potiuk commented on PR #33598:
URL: https://github.com/apache/airflow/pull/33598#issuecomment-1688599061

   >  Wow that’s a big leap. Would it be reasonable/meaningful to include some 
of the versions between 4.1 and 1.0?
   > yeah i feel that is more reasonable.
   
   Comment:
   
   I don't have, any problems with big jumps like that to be honest - 
especially in providers. IMHO - if we need a new functionality from a new 
version of dependency and we are already (via constraints) using newer version 
of it, this is no brainer and we should simply bump the min version - even if 
it is few major versions.
   
   We're doing it quite often when we see that we want to add a new feature or 
make sure we do not want to trigger som bugs. For example I **just** 2 days ago 
Bumped Celery to 5.3.0+ - relatively new releease from Jun - because it 
unblocked some other dependencies and actually allowed us to upgrade 
opentelemetry to newer versions (we were held back before).
   
   I think there are very little risks involved in  bumping individual 
dependencies with their minimum versions. For 9X% of our users they are anyhow 
already using constraints, and if we are bumping the min version to somethign 
"below" our constraints, it's rather "safe" bet.  Of course there might be edge 
cases that users won't be able to upgrade because they are somehow tied to 
older versions of the dependency, but:
   
   a) risk of it is really small
   
   b) effect of it is that simply won't be able upgrade to a new version of 
provider (but if they are somehow linked to the old version they will know that 
during the installation and even - if they properly follow our best practices 
of ours, the provider will be automatically downgraded to the right version by 
`pip` resolver when they install older version of the dependency - they might 
even not notice that they downgrade happened.
   
   c) they can still use the old version of provider for a long time according 
to our policies
   
   d) the problem with keeping min-version for a long time is that we do not 
actually KNOW if the min version is right. We might have added a new feature or 
incompatible code, accidentally and might have not realised that because we are 
using the latest "matching" version of the dependency in our tests. So bu 
bumping min version we are actually making it MORE likely that someone will 
avoid problems by using older version that we broke compatibility with 
accidentallly
   
   So for me I see no reason at all to avoid such jump. IMHO This is completely 
no brainer that we should bump min version unconditionally here if we want to 
use the new feature.


-- 
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