lance-format
diff --git a/‎.github/workflows/MainDistributionPipeline.yml‎
Lines changed: 20 additions & 0 deletions b/‎.github/workflows/MainDistributionPipeline.yml‎
Lines changed: 20 additions & 0 deletions
diff --git a/‎.github/workflows/build.yml‎
Lines changed: 0 additions & 71 deletions b/‎.github/workflows/build.yml‎
Lines changed: 0 additions & 71 deletions
diff --git a/‎.gitmodules‎
Lines changed: 5 additions & 0 deletions b/‎.gitmodules‎
Lines changed: 5 additions & 0 deletions
diff --git a/‎CLAUDE.md‎
Lines changed: 11 additions & 85 deletions b/‎CLAUDE.md‎
Lines changed: 11 additions & 85 deletions
diff --git a/‎CMakeLists.txt‎
Lines changed: 25 additions & 0 deletions b/‎CMakeLists.txt‎
Lines changed: 25 additions & 0 deletions
@@ -0,0 +1,20 @@
+name: Main Extension Distribution Pipeline
+on:
+  push:
+    branches: [main]
+  pull_request:
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || '' }}-${{ github.base_ref || '' }}-${{ github.ref != 'refs/heads/main' && github.sha || '' }}
+  cancel-in-progress: true
+
+jobs:
+  duckdb-stable-build:
+    name: Build extension binaries
+    uses: duckdb/extension-ci-tools/.github/workflows/_extension_distribution.yml@v1.3.2
+    with:
+      duckdb_version: v1.3.2
+      ci_tools_version: v1.3.2
+      extension_name: lance
+      exclude_archs: "wasm_mvp;wasm_eh;wasm_threads"
@@ -1,3 +1,8 @@
+[submodule "duckdb"]
+	path = duckdb
+	url = https://github.com/duckdb/duckdb
+	branch = main
 [submodule "extension-ci-tools"]
 	path = extension-ci-tools
 	url = https://github.com/duckdb/extension-ci-tools
+	branch = main
@@ -12,27 +12,29 @@ This is a DuckDB extension written in Rust that enables native SQL querying of L
 ```bash
 # Initial setup (only needed once)
 git submodule update --init --recursive
-make configure
 
 # Build commands
-make release         # Production build → build/release/lance.duckdb_extension
-make debug          # Debug build → build/debug/lance.duckdb_extension
-make clean          # Clean build artifacts
-make clean_all      # Clean everything including configure
+GEN=ninja make release         # Production build → build/release/lance.duckdb_extension
+GEN=ninja make debug          # Debug build → build/debug/lance.duckdb_extension
+GEN=ninja make clean          # Clean build artifacts
+GEN=ninja make clean_all      # Clean everything including configure
 
 # Quick Rust checks (without full build)
 cargo check
 cargo clippy --all-targets --all-features
 ```
 
 ### Testing
+
+release build can be slow, use `test_debug` for quick test.
+
 ```bash
 # Run all tests (builds release and runs sqllogictest)
-make test
+GEN=ninja make test
 
 # Run with specific build
-make test_debug     # Test with debug build
-make test_release   # Test with release build
+GEN=ninja make test_debug     # Test with debug build
+GEN=ninja make test_release   # Test with release build
 
 # Run DuckDB with extension for manual testing
 duckdb -unsigned -c "LOAD 'build/release/lance.duckdb_extension'; SELECT * FROM lance_scan('test/test_data.lance');"
@@ -41,7 +43,7 @@ duckdb -unsigned -c "LOAD 'build/release/lance.duckdb_extension'; SELECT * FROM
 ### Development Iteration
 ```bash
 # Fast iteration cycle
-cargo build --release && make test_release
+cargo build --release && make test_debug
 
 # Check for issues without full build
 cargo clippy --all-targets --all-features
@@ -73,59 +75,11 @@ The extension follows a three-layer architecture:
 The extension uses different names to avoid conflicts:
 - **Extension name**: `lance` (what users see)
 - **Rust crate name**: `lance_duckdb` (avoids crate conflict)
-- **Entry point**: `lance_init_c_api` (generated from extension name)
-
-This is controlled in `Makefile`:
-```makefile
-EXTENSION_NAME=lance
-RUST_CRATE_NAME=lance_duckdb
-```
-
-#### Async Bridge Pattern
-Lance uses async APIs while DuckDB extensions are synchronous:
-```rust
-// Create runtime in init
-let runtime = Arc::new(Runtime::new()?);
-
-// Block on async operations
-let dataset = runtime.block_on(async {
-    Dataset::open(&path).await
-})?;
-```
-
-#### Current Data Loading Strategy
-**Important**: Currently loads ALL data into memory during `init()`:
-```rust
-// In LanceScanInitData
-batches: Arc<Mutex<Vec<RecordBatch>>>,  // All data loaded here
-```
-
-This works for small-medium datasets but needs streaming for production.
-
-### Dependency Version Constraints
-
-**Critical**: Arrow versions MUST match exactly between Lance and DuckDB:
-- Lance 0.32.1 → Arrow 55.1
-- No version ranges allowed (use exact versions)
-
-### Known Limitations
-
-1. **Replacement Scan**: Not implemented due to `duckdb-rs` API limitations
-   - Users must use `lance_scan('file.lance')` instead of `FROM 'file.lance'`
-   - Requires access to raw database handle not exposed by duckdb-rs
-
-2. **Type Conversion**: Currently simplified to strings
-   - Production needs direct Arrow→DuckDB memory mapping
-
-3. **Memory Usage**: Loads entire dataset into memory
-   - Needs streaming implementation for large datasets
 
 ## Test Data & Testing
 
 ### Test Dataset
 Location: `test/test_data.lance`
-- 5 records: id (1-5), name (Alice-Eve), age (25-45), score (78.5-95.5)
-- Created by: `cargo run --example create_test_data`
 
 ### Test Format
 Uses DuckDB's sqllogictest format in `test/sql/`:
@@ -135,37 +89,9 @@ Uses DuckDB's sqllogictest format in `test/sql/`:
 
 ## Common Issues & Solutions
 
-### Build Failures
-1. **Cargo hangs**: Kill with `pkill -9 cargo rustc`, then `make clean`
-2. **Version mismatch**: Check `TARGET_DUCKDB_VERSION=v1.3.2` in Makefile
-3. **Missing symbols**: Ensure `USE_UNSTABLE_C_API=1` is set
-
 ### Extension Loading
 ```sql
 -- Always use -unsigned flag for local builds
 duckdb -unsigned
 LOAD 'build/release/lance.duckdb_extension';
 ```
-
-### Type Errors
-Current implementation converts to strings. If seeing type mismatches, check:
-1. Arrow schema extraction in `bind()`
-2. Type mapping in `types.rs`
-3. Data conversion in `func()`
-
-## Future Improvements Priority
-
-1. **High Priority**
-   - Streaming reads (replace Vec<RecordBatch> with iterator)
-   - Proper Arrow→DuckDB type mapping
-   - Predicate pushdown to Lance
-
-2. **Medium Priority**
-   - Replacement scan when API available
-   - Projection pushdown
-   - Better error messages
-
-3. **Low Priority**
-   - Write support (COPY TO)
-   - Vector index integration
-   - Statistics for query optimization
@@ -0,0 +1,25 @@
+cmake_minimum_required(VERSION 3.5)
+
+# Set extension name here
+set(TARGET_NAME lance)
+
+# No external dependencies for now
+
+set(EXTENSION_NAME ${TARGET_NAME}_extension)
+set(LOADABLE_EXTENSION_NAME ${TARGET_NAME}_loadable_extension)
+
+project(${TARGET_NAME})
+include_directories(src/include)
+
+set(EXTENSION_SOURCES src/lance_extension.cpp src/lance_scan.cpp)
+
+build_static_extension(${TARGET_NAME} ${EXTENSION_SOURCES})
+build_loadable_extension(${TARGET_NAME} " " ${EXTENSION_SOURCES})
+
+# No external libraries to link
+
+install(
+  TARGETS ${EXTENSION_NAME}
+  EXPORT "${DUCKDB_EXPORT_SET}"
+  LIBRARY DESTINATION "${INSTALL_LIB_DIR}"
+  ARCHIVE DESTINATION "${INSTALL_LIB_DIR}")