pg-clickhouse-c

.github/workflows/bake.yml

@@ -20,7 +20,7 @@ jobs:
     steps:
       - name: Checkout the Repository
         uses: actions/checkout@v6
-        with: { submodules: true }
+        with: { submodules: recursive }
       - name: Set up Depot CLI
         uses: depot/setup-action@v1
       - name: Log in to GHCR

.github/workflows/clickhouse.yml

@@ -36,7 +36,7 @@ jobs:
         run: pgxn install re2
       - name: Checkout the Repository
         uses: actions/checkout@v6
-        with: { submodules: true }
+        with: { submodules: recursive }
       - name: Start ClickHouse
         env: { CH_RELEASE: "${{ matrix.ch }}" }
         run: .github/ubuntu/clickhouse.sh

.github/workflows/lint.yml

@@ -17,7 +17,7 @@ jobs:
         run: NO_CLUSTER=1 pg-start ${{ env.pg }} libcurl4-openssl-dev uuid-dev liblz4-dev libzstd-dev clang-tidy bear shellcheck
       - name: Checkout the Repository
         uses: actions/checkout@v6
-        with: { submodules: true }
+        with: { submodules: recursive }
       - name: Install pre-commit
         run: make debian-install-lint
       - name: Add Postgres to Path

.github/workflows/postgres.yml

@@ -27,7 +27,7 @@ jobs:
         run: pgxn install re2
       - name: Checkout the Repository
         uses: actions/checkout@v6
-        with: { submodules: true }
+        with: { submodules: recursive }
       - name: Start ClickHouse
         run: .github/ubuntu/clickhouse.sh
       - name: Test

.gitmodules

@@ -1,3 +1,3 @@
-[submodule "vendor/clickhouse-c"]
-	path = vendor/clickhouse-c
-	url = https://github.com/ClickHouse/clickhouse-c
+[submodule "vendor/pg-clickhouse-c"]
+	path = vendor/pg-clickhouse-c
+	url = https://github.com/ClickHouse/pg-clickhouse-c

CHANGELOG.md

@@ -78,6 +78,12 @@ All notable changes to this project will be documented in this file. It uses the
     rather than understating foreign scan cost ([#310]).
 *   Added mapping to push down the re2 v0.4 `@~` operator to ClickHouse as
     the `match()` function ([#318]).
+*   Binary driver errors now name the column and type involved, for example
+    `cannot encode integer into ClickHouse Array(Array(Int32)) (column "c2")`
+    in place of `unexpected PG/CH type pair for column 0` ([#326]).
+*   Coerce array elements in binary driver. `Array(Int32)` to `bigint[]`, or
+    `quantilesExactLow()` results into `double precision[]`, no longer fails
+    with `could not cast value from integer[] to bigint[]` ([#326]).
 
 ### 🐞 Bug Fixes
 
@@ -160,6 +166,8 @@ All notable changes to this project will be documented in this file. It uses the
     "ClickHouse/pg_clickhouse#317 Push down the array IN family unconditionally"
   [#319]: https://github.com/ClickHouse/pg_clickhouse/pull/319
     "ClickHouse/pg_clickhouse#319 Fix foreign scan RTE selection"
+  [#326]: https://github.com/ClickHouse/pg_clickhouse/pull/326
+    "ClickHouse/pg_clickhouse#326 pg-clickhouse-c"
 
 ## [v0.3.2] — 2026-06-16
 

Makefile

@@ -19,13 +19,18 @@ ARCH         = $(shell uname -m)
 # Collect all the C files to compile into MODULE_big.
 OBJS = $(subst .c,.o, $(wildcard src/*.c src/*/*.c))
 
-# clickhouse-c is a header-only single-header library. Override
-# CH_C_DIR to point elsewhere when developing against a local checkout.
-CH_C_DIR ?= vendor/clickhouse-c
+# clickhouse-c and pg-clickhouse-c are header-only. pg-clickhouse-c vendors the
+# clickhouse-c commit its own signatures compile against, so take that pin
+# rather than a second checkout. Override CH_C_DIR / PGCH_DIR to point
+# elsewhere when developing against a local checkout.
+PGCH_DIR = vendor/pg-clickhouse-c
+CH_C_DIR = $(PGCH_DIR)/clickhouse-c
 
 # Own headers via -I; vendored and PG server headers via -isystem so their
 # own warnings (eg -Wsign-compare in PG array.h macros) don't trip -Werror.
-PG_CPPFLAGS = -I./src/include -isystem $(CH_C_DIR) -isystem $(shell $(PG_CONFIG) --includedir-server)
+# PGCH_MSG_PREFIX prefixes every message pg-clickhouse-c raises; it expands
+# inside the ereport macros, so every TU must see the same definition.
+PG_CPPFLAGS = -I./src/include -isystem $(CH_C_DIR) -isystem $(PGCH_DIR) -isystem $(shell $(PG_CONFIG) --includedir-server) -DPGCH_MSG_PREFIX='"pg_clickhouse: "'
 
 # Link OpenSSL (for TLS in the binary driver), curl (for the HTTP driver),
 # libuuid (for http_streaming.c's query-id generator), and lz4 / zstd
@@ -50,13 +55,15 @@ EXTRA_CLEAN = sql/$(EXTENSION)--$(EXTVERSION).sql src/include/version.h compile_
 PGXS := $(shell $(PG_CONFIG) --pgxs)
 include $(PGXS)
 
-# Clone clickhouse-c submodule.
-$(CH_C_DIR)/clickhouse.h: .gitmodules
-	git submodule update --init
+# Clone vendored submodules, clickhouse-c nested inside pg-clickhouse-c. Spelled
+# out rather than via CH_C_DIR / PGCH_DIR: an override points at a checkout git
+# knows nothing about.
+vendor/pg-clickhouse-c/clickhouse-c/clickhouse.h vendor/pg-clickhouse-c/pg-clickhouse.h: .gitmodules
+	git submodule update --init --recursive
 
-# Require clickhouse-c and the version header. The bitcode twins compile the
-# same sources, so they need the same generated header.
-$(OBJS) $(OBJS:.o=.bc): $(CH_C_DIR)/clickhouse.h src/include/version.h
+# Require both vendored libraries and the version header. The bitcode twins
+# compile the same sources, so they need the same generated header.
+$(OBJS) $(OBJS:.o=.bc): $(CH_C_DIR)/clickhouse.h $(PGCH_DIR)/pg-clickhouse.h src/include/version.h
 
 # Require the versioned SQL script.
 all: sql/$(EXTENSION)--$(EXTVERSION).sql

src/binary/binary.c

@@ -1,15 +1,11 @@
 /*
  * binary.c
  *
- * Core glue for the binary driver. Owns the chc_alloc thunks (palloc on
- * CurrentMemoryContext, MCXT_ALLOC_HUGE so block buffers can escape the 1GB
- * cap), error mapping into ereport, and a couple of small type helpers shared
- * across the subdir.
- *
- * This also holds the CHC_IMPLEMENTATION define for the clickhouse-c
- * header-only library; the other .c files in src/binary include the
- * same headers without that define and link against the bodies emitted
- * here.
+ * Core glue for the binary driver. Holds the CHC_IMPLEMENTATION and
+ * PGCH_IMPLEMENTATION defines for the header-only clickhouse-c and
+ * pg-clickhouse-c libraries; the other .c files in src/binary include the
+ * same headers without those defines and link against the bodies emitted
+ * here. pgch_in_alloc needs both in the same TU, so they stay together.
  *
  * API lives in src/include/binary.h. Driver internals shared between
  * connection.c / select.c / insert.c live in binary_internal.h
@@ -28,50 +24,11 @@
 #include "clickhouse-posix-io.h"
 #include "clickhouse.h"
 
-#include "binary_internal.h"
-
-static void*
-pg_chc_palloc(void* ud pg_attribute_unused(), size_t n) {
-    return MemoryContextAllocExtended(CurrentMemoryContext, n, MCXT_ALLOC_HUGE);
-}
-
-static void*
-pg_chc_repalloc(
-    void* ud pg_attribute_unused(),
-    void* p,
-    size_t old_bytes pg_attribute_unused(),
-    size_t new_bytes
-) {
-    if (!p) {
-        return MemoryContextAllocExtended(
-            CurrentMemoryContext, new_bytes, MCXT_ALLOC_HUGE
-        );
-    }
-    return repalloc_huge(p, new_bytes);
-}
+#define PGCH_IMPLEMENTATION
+#include "pg-clickhouse-decode.h"
+#include "pg-clickhouse-encode.h"
 
-static void
-pg_chc_pfree(
-    void* ud pg_attribute_unused(),
-    void* p,
-    size_t bytes pg_attribute_unused()
-) {
-    if (p) {
-        pfree(p);
-    }
-}
-
-const chc_alloc pg_chc_alloc = {
-    .ud      = NULL,
-    .alloc   = pg_chc_palloc,
-    .realloc = pg_chc_repalloc,
-    .free    = pg_chc_pfree,
-};
-
-/* power-of-10 lookup; CH bounds DateTime64 / Decimal scale to [0, 9] */
-const int64_t pow10i[10] = {
-    1, 10, 100, 1000, 10000, 100000, 1000000, 10000000, 100000000, 1000000000,
-};
+#include "binary_internal.h"
 
 /*
  * output_format_native_write_json_as_string exists on the server from
@@ -93,12 +50,3 @@ server_supports_json_as_string(const chc_client* c) {
     }
     return false;
 }
-
-void
-raise_chc(const chc_err* err, int sqlstate, const char* prefix) {
-    const char* m = (err && err->msg[0]) ? err->msg : "unknown error";
-
-    ereport(
-        ERROR, errcode(sqlstate), errmsg("pg_clickhouse: %s%s", prefix ? prefix : "", m)
-    );
-}