[erlang-questions] modular otp concerns

[Tuncer Ayaz]

On Tue, 18 Feb 2014 21:18:51 +0100, Björn-Egil Dahlberg wrote:

> =) I have no illusions and don't stop a discussion on my account.
> I'm just saying I won't be part of it, and I'm guessing nor anyone
> else from OTP, until after 17.0. I'm just hoping someone is taking
> notes.

To be honest, I don't think there's much to discuss without you being
involved :). I was under the impression that this project may already
be cooking, and could soon be revealed without a chance of community
involvement, so I opened the thread.

> 2014-02-18 20:24 GMT+01:00 Tuncer Ayaz <tuncer.ayaz@REDACTED>:
>
>> On Tue, Feb 18, 2014 at 7:02 PM, Björn-Egil Dahlberg wrote:
>> > On 2014-02-18 18:46, Michael Truog wrote:
>> >>
>> >> On 02/18/2014 09:30 AM, Björn-Egil Dahlberg wrote:
>> >>>
>> >>> If/when we split the OTP repository, and this split should be on
>> >>> the level where we have specific versions i.e. on application
>> >>> level. It has solve the issue of easy delivery of new patches and
>> >>> release on these specific applications. Some have voiced a concern
>> >>> that it will simply be to many repos and we should instead group
>> >>> them, i.e. Orber + cos* etc. I say fine, if so then those
>> >>> applications should have one version (and be a single
>> >>> application). One versioned entity per repository thank you.
>> >>
>> >> My main concern is: if the OTP repository is split, we still need
>> >> the concept of a "version set", so a set of versions for all of OTP
>> >> that are known to work together, just to help people avoid any
>> >> potential for instability. With that concept in place, it shouldn't
>> >> be a problem either way, and you are probably aware of the issue,
>> >> but I wanted to mention it due to the added complexity that not
>> >> having a "version set" concept can cause. If you don't have that
>> >> concept, of versions grouped into a release, basically, then you
>> >> run into a combinatorial problem during testing, which makes
>> >> testing take longer (i.e., for all end-users).
>> >>
>> >>
>> >
>> > I'm aware. The "set" you talk about would ideally be controlled by
>> > the application dependencies.
>> >
>> > But OTP releases would be the set you talk about though and a top
>> > repo, i.e. otp would control the set. I imagine that we would still
>> > have releases just like today with a set of applications that we
>> > have tested together.
>> >
>> > As for patches, those are tested with the dependencies specified,
>> > typically with those applications in the previous release. It would
>> > have the same stability as before.
>> >
>> > It is unfortunate that this debate got started now. I have actually
>> > zero time to spend on it .. i'll can do is watch it spin into
>> > another dimension.
>>
>> Sorry, Bjoern-Egil, and it's totally fine to postpone the discussion
>> until after R17 is released. Let's just just pause and start a new
>> discussion when R17 is out.
>>
>> BTW, your arguments for splitting up the repo sound reasonable.