reader and
+          # skip themselves without it, which would be a silent loss of coverage.
+          pip3 install --break-system-packages --quiet pyarrow || pip3 install --quiet pyarrow
+
+      - name: Stop the packaged cluster
+        # The suites start their own throwaway clusters on private ports; the
+        # packaged one is unused and only competes for shared memory.
+        run: sudo systemctl stop postgresql || true
+
+      - name: Run the suite matrix on PG 17
+        run: |
+          set -euo pipefail
+          # PGC_SKIP_TIMING drops the three wall-clock suites: a shared runner
+          # cannot hold a ratio still, and a gate that reds for reasons unrelated
+          # to the change is worse than one that does not run. They stay in the
+          # local matrix, which is where those numbers mean anything.
+          # PGC_JOBS is sized for a 4-core runner rather than the local 8.
+          sudo -E env "PATH=$PATH" PGC_SKIP_TIMING=1 PGC_JOBS=4 \
+            bash test/run_all_versions.sh /usr/lib/postgresql/17/bin/pg_config
+
+      - name: Collect logs on failure
+        if: failure()
+        run: |
+          for f in /tmp/pgcolumnar-matrix-*/*.log; do
+            [ -e "$f" ] || continue
+            echo "===== $f ====="
+            tail -60 "$f"
+          done

test/run_all_versions.sh

@@ -228,10 +228,22 @@ for pgc in "${CONFIGS[@]}"; do
 	wait
 
 	# Then the timing-sensitive suites, one at a time, with nothing else running.
+	#
+	# PGC_SKIP_TIMING=1 drops them entirely. That exists for shared CI hardware,
+	# where the wall-clock ratios these three assert cannot be trusted: a runner
+	# is noisy by construction and a gate that goes red for reasons unrelated to
+	# the change teaches its readers to discount red, which is worse than not
+	# running it. They stay in every local run, which is where the numbers mean
+	# something.
 	for s in "${SUITES[@]}"; do
 		if ! is_timing_suite "$s"; then
 			continue
 		fi
+		if [ "${PGC_SKIP_TIMING:-0}" = 1 ]; then
+			echo "  SKIP  $s (PGC_SKIP_TIMING)"
+			echo 0 >"$builddir/${s}.rc"
+			continue
+		fi
 		port=$((BASE_PORT++))
 		PGC_SKIP_BUILD=1 PGC_PORT="$port" \
 			bash "$builddir/test/${s}.sh" "$pgc" >"$builddir/${s}.log" 2>&1

test/wal_envelope.sh

@@ -66,19 +66,29 @@ end="$(awk -v s="$start" 'NR > s && /^[A-Za-z_][A-Za-z0-9_]*\(/ {print NR; exit}
 
 check "ColumnarTruncateMainFork was found" "$([ -n "$start" ] && echo yes || echo no)" "yes"
 
-# line number of the first match of a pattern inside the function, or 0
+# Line number of the first line inside the function that CONTAINS the literal
+# string, or empty.
+#
+# Deliberately a substring test rather than a regex match. Every pattern below is
+# literal C text, and passing one through awk's -v made the backslashes a string
+# escape before the regex ever saw them: awk turns "\(" into "(", so
+# 'XLogInsert\(RM_SMGR_ID' reached the matcher as 'XLogInsert(RM_SMGR_ID' with an
+# unterminated group, and never matched. Whether that happened at all depended on
+# the awk build, so this suite passed locally and failed on a runner with a
+# different mawk, reporting three steps of the envelope as missing when the code
+# was fine. index() has no escaping question to get wrong.
 at() {
 	awk -v s="$start" -v e="$end" -v pat="$1" \
-		'NR >= s && NR <= e && $0 ~ pat { print NR; exit }' "$FN"
+		'NR >= s && NR <= e && index($0, pat) { print NR; exit }' "$FN"
 }
 
-delay_set="$(at 'delayChkptFlags \|= DELAY_CHKPT_COMPLETE')"
-crit_in="$(at 'START_CRIT_SECTION\(\)')"
-insert="$(at 'XLogInsert\(RM_SMGR_ID')"
-flush="$(at 'XLogFlush\(')"
-trunc="$(at 'COLUMNAR_SMGRTRUNCATE\(')"
+delay_set="$(at 'delayChkptFlags |= DELAY_CHKPT_COMPLETE')"
+crit_in="$(at 'START_CRIT_SECTION()')"
+insert="$(at 'XLogInsert(RM_SMGR_ID')"
+flush="$(at 'XLogFlush(')"
+trunc="$(at 'COLUMNAR_SMGRTRUNCATE(')"
 delay_clear="$(at 'delayChkptFlags &= ~DELAY_CHKPT_COMPLETE')"
-crit_out="$(at 'END_CRIT_SECTION\(\)')"
+crit_out="$(at 'END_CRIT_SECTION()')"
 
 for v in delay_set crit_in insert flush trunc delay_clear crit_out; do
 	eval "val=\$$v"