Yeah. Removing them from IDE is PRECISELY the goal :). On Fri, Aug 5, 2022 at 11:01 AM Ash Berlin-Taylor <a...@apache.org> wrote:
> And by removing them/making them dynamic the ide will think they don't > even exist anymore. > > (Also: these have been showing as deprecated in IDEs for ~2 years; If > someone was going to update their code they would have already) > > On 5 August 2022 09:58:38 BST, Ash Berlin-Taylor <a...@apache.org> wrote: >> >> I think the _goal_ is to not have the deprecated classes show up in IDEs >> - i.e. we want to discourage people from using them. >> >> On 5 August 2022 09:44:42 BST, "Kamil Breguła" <dzaku...@gmail.com> >> wrote: >>> >>> I am concerned that the use of dynamic attributes will prevent the IDE >>> from recognizing these deprecated modules and marking them in the IDE. >>> This is what it looks like in my IDE now: https://imgur.com/a/OnRjiKs >>> >>> pt., 5 sie 2022 o 00:28 Ferruzzi, Dennis <ferru...@amazon.com.invalid> >>> napisał(a): >>> >>>> > add dynamic attributes to __init__ of airflow.contrib, >>>> airflow.operators, airflow.hooks. (to resolve to provider operators). >>>> >>>> This sounds promising to me. >>>> >>>> ------------------------------ >>>> *From:* Jarek Potiuk <ja...@potiuk.com> >>>> *Sent:* Thursday, August 4, 2022 6:52 AM >>>> *To:* dev@airflow.apache.org >>>> *Subject:* RE: [EXTERNAL][DISCUSS] Move "contrib" and all old classes >>>> to a separate package >>>> >>>> >>>> *CAUTION*: This email originated from outside of the organization. Do >>>> not click links or open attachments unless you can confirm the sender and >>>> know the content is safe. >>>> >>>> > This is the issue though. apache-airflow does not depend on >>>> apache-airflow-providers-google >= 8.0, so you can still be on the same >>>> version of the provider on 2.3 and 2.4, but by _only_ upgrading >>>> apache-airflow package your DAG now breaks. >>>> >>>> Correct. The difference is that by only upgrading apache-airflow alone >>>> you will break it. And I see why this might be seen as "problematic". >>>> >>>> But I have another idea. Why - rather than keeping the deprecations in >>>> the code we load such deprecated we add dynamic attributes to __init__ of >>>> airflow.contrib, airflow.operators, airflow.hooks. (to resolve to provider >>>> operators). >>>> >>>> Such dynamic attributes will be invisible to autocompletion, will not >>>> be documented with APIs nor visible in the source code (Except deep >>>> inspection of __init__ in those packages). >>>> >>>> This has no compatibility breaking potential whatsoever >>>> >>>> >>>> J. >>>> >>>> >>>> On Thu, Aug 4, 2022 at 3:47 PM Ash Berlin-Taylor <a...@apache.org> >>>> wrote: >>>> >>>>> On Thu, Aug 4 2022 at 13:59:23 +02:00:00, Jarek Potiuk < >>>>> ja...@potiuk.com> wrote: >>>>> >>>>> When you migrate to Airflow 2.4 and it links to the 8.0 version of >>>>> Google provider (assume 2.3 linked to 7) you have to make changes to make >>>>> it work in backwards incompatible way - you have to downgrade google >>>>> provider to 7.0. You still can do it, but it is not "out-of-the-box". >>>>> >>>>> >>>>> This is the issue though. apache-airflow does not depend on >>>>> apache-airflow-providers-google >= 8.0, so you can still be on the same >>>>> version of the provider on 2.3 and 2.4, but by _only_ upgrading >>>>> apache-airflow package your DAG now breaks. >>>>> >>>>> >>>>>
