I recall that we were stuck with Maven 3.2.5 for a very long time because of the way Flink shaded dependencies, and that required an extensive rework before we could upgrade. When that work was done, Maven 3.8.6 was new (it was released in June 2022). Maven 3.9 didn't exist at that time, the first version was only released in January 2023). Given how much issues Maven upgrades were, I don't think any contributor wanted to spent time on something that wasn't necessarily broken.
Op wo 19 aug 2026 om 00:17 schreef David Anderson <[email protected]>: > > +1 I can only see upsides to this, but I could be missing something. > > I believe we ended up on 3.8.6 because that was the newest version that > Chesnay deemed trustworthy at the time of the last maven upgrade [1]. > > [1] https://lists.apache.org/thread/jbw3lzzoq5w16ckco3fc9xokycs3f22x > > > On Tue, Aug 18, 2026 at 7:07 AM Gabor Somogyi <[email protected]> > wrote: > > > +1 from my side. > > Just for my own understanding why we've stuck with such an old maven? > > For me it's always a pain to have such an old version. > > > > G > > > > > > On Tue, Aug 18, 2026 at 2:25 PM Martijn Visser <[email protected]> > > wrote: > > > > > Hi all, > > > > > > I'd like to propose moving Flink's build from the currently pinned > > > Maven 3.8.6 to the Maven 3.9 line (3.9.16). This was left for "a > > > separate discussion" when we adopted Apache Parent POM 35 in > > > FLINK-39427 [1] > > > > > > I'd propose to move it because: > > > - Maven 3.8.6 (2022) is aging; 3.9.x is the current stable 3.x line > > > and a prerequisite for future Apache Parent POM upgrades (36+). > > > - It moves us onto Maven's "consumer POM" model behavior that > > > anticipates Maven 4. > > > > > > I've ran a sample/local red/green diff of 3.8.6 version 3.9.16, > > > building flink-s3-fs-hadoop -am on JDK 17 and 11, the full flink-dist > > > -am reactor on JDK 17, and a -Prelease build on > > > Java 11 to match how releases are actually cut. > > > > > > Both are building cleanly on JDK 17 and 11, the shaded uber-jars seem > > > to be byte-identical in contents and the aggregated NOTICE files also > > > seem identical, except the manifest "Created-By" changes. The only > > > change I saw is that the flink-parent pom inlines the release profile > > > plugin config, while 3.9 emits the pom with re-inlining them. Perhaps > > > there will be more issues when opening a PR, but from a first glance > > > it seemed to be OK locally. > > > > > > While running the tests I did see that there is a warning about > > > gpgArguments, which now surfaces. I'll file a separate Jira on this, > > > it's because of the maven-gpg-plugin version we're using. > > > > > > Thanks, > > > > > > Martijn > > > > > > [1] https://issues.apache.org/jira/browse/FLINK-39427 > > > > >
