Consuming clink from source with FetchContent¶

The examples one directory up consume an installed clink through find_package(clink). This project is the other route: it pulls clink's source from GitHub at a pinned tag with CMake's FetchContent, builds the engine inside its own build tree and links it as clink::core, the same target the installed package exports. Nothing needs installing first, and the consumer chooses clink's build options.

It is a complete project to copy from: CMakeLists.txt is the integration, and fetched_pipeline.cpp is an ordinary self-checking pipeline (bounded source, key by customer, one-second tumbling window summing amounts, collecting sink) that proves the linked engine runs. It registers with CTest and exits non-zero if the window totals differ from the input.

Prerequisites¶

The same as building clink itself: a C++23 compiler, CMake 3.24 or newer, and the pinned Arrow and Parquet under CLINK_DEPS_PREFIX (host default ~/.clink-deps, bootstrapped once with scripts/build-arrow.sh from a clink checkout). clink's CMake finds that prefix on its own. No connector SDKs are needed, because this project builds clink with CLINK_BUILD_IMPLS=OFF.

Build and run¶

cd docs/consumer-examples/fetchcontent
cmake -S . -B build                      # clones clink at the pinned tag
cmake --build build --parallel 10        # builds clink::core, then the program
ctest --test-dir build --output-on-failure

The first configure clones clink and the third-party sources clink fetches itself. The build compiles the engine, so expect it to take as long as a core-only clink build.

Choosing what to build¶

Two cache variables select the source, and two ordinary variables set before the fetch shape clink's build:

Setting	Default	Meaning
`CLINK_GIT_REPOSITORY`	`https://github.com/orhaugh/clink.git`	Repository FetchContent clones
`CLINK_GIT_TAG`	`v0.10.0`	Tag or branch to build. v0.9.0 is the first release that supports build-tree consumption; v0.8.0 predates it. A commit hash works once `GIT_SHALLOW` is dropped from the declaration
`CLINK_BUILD_IMPLS`	`OFF`	Connector and backend impls. Set it on and link `clink::kafka` and friends to use one
`CLINK_BUILD_SQL`	`OFF`	The SQL frontend

clink's own tests and in-tree examples are off by default when clink is not the top-level project, so this project does not mention them. Set CLINK_BUILD_TESTS to ON before the fetch to build them too.

To build against a clink checkout you already have, rather than cloning, point FetchContent at it. clink's CI does exactly this with the commit under test:

cmake -S . -B build -DFETCHCONTENT_SOURCE_DIR_CLINK=/path/to/clink

When to prefer find_package¶

Fetching from source suits a single application that wants one build, one toolchain and a pinned clink. Several projects on one machine, or a runtime image that ships the clink CLI and clink_node, are better served by installing clink once and consuming it with find_package(clink), as the consumer examples do.