Add RAPIDS-based devcontainer for velox/cuDF/presto development

.devcontainer/Dockerfile

@@ -0,0 +1,116 @@
+# Velox development container based on RAPIDS devcontainers
+#
+# Uses RAPIDS devcontainer base image (Ubuntu) for compatibility with
+# RAPIDS build utilities and cuDF development.
+
+ARG CUDA_VERSION=12.9
+ARG BASE=rapidsai/devcontainers:latest-cpp-cuda12.9-ucx1.19.0-openmpi5.0.7
+
+FROM ${BASE}
+
+ARG TARGETARCH
+ARG CUDA_VERSION
+
+# Install velox build dependencies
+RUN apt-get update -y && \
+    DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
+    # GCC 14 (avoids GCC 13 false-positive -Wstringop-overflow in fmt)
+    gcc-14 \
+    g++-14 \
+    # Velox dependencies not in RAPIDS base
+    libc-ares-dev \
+    libboost-all-dev \
+    libdouble-conversion-dev \
+    libevent-dev \
+    libfmt-dev \
+    libgflags-dev \
+    libgoogle-glog-dev \
+    liblz4-dev \
+    liblzo2-dev \
+    libprotobuf-dev \
+    libprotoc-dev \
+    protobuf-compiler \
+    librdkafka-dev \
+    libre2-dev \
+    libsnappy-dev \
+    libsodium-dev \
+    libstemmer-dev \
+    libthrift-dev \
+    libxxhash-dev \
+    libzstd-dev \
+    # Build tools
+    bison \
+    flex \
+    libfl-dev \
+    gperf \
+    ninja-build \
+    ccache \
+    && rm -rf /tmp/* /var/tmp/* /var/cache/apt/* /var/lib/apt/lists/*
+
+# Set GCC 14 as default (GCC 13 has false-positive -Wstringop-overflow in fmt)
+RUN update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 140 \
+                        --slave /usr/bin/g++ g++ /usr/bin/g++-14 \
+                        --slave /usr/bin/gcov gcov /usr/bin/gcov-14 && \
+    update-alternatives --set gcc /usr/bin/gcc-14
+
+# Install velox helper scripts
+COPY velox-testing/scripts/devcontainer/* /usr/local/bin/
+RUN chmod 755 /usr/local/bin/build-* /usr/local/bin/configure-* \
+              /usr/local/bin/test-* /usr/local/bin/clean-* /usr/local/bin/post-create \
+              /usr/local/bin/_devcontainer_common_utils.sh
+
+# Environment for velox builds
+# VELOX_DEPENDENCY_SOURCE is read from env by CMake
+# Individual _SOURCE overrides must be passed as -D flags in configure-velox
+ENV VELOX_DEPENDENCY_SOURCE=SYSTEM \
+    CUDA_VERSION=${CUDA_VERSION} \
+    CUDAARCHS="RAPIDS" \
+    DEFAULT_VIRTUAL_ENV=rapids \
+    PYTHON_PACKAGE_MANAGER=pip
+
+# Python environment
+ENV PYTHONSAFEPATH="1" \
+    PYTHONUNBUFFERED="1" \
+    PYTHONDONTWRITEBYTECODE="1"
+
+# Persist bash history to cache
+ENV HISTFILE="/home/coder/.cache/._bash_history"
+
+###
+# sccache configuration
+###
+ENV AWS_ROLE_ARN="arn:aws:iam::279114543810:role/nv-gha-token-sccache-devs" \
+    SCCACHE_REGION="us-east-2" \
+    SCCACHE_BUCKET="rapids-sccache-devs" \
+    SCCACHE_S3_USE_PREPROCESSOR_CACHE_MODE=true \
+    SCCACHE_IDLE_TIMEOUT=0
+
+###
+# sccache-dist configuration
+###
+# Enable sccache-dist by default
+ENV DEVCONTAINER_UTILS_ENABLE_SCCACHE_DIST=1
+# Compile locally if max retries exceeded
+ENV SCCACHE_DIST_FALLBACK_TO_LOCAL_COMPILE=true
+# Retry transient errors 4 times (for a total of 5 attempts)
+ENV SCCACHE_DIST_MAX_RETRIES=4
+# 1hr 59min (to accommodate debug builds)
+ENV SCCACHE_DIST_REQUEST_TIMEOUT=7140
+ENV SCCACHE_DIST_URL="https://${TARGETARCH}.linux.sccache.rapids.nvidia.com"
+
+# Build as much in parallel as possible
+ENV INFER_NUM_DEVICE_ARCHITECTURES=1 \
+    MAX_DEVICE_OBJ_TO_COMPILE_IN_PARALLEL=20
+
+# Add velox environment to bashrc
+RUN echo "export CUDA_VERSION=${CUDA_VERSION}" >> /home/coder/.bashrc && \
+    echo 'export CMAKE_C_COMPILER_LAUNCHER=sccache' >> /home/coder/.bashrc && \
+    echo 'export CMAKE_CXX_COMPILER_LAUNCHER=sccache' >> /home/coder/.bashrc && \
+    echo 'export CMAKE_CUDA_COMPILER_LAUNCHER=sccache' >> /home/coder/.bashrc
+
+# Create build directories before USER switch; chmod 0777 so any uid works after
+# the devcontainer feature remaps the coder uid to match the host user.
+RUN mkdir -m 0777 -p /opt/velox-build /opt/presto-build /opt/fb-deps
+
+USER coder
+WORKDIR /home/coder

.devcontainer/README.md

@@ -0,0 +1,161 @@
+# Velox + Presto Devcontainer
+
+A GPU-ready development environment for building [Velox](https://github.com/facebookincubator/velox) with cuDF acceleration and [Presto Native Execution](https://github.com/prestodb/presto/tree/master/presto-native-execution). Based on the RAPIDS devcontainer image with pre-built cuDF, RMM, and KvikIO.
+
+## Quick Start
+
+```bash
+# 1. Clone sibling repos under the same parent directory
+mkdir ~/code && cd ~/code
+git clone https://github.com/<org>/velox-testing.git
+git clone https://github.com/facebookincubator/velox.git
+git clone https://github.com/prestodb/presto.git
+git clone https://github.com/rapidsai/rmm.git
+git clone https://github.com/rapidsai/cudf.git
+git clone https://github.com/rapidsai/kvikio.git
+
+# 2. Open in VS Code (or any devcontainer-compatible editor)
+code velox-testing
+
+# 3. Reopen in container (pick CUDA 13.1 or 12.9 variant)
+
+# 4. Build
+build-velox            # ~25 min — standalone velox with cuDF
+build-presto           # ~40 min first run (builds FB deps), ~25 min after
+
+# 5. Test
+test-velox             # run velox test suite
+test-presto            # run presto test suite
+```
+
+## Directory Layout
+
+The devcontainer expects this layout on the host:
+
+```
+~/code/
+├── velox-testing/     # this repo (devcontainer workspace root)
+├── velox/             # facebookincubator/velox
+├── presto/            # prestodb/presto
+├── rmm/               # rapidsai/rmm
+├── cudf/              # rapidsai/cudf
+└── kvikio/            # rapidsai/kvikio
+```
+
+All repos are bind-mounted into the container under `/home/coder/`.
+
+## Build Architecture
+
+```
+  ┌───────────────┐           ┌───────────────────────────────┐
+  │  build-velox  │           │         build-presto          │
+  │               │           │                               │
+  │ ~/velox src   │           │  1. FB deps (auto, once)      │
+  │ all BUNDLED   │           │     folly, fbthrift, proxygen │
+  │ MONO_LIB=ON  │           │     → /opt/fb-deps            │
+  │ SHARED=ON    │           │                               │
+  │               │           │  2. rsync ~/velox → submodule │
+  │               │           │     add_subdirectory(velox)   │
+  │               │           │     MONO_LIB=OFF              │
+  └───────┬───────┘           └──────────────┬────────────────┘
+          │                                  │
+          ▼                                  ▼
+  /opt/velox-build/              /opt/presto-build/
+     release/                       presto_server
+     3879 targets                   1978 targets
+```
+
+Both builds consume pre-built **cudf**, **rmm**, and **kvikio** from the RAPIDS devcontainer.
+
+**Standalone velox** bundles all its dependencies (folly, xsimd, Arrow, etc.) — no prerequisite steps.
+
+**Presto** requires Facebook's OSS stack (folly, fbthrift, proxygen, etc.) for its thrift RPC layer. `build-presto` builds these automatically on first run and caches them at `/opt/fb-deps`. Subsequent runs skip this step unless `--rebuild-deps` is passed.
+
+### Why velox builds twice
+
+Presto integrates velox via `add_subdirectory()` with different options (`VELOX_MONO_LIBRARY=OFF`, no testing, etc.). The rsync in `build-presto` copies `~/velox` into presto's git submodule directory so both repos share the same source while keeping presto's git state clean.
+
+## Commands
+
+| Command | Description |
+|---------|-------------|
+| `build-velox` | Build standalone velox with cuDF (fully self-contained) |
+| `build-presto` | Build presto-native-execution (auto-builds FB deps on first run) |
+| `configure-velox` | CMake configure only (for IDE integration) |
+| `test-velox` | Run velox tests via ctest |
+| `test-presto` | Run presto tests via ctest |
+| `clean-velox` | Delete velox build artifacts |
+| `clean-presto` | Delete presto build artifacts |
+
+All commands accept `--help`. Common options:
+
+```bash
+build-velox --debug            # debug build
+build-velox -j 16              # limit parallelism
+build-presto --release         # release build (default)
+build-presto --rebuild-deps    # force rebuild of FB deps
+```
+
+## CUDA Variants
+
+Two devcontainer configurations are provided:
+
+| Path | CUDA | Base Image |
+|------|------|------------|
+| `.devcontainer/cuda13.1/` | 13.1 | `rapidsai/devcontainers:latest-cpp-cuda13.1-*` |
+| `.devcontainer/cuda12.9/` | 12.9 | `rapidsai/devcontainers:latest-cpp-cuda12.9-*` |
+
+VS Code will prompt you to choose when opening the workspace. The `CUDAARCHS` environment variable defaults to `RAPIDS`, which expands to all RAPIDS-supported architectures.
+
+## Build Outputs
+
+| Build | Location | Contents |
+|-------|----------|----------|
+| FB deps | `/opt/fb-deps/` | folly, fbthrift, proxygen (auto-built by `build-presto`) |
+| Velox | `/opt/velox-build/release/` | Mono shared library, tests |
+| Presto | `/opt/presto-build/` | `presto_server` binary, tests |
+
+Build directories are under `/opt/` so they don't pollute mounted source trees and persist across container sessions (unless the container is recreated).
+
+## Dependency Resolution
+
+Velox has many dependencies. Each can be `SYSTEM` (pre-installed) or `BUNDLED` (built from source via FetchContent). The build scripts set these automatically:
+
+| Dependency | Velox Build | Presto Build | Source |
+|------------|------------|--------------|--------|
+| folly | BUNDLED | SYSTEM | Built from source / `/opt/fb-deps` |
+| xsimd | BUNDLED | SYSTEM | Built from source / `/opt/fb-deps` |
+| cudf | SYSTEM | SYSTEM | RAPIDS pre-built |
+| rmm | SYSTEM | SYSTEM | RAPIDS pre-built |
+| kvikio | SYSTEM | SYSTEM | RAPIDS pre-built |
+| Arrow | BUNDLED | BUNDLED | Built from source |
+| DuckDB | BUNDLED | BUNDLED | Built from source |
+| GTest | BUNDLED | BUNDLED | Built from source |
+| simdjson | BUNDLED | BUNDLED | Built from source |
+| geos | BUNDLED | BUNDLED | Built from source |
+
+## Workarounds
+
+The build scripts include several workarounds for toolchain issues:
+
+- **GCC 14** instead of GCC 13: avoids false-positive `-Wstringop-overflow` in system `fmt` v9.
+- **`-DCMAKE_CXX_SCAN_FOR_MODULES=OFF`**: CMake 4.x + GCC 14 + Ninja triggers `-fmodules-ts` which causes GCC 14 ICE (segfault).
+- **`-no-pie` linker flag** (presto only): fbthrift static archives have construction vtables with hidden visibility from virtual inheritance in `apache::thrift` exception classes. The linker cannot resolve `R_X86_64_PC32` relocations against hidden symbols in PIE executables.
+- **`-Wno-error=nonnull`** (presto only): presto's `SystemConnector.cpp` triggers a false-positive `this` null check warning.
+
+## Scripts
+
+All scripts live in `scripts/devcontainer/` and are installed to `/usr/local/bin/` in the container image. They share common functions via `_common.sh`.
+
+```
+scripts/devcontainer/
+├── _common.sh         # Shared constants (CUDA archs, RAPIDS detection)
+├── build-velox        # Standalone velox build (all deps bundled)
+├── build-presto       # Presto + velox build (includes FB deps)
+├── configure-velox    # CMake configure only
+├── test-velox         # Run velox tests
+├── test-presto        # Run presto tests
+├── clean-velox        # Clean velox build dir
+├── clean-presto       # Clean presto build dir
+└── post-create        # Devcontainer post-create hook
+```

.devcontainer/cuda12.9/devcontainer.json

@@ -0,0 +1,61 @@
+{
+  "name": "velox-dev-cuda12.9",
+  "build": {
+    "context": "${localWorkspaceFolder}/..",
+    "dockerfile": "${localWorkspaceFolder}/.devcontainer/Dockerfile",
+    "args": {
+      "CUDA_VERSION": "12.9",
+      "BASE": "rapidsai/devcontainers:latest-cpp-cuda12.9-ucx1.19.0-openmpi5.0.7"
+    }
+  },
+  "features": {
+    "ghcr.io/rapidsai/devcontainers/features/rapids-build-utils:latest": {}
+  },
+  "overrideFeatureInstallOrder": [
+    "ghcr.io/rapidsai/devcontainers/features/rapids-build-utils"
+  ],
+  "initializeCommand": ["/bin/bash", "-c", "mkdir -p ${localWorkspaceFolder}/../.{aws,cache,config/pip,local/share/velox-testing-cuda12.9-venvs}"],
+  "workspaceFolder": "/home/coder",
+  "workspaceMount": "source=${localWorkspaceFolder},target=/home/coder/velox-testing,type=bind,consistency=consistent",
+  "mounts": [
+    "source=${localWorkspaceFolder}/../velox,target=/home/coder/velox,type=bind,consistency=consistent",
+    "source=${localWorkspaceFolder}/../presto,target=/home/coder/presto,type=bind,consistency=consistent",
+    "source=${localWorkspaceFolder}/../rmm,target=/home/coder/rmm,type=bind,consistency=consistent",
+    "source=${localWorkspaceFolder}/../cudf,target=/home/coder/cudf,type=bind,consistency=consistent",
+    "source=${localWorkspaceFolder}/../kvikio,target=/home/coder/kvikio,type=bind,consistency=consistent",
+    "source=${localWorkspaceFolder}/../.aws,target=/home/coder/.aws,type=bind,consistency=consistent",
+    "source=${localWorkspaceFolder}/../.cache,target=/home/coder/.cache,type=bind,consistency=consistent",
+    "source=${localWorkspaceFolder}/../.config,target=/home/coder/.config,type=bind,consistency=consistent",
+    "source=${localWorkspaceFolder}/../.local/share/velox-testing-cuda12.9-venvs,target=/home/coder/.local/share/venvs,type=bind,consistency=consistent"
+  ],
+  "hostRequirements": {
+    "gpu": "optional"
+  },
+  "runArgs": [
+    "--rm",
+    "--name", "${localEnv:USER:anon}-velox-dev-cuda12.9",
+    "--ulimit", "nofile=500000"
+  ],
+  "postCreateCommand": "/home/coder/velox-testing/scripts/devcontainer/post-create",
+  "postAttachCommand": [
+    "/bin/bash",
+    "-c",
+    "if [ ${CODESPACES:-false} = 'true' ]; then . devcontainer-utils-post-attach-command; . rapids-post-attach-command; fi"
+  ],
+  "customizations": {
+    "vscode": {
+      "extensions": [
+        "ms-vscode.cpptools",
+        "ms-vscode.cmake-tools",
+        "llvm-vs-code-extensions.vscode-clangd"
+      ],
+      "settings": {
+        "cmake.buildDirectory": "/opt/velox-build/${buildType}",
+        "cmake.configureOnOpen": false
+      }
+    }
+  },
+  "containerEnv": {
+    "VELOX_DEPENDENCY_SOURCE": "SYSTEM"
+  }
+}