Split vortex-duckdb from the extension

Signed-off-by: Adam Gutglick <adam@spiraldb.com>

Author: AdamGS

Cargo.lock

@@ -26,6 +26,7 @@ members = [
     "vortex-bench",
     "vortex-datafusion",
     "vortex-duckdb",
+    "vortex-duckdb-extension",
     "vortex-cuda",
     "vortex-cuda/cub",
     "vortex-cuda/ffi",

Cargo.toml

@@ -0,0 +1,26 @@
+[package]
+name = "vortex-duckdb-extension"
+version = { workspace = true }
+description = "C-compatible Vortex DuckDB extension artifacts"
+homepage = { workspace = true }
+repository = { workspace = true }
+authors = { workspace = true }
+license = { workspace = true }
+publish = false
+keywords = { workspace = true }
+include = { workspace = true }
+edition = { workspace = true }
+rust-version = { workspace = true }
+categories = { workspace = true }
+readme = { workspace = true }
+
+[lib]
+name = "vortex_duckdb_extension"
+path = "src/lib.rs"
+crate-type = ["staticlib", "cdylib"]
+
+[dependencies]
+vortex-duckdb = { workspace = true }
+
+[lints]
+workspace = true

vortex-duckdb-extension/Cargo.toml

@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright the Vortex contributors
+
+#![expect(clippy::missing_safety_doc)]
+
+use std::ffi::c_char;
+use std::ffi::c_void;
+
+/// Global symbol visibility in the Vortex extension:
+/// - Rust functions use C ABI with "_rust" suffix (e.g., vortex_init_rust)
+/// - C++ wrapper functions have the expected name without suffix (e.g., vortex_init)
+/// - C++ wrappers are annotated with DUCKDB_EXTENSION_API to ensure global visibility
+/// - C++ wrappers call the corresponding Rust functions
+///
+/// This ensures DuckDB can find the symbols when loading the extension.
+///
+/// The DuckDB extension ABI initialization function.
+#[unsafe(no_mangle)]
+pub unsafe extern "C" fn vortex_init_rust(db: *mut c_void) {
+    unsafe { vortex_duckdb::initialize_extension_from_raw(db) };
+}
+
+/// The DuckDB extension ABI version function.
+/// This function returns the version of the DuckDB library the extension is built against.
+#[unsafe(no_mangle)]
+pub extern "C" fn vortex_version_rust() -> *const c_char {
+    vortex_duckdb::duckdb_library_version()
+}
+
+/// An additional function we export to expose the version of the extension itself to C++ code.
+#[unsafe(no_mangle)]
+pub extern "C" fn vortex_extension_version_rust() -> *const c_char {
+    vortex_duckdb::extension_version()
+}

vortex-duckdb-extension/src/lib.rs

@@ -21,7 +21,7 @@ links = "duckdb"
 [lib]
 name = "vortex_duckdb"
 path = "src/lib.rs"
-crate-type = ["staticlib", "cdylib", "rlib"]
+crate-type = ["rlib"]
 
 [dependencies]
 async-fs = { workspace = true }

vortex-duckdb/Cargo.toml

@@ -18,20 +18,26 @@ Link against the precompiled DuckDB release build.
 cargo build -p vortex-duckdb
 ```
 
+Build the C-compatible DuckDB extension artifacts with:
+
+```bash
+cargo build -p vortex-duckdb-extension
+```
+
 ### Debug Build
 
 Opt into DuckDB debug build: `VX_DUCKDB_DEBUG=1`.
 
 ```bash
-VX_DUCKDB_DEBUG=1 cargo build -p vortex-duckdb
+VX_DUCKDB_DEBUG=1 cargo build -p vortex-duckdb-extension
 ```
 
 ### AddressSanitizer & ThreadSanitizer
 
 Enable both ASAN & TSAN: `VX_DUCKDB_SAN=1`.
 
 ```bash
-VX_DUCKDB_DEBUG=1 VX_DUCKDB_SAN=1 cargo build -p vortex-duckdb
+VX_DUCKDB_DEBUG=1 VX_DUCKDB_SAN=1 cargo build -p vortex-duckdb-extension
 ```
 
 ## Environment Variables

vortex-duckdb/README.md

@@ -15,12 +15,6 @@
 extern "C" {
 #endif // __cplusplus
 
-extern void vortex_init_rust(duckdb_database db);
-
-extern const char *vortex_version_rust(void);
-
-extern const char *vortex_extension_version_rust(void);
-
 extern void duckdb_table_function_to_string(void *bind_data, duckdb_vx_string_map map);
 
 extern

vortex-duckdb/include/vortex.h

@@ -3,8 +3,8 @@
 
 #![expect(clippy::missing_safety_doc)]
 
-use std::ffi::CStr;
 use std::ffi::c_char;
+use std::ffi::c_void;
 use std::sync::LazyLock;
 use std::sync::OnceLock;
 
@@ -79,40 +79,23 @@ pub fn initialize(db: &DatabaseRef) -> VortexResult<()> {
     db.register_copy_function()
 }
 
-/// Global symbol visibility in the Vortex extension:
-/// - Rust functions use C ABI with "_rust" suffix (e.g., vortex_init_rust)
-/// - C++ wrapper functions have the expected name without suffix (e.g., vortex_init)
-/// - C++ wrappers are annotated with DUCKDB_EXTENSION_API to ensure global visibility
-/// - C++ wrappers call the corresponding Rust functions
-///
-/// This ensures DuckDB can find the symbols when loading the extension.
-///
-/// The DuckDB extension ABI initialization function.
-#[unsafe(no_mangle)]
-pub unsafe extern "C" fn vortex_init_rust(db: cpp::duckdb_database) {
+/// Initialize the DuckDB extension from a raw DuckDB database pointer.
+pub unsafe fn initialize_extension_from_raw(db: *mut c_void) {
     init_tracing();
-    let database = unsafe { Database::borrow(db) };
+    let database = unsafe { Database::borrow(db.cast()) };
 
     database
         .register_vortex_scan_replacement()
         .vortex_expect("failed to register vortex scan replacement");
     initialize(database).vortex_expect("Failed to initialize Vortex extension");
 }
 
-/// The DuckDB extension ABI version function.
-/// This function returns the version of the DuckDB library the extension is built against.
-#[unsafe(no_mangle)]
-pub extern "C" fn vortex_version_rust() -> *const c_char {
+/// Returns the version of the DuckDB library the extension is built against.
+pub fn duckdb_library_version() -> *const c_char {
     unsafe { cpp::duckdb_library_version() }
 }
 
-/// An additional function we export to expose the version of the extension itself to C++ code.
-#[unsafe(no_mangle)]
-pub extern "C" fn vortex_extension_version_rust() -> *const c_char {
-    // We do some fiddly macros here to get ourselves a _static_ C-style string.
-    // Otherwise, we'd be leaking memory.
-    unsafe {
-        CStr::from_bytes_with_nul_unchecked(concat!(env!("CARGO_PKG_VERSION"), "\0").as_bytes())
-    }
-    .as_ptr()
+/// Returns the version of the Vortex DuckDB extension.
+pub fn extension_version() -> *const c_char {
+    concat!(env!("CARGO_PKG_VERSION"), "\0").as_ptr().cast()
 }