Groovy Compilation: AST and CPS from a Jenkins point of view

When you are trying to plugin in your pure Groovy library into your Jenkins CI pipelines, you often run into all kinds of compilation issues due to that Jenkins is leverging a customized CPS groovy compiler to parse pipeline steps instead of using native Groovy compiler. Even worse, those compilation issues printed from Jenkins console is not verbose enough for you figure out what actually went wrong. Here, I am writting down some frustrations I have encountered and also share some tips that I found helpful. To show a clear view and only demonstrate what I know, I am showing a list of bullet points.

  1. [ Tip ] I found following way of writting a pure Groovy library works pretty well for me. First, directory structure is as follow:
<GIT_ROOT>
|
|--src
    |--com
        |--<CompanyName>
                |
                |-------lib
                |        |---------deps
                |        |          |---<Self vendored dependent binary goes here>
                |        |
                |        |---------tests
                |        |          |---<Groovy Unit tests go here>
                |        |
                |        |---------test.sh
                |        |          |---<I usually have a bash to automate running of all tests>
                |        |
                |        <Groovy classes go in this level>
                |
                |-------<You might also have some groovy at this level>

I manage all tests in src/com/<CompanyName>/lib/tests. To run single tests, execute a command like groovy -classpath src/com/<CompanyName>/lib/ src/com/<CompanyName>/lib/tests/SomeTest.groovy. What worth mentioning here is that pure groovy is compiled by official Groovy compiler in AST. However, the groovy pipelines on Jenkns are compiled with a customized Groovy compiler wich compiles in CPS (continue passing style).

  1. [ Frustration ] As Jenkins is using a customized Cps Groovy compiler, compiling our pure Groovy lib is likely hit a bug here and there, especially when Jenkins pipeline is still evolving and undergoing fast pace of development. In addition, there is too much freedoom when writting in pure groovy, which increases chances of hitting issue with integrating into Jenkins pipelines.

  2. [ Frustration ] The @NonCPS annotation, as recommended by Cloudbees support for pure Groovy code block, doesn’t solve problems magically. Take s step back, lets think about and discuss what is this annotation for ? The point I heard most is that adding this annotation enforces Jenkins to run annotated block of code in a one-off manner. To further clarify, let me begin with claiming that the key here is about serialization. Normal objects created in Jenkins pipelines must be serializable so that Jenkins can serialize the state of the script in order to save it onto and restore from disk. Use case like: Jenkins need to install new patches and ongoing jobs need to be temporarily paused, and restarted after Jenkins upgrade. Then, lets go back to our original point: with @NonCPS annotation for non-standard Jenkins groovy, it tells Jenkins that such bocks are not serializable and need to be run till finish. Also, under this annotation, code bloakcs will be compiled in AST. Ideally, that should help the situation when we want to plug in a bunch of Groovy classes. However, from what I have tried, @NonCPS is not doing the magic as claimed.

-10000TB


References:
1, Image credit: code compiler image (https://blog.programminghub.io/blog/2017/02/28/offline-compiler-apps-solution/)

comments powered by Disqus