potiuk commented on PR #30727:
URL: https://github.com/apache/airflow/pull/30727#issuecomment-1535974200

   > I don't personally see this PR as being exceptionally dangerous or 
difficult to review beyond a standard PR that shuffles a few classes around. I 
think we've all seen and reviewed plenty of PRs that are WAY larger.
   
   Yeah. We did. But we often very carefully debated when to merge the change 
in the past and the question of cherry-pickability has been an important 
decision making factor. Usually when we did big refactors like that we parked 
such changes to before the cut-off in the past. This is nothing new (usually it 
happened when we had bigger refactors across the whole codebase though). This 
one is a bit different as it is localized.
   
   > In my opinion, we should proceed and merge this PR to main. If it results 
in issues while cherry-picking changes, we can ask the involved contributors 
for help in navigating those. Does it add more responsibility to the Release 
managers? Yes, it does. Nonetheless, I believe it is the lesser of the two 
evils because we should expect only a few changes to be cherry-picked in patch 
releases. Conversely, in the other scenario, the contributor will need to keep 
an eye on ALL the changes.
   
   Not really. I disagree. It only takes to PARK this change and redo it once 
before the cuttof. It does not have to be continuosly rebased. With any change 
like that there are two things that can be carried over:
   
   1) the actually changed code
   2) the IDEA what should be changed 
   
   I like the idea behind the change. It's cool. But if the changed code is 
merged now, it will either carry the risk in the bugfix branch or make it more 
difficult for release managers. The only "evil" we are weighting here is 
whether to implement the idea now or later. There is no more "evil" happening 
than choosing implementing the idea later (And yes, sinking some work already 
done by @o-nikolas) . However I'd argue (and I've done that multiple times in 
the past) that redoing such a change some time later that you already have 
proven that it can be done and looking at your past changes, is multiple times 
faster (and less error prone) than doing it for the first time. I even 
sometimes deliberately drop the change like that I've done to do it again 
starting from newer codebase, and it is usually cleaner and super-fast to do.
   
   So my proposal is not to carry the code and look at all the other changes, 
but to carry the idea of the change and implement it when the time is ripe. I 
even offer to do it together with @o-nikolas - when we will be extracting the 
k8S executor to provider - as @eladkal suggested, then we can both together 
iterate on the change and it will even have the nice angle of pair-programming 
aspect which we could do together.
   
   Maybe that is the best way?
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to