> we are talking about the same thing & on the same page with versioning, 
> though releasing clients with each Airflow release confuses me.

Not with every release. I am talking about releasing it with every
release "if it is needed". I do not see anything confusing here. The
logic is simple:

* release Airflow
* do we need to release Python API Client -> if yes, release
* do we need to release Go API Client -> if yes, release
* release Docker Image

I hope this "algorithm" is pretty straightforward.

J.


On Tue, Nov 15, 2022 at 12:21 PM Sumit Maheshwari
<[email protected]> wrote:
>
> @Jarek Potiuk we are talking about the same thing & on the same page with 
> versioning, though releasing clients with each Airflow release confuses me.
>
> On Tue, Nov 15, 2022 at 4:04 PM Jarek Potiuk <[email protected]> wrote:
>>
>> >
>> > There had been a discussion around this earlier, though I couldn't find 
>> > the thread. So if we want to release clients with each Airflow release, 
>> > then we've to move away from current semantic versioning, something which 
>> > we had decided in that thread.
>>
>> Why ? I think we do not have to do that  - the point I made is that we
>> do not "have" to release the client - but we SHOULD do it if there is
>> an outstanding change.
>>
>> As I mentioned above:
>>
>> 1) when we release 2.5.0 Airflow -> we always make 2.5.0 API - no
>> exceptions here.
>> 2) when we release 2.5.n (n != 0) - we MIGHT release API if there are
>> some bug-fixes that are relevant (for example we found out that there
>> is a documentation fix needed or Python generator fixed some mistake
>> in generated code (happened in the past). But this could easily be the
>> case that when we release Airflow 2.5.3 we also release Python API
>> client 2.5.1 and Python Go client 2.5.4 (for example, because in the
>> meantime we independently released fixed to Python Go Client 3 times.
>>
>> Generally our users should install the latest released Python/ Go
>> clients for the 2.5 line - no matter which patchlevel they have.
>>
>> I do not think it requires any changes to our agreed SemVer approach.
>>
>> But maybe I have not thought about something?
>>
>> J.

Reply via email to