开源软件名称（OpenSource Name）：

开源软件地址(OpenSource Url)：

开源编程语言(OpenSource Language)：

开源软件介绍(OpenSource Introduction)：

Monorepo sample

This repo showcases how one could structure and build monorepos with either Apache Maven or Bazel.

The repo

A couple of toy applications under apps/ depend on shared code under libs/.

.

├── apps
│   ├── app1
│   └── app2
└── libs
    ├── lib1
    └── lib2

Continous integration

The ci/ sub folder contains a self-contained ci-stack for building the monorepo with Jenkins.

Maven and Bazel: a comparison of basic actions

Maven is really not a monorepo-native build tool (e.g. lacks trustworthy incremental builds, can only build java code natively, is recursive and struggles with partial repo checkouts) but can be made good use of with some tricks and usage of a couple of lesser known command line switches.

Notes

• Maven is a recursive build tool that needs an reference to the root module with --file when building from a sub folder. In contrast will Bazel discover the entire project from any sub folder and packages to be built can be specified both absolutely from root (//) and relative the current working directory.
• Bazel doesn't build the so-called implicit deploy jar-target (:app1_deploy.jar) if we refrain from building with the :* suffix. Deploy jar is Bazel speak før über-/super-/fat-jar.
• Maven has no trustworthy support for incremental builds, hence one typically always build from a clean state.

Sparse checkouts

A monorepo will naturally grow quite large quite fast and for many reasons engineers will for the most time prefer to only have a subset of the tree checked out at once. Git sparse checkouts provides a mechanism for archieving this.

Executing:

git config core.sparseCheckout true
echo '/*'           >  .git/info/sparse-checkout
echo '!/apps/*'     >> .git/info/sparse-checkout
echo '/apps/app1/*' >> .git/info/sparse-checkout

enables sparse checkouts and configures git to checkout the the whole tree except all apps other than app1. To make it happen, either run a checkout command or to update the work tree in-place:

git read-tree -mu HEAD

Notes for Maven

Sparse checkouts

While Bazel is fine with sparse checkouts, Maven struggles as a sparse checkout typically implies that there will be pom.xml-references to submodules that no longer are present on disc. A hack to work-around this is to use Maven profiles that only activates when the corresponding pom.xml-file is available on disc:

<profile>
  <id>app2</id>
  <activation>
    <file>
      <exists>app2/pom.xml</exists>
    </file>
  </activation>
  <modules>
    <module>app2</module>
  </modules>
</profile>

Build parallelization

Maven requires explicit setup to build modules in parallell. This can be provided as a switch on the command-line (--threads 2) or in .mvn/maven.config.

Versioning

Maven requires you to provide an version for every module in its Project Object Model. This is confusing and superflous in monorepos as the they are typically exclusively versioned by the version control system. This repo takes advantage of Maven's newish support for so-called CI Friendly Versions. This makes it possible to ensure that the version of all Maven modules are set to current git HEAD at build-time:

$ mvn clean install -Drevision=$(git rev-parse HEAD)

To ensure that no CI-friendly version placeholders (<version>${revision}</version>) are present in the pom-files that are installed, this repo utilizes the flatten-maven-plugin to resolve them to their values before install.

Partial builds

Maven's lack of incremental build support can lead to long build times in a large monorepo. The partial-build-plugin can be used to only build modules that has a diff against the trunk branch.