You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@arrow.apache.org by Micah Kornfield <em...@gmail.com> on 2021/05/02 04:52:14 UTC

Re: [DISCUSS] experimental repos

Jorge, do you think  linking the doc or copying this to someplace on the
Arrow website to make it more visible would make sense?

On Tue, Apr 27, 2021 at 2:51 PM Wes McKinney <we...@gmail.com> wrote:

> This process seems pretty reasonable to me. Thanks for writing up the
> document.
>
> On Tue, Apr 27, 2021 at 10:56 AM Micah Kornfield <em...@gmail.com>
> wrote:
>
>> Hi Jorge,
>> I think especially for the second case, it might be better to keep things
>> on branches in the repro even if they aren't quite mergeable.  Even for
>> the
>> first case, I would potentially aim for the "closest possible" repo with a
>> new branch.
>>
>> I think standalone repos tend to indicate a higher level of
>> maturity/commitment to casual observers that these experiments are meant
>> to
>> represent (even clearly documented experimental repos).  I think once the
>> project reaches some level of maturity and there is broader community
>> commitment the decision/technical part can be made to spinning new repos.
>>
>> Again, this isn't based on any data or any real experience so I'm happy
>> for
>> whoever wants to try out the process to go the way they see fit.  I also
>> guess there might be some technical implications for some languages based
>> on branches/repos.
>>
>> Cheers,
>> Micah
>>
>> On Mon, Apr 26, 2021 at 8:44 AM Jorge Cardoso Leitão <
>> jorgecarleitao@gmail.com> wrote:
>>
>> > Hi Micah,
>> >
>> > For code that is mergeable, I would say that a branch is superior, as it
>> > keeps lineage and thus enables rebasing. IMO there are two use-cases for
>> > this mechanism:
>> >
>> > * create a new component from scratch (e.g. Ballista, bindings to
>> language
>> > Z, python-datafusion).
>> > * re-write an existing implementation / component, with no intentions of
>> > being mergeable in the traditional sense.
>> >
>> > Operationally, the first case would be things like mv new_repo/*
>> > old_repo/component/
>> > The second case would be for things like rm -rf old_repo/*; mv
>> new_repo/*
>> > old_repo/ or "this repo is now official code, old is in maintenance
>> mode".
>> >
>> > Best,
>> > Jorge
>> >
>> >
>> >
>> >
>> > On Mon, Apr 26, 2021 at 6:40 AM Micah Kornfield <em...@gmail.com>
>> > wrote:
>> >
>> >> Hi Jorge,
>> >> Thanks for doing this.
>> >>
>> >> One question that springs to mind: For work that is mostly intended to
>> be
>> >> merged back to an existing Arrow repo what are the trade-offs between
>> >> brand
>> >> new repos as compared to a separate branch in existing one in an
>> existing
>> >> one?
>> >>
>> >> Cheers,
>> >> Micah
>> >>
>> >> On Sun, Apr 25, 2021 at 9:31 PM Jorge Cardoso Leitão <
>> >> jorgecarleitao@gmail.com> wrote:
>> >>
>> >> > Hi,
>> >> >
>> >> > As discussed in other threads (rust sync and parquet2), I would like
>> to
>> >> > open the discussion around opening repos for experimental work that
>> may
>> >> or
>> >> > may not be merged / used.
>> >> >
>> >> > The rationale is that we incentivise work to be conducted within ASF
>> and
>> >> > Apache Arrows' governance, thereby clarifying the context and
>> >> governance of
>> >> > that work, while still offering the freedoms that people enjoy when
>> >> > creating a new repo on their personal accounts.
>> >> >
>> >> > I wrote a draft proposal here: [1]
>> >> >
>> >> > [1]
>> >> >
>> >> >
>> >>
>> https://docs.google.com/document/d/1rDTWezkKkmOQ3HQeX8NhaXbKZLeuwcyFdC4ZA7ZatDY/edit?usp=sharing
>> >> >
>> >> > Best,
>> >> > Jorge
>> >> >
>> >>
>> >
>>
>