[ 
https://issues.apache.org/jira/browse/FLINK-8036?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16246715#comment-16246715
 ] 

Chesnay Schepler commented on FLINK-8036:
-----------------------------------------

I can see the benefit, but as it stands I'm against it. As Fabian said, this 
would entail changing many integral parts of the system, not to mention how 
users build custom flink versions.

Furthermore, there are just really trivial workarounds for the described 
use-case:
Step1: compile Flink (without checkstyle etc.)
Step2: invoke checkstyle, findbugs, tests etc. for each submodule in a separate 
process

This is rather trivial to implement and should buy you plenty of time.

> Consider using gradle to build Flink
> ------------------------------------
>
>                 Key: FLINK-8036
>                 URL: https://issues.apache.org/jira/browse/FLINK-8036
>             Project: Flink
>          Issue Type: Improvement
>            Reporter: Ted Yu
>
> Here is summary from Lukasz over this thread 
> (http://search-hadoop.com/m/Beam/gfKHFVh4NM151XIu1?subj=Re+DISCUSS+Move+away+from+Apache+Maven+as+build+tool)
>  w.r.t. performance boost from using gradle:
> Maven performs parallelization at the module level, an entire module needs
> to complete before any dependent modules can start, this means running all
> the checks like findbugs, checkstyle, tests need to finish. Gradle has task
> level parallelism between subprojects which means that as soon as the
> compile and shade steps are done for a project, and dependent subprojects
> can typically start. This means that we get increased parallelism due to
> not needing to wait for findbugs, checkstyle, tests to run. I typically see
> ~20 tasks (at peak) running on my desktop in parallel.
> Flink should consider using gradle - on Linux with SSD, a clean build takes 
> an hour.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to