1.8 release planning

[Tim Graham]

I'd like to kickoff the discussion on the timetable and process for the 1.8 release. I am also volunteering to be the release manager.

First, a retrospective on the 1.7 release with planned release dates and (actual):

Jan. 20: alpha (Jan. 22)
March 6: beta (March 20)
May 1: RC (June 26)
May 15: final (Sept. 2)

One observation I have is that each stage of the release does not really do a good job at accurately reflecting our belief about the quality of the code. For example, we have an "alpha" in order to have a major feature freeze, but we still allow a significant amount of minor features (3 months worth in the last release) such that the alpha and beta are hardly comparable. Likewise, we had little confidence that the "RC" would actually be released without further changes, but rather we needed to do the release in order to get to the stage where we would only backport release blocking bugs. Therefore, I am going to propose returning to a process that is closer to what's documented in the Release cycle docs [1]. The idea is to front-load all feature work to pre-alpha so that we can become more conservative with backports sooner.

Here is my proposed schedule:

Jan. 12: alpha
  - Feature freeze including minor features (minor features were allowed until beta in the past)
  - fork stable/1.8.x from master (in the past we forked after beta, but now that we'd no longer accept minor features after alpha, we'd need to fork sooner).
  - I picked this date since it is after the end of the year when I imagine many people are on holiday and therefore able to contribute more to open source.
  - Non-release blocking bug fixes may be backported at the committer's discretion after this date.

Feb. 16: beta
  - Only release blocking bugs are allowed to be backported after this date.
  - Aggressively advertise it for testing

March 16: release candidate
  - Hopefully a true release candidate. If there is still a consistent stream of release blockers coming in at this date; we'd release beta 2 to encourage further testing and push the release candidate date out ~1 month.

March 30: final
  - Release a final as long as the release blocker stream is sufficiently low. If not, give an update about the status and make a plan as to how to proceed from there.

On a related note, I believe we should give some guidance on our thinking regard LTS. Currently our docs say, "Django 1.4, supported until at least March 2015." If we adopt 1.8 as the next LTS, I propose to support 1.4 until 6 months after 1.8 is released, which would be at least September 2015. Like 1.4, we'd advertise LTS support for 1.8 for at least 3 years after it's released with a decision on the next LTS to be made as we approach that date.

Feedback on the proposed schedule and handling of the LTS cycle would be appreciated!

If you have any major features you plan to shepherd for this cycle, please ensure they are listed on the roadmap: https://code.djangoproject.com/wiki/Version1.8Roadmap

[1] https://docs.djangoproject.com/en/dev/internals/release-process/#release-cycle

[Josh Smeaton]

I was wondering when this discussion was going to happen - it's good to see some dates finally proposed, thanks Tim.

The LTS proposal is a good one. It should give people enough time (6 months) to jump to the next LTS release. I also think that 1.8 should be a good candidate for LTS. Migrations and App Loading should have enough time to mature after a full release in the wild, and the proposed 1.8 features seem mainly focused on stability and compatibility rather than big sweeping public changes.

Perhaps it'd be a good idea to write a 1.4 - 1.8 migration guide to ease the transition between LTS releases? There have been some concerns raised on the ML lately about backwards compatibility and the difficulty in porting between versions. The current way of migrating to a new release involves reading each release note and checking your code base for those changes, and move one release at a time. Maybe a simplified all-in-one document that shows the most common backwards compatibility changes, and has guidance on how to address each of them.

[Tim Graham]

If you are upgrading from 1.4, I would recommend going step by step through each intermediate Django version. I don't believe trying to upgrade straight away to 1.8 would be easier. I'd encourage anyone on 1.4 to start doing intermediate upgrades soon, so if there are still undetected regressions from 1.4, we can start solving them now.

[Josh Smeaton]

I didn't mean to imply jumping from 1.4 - 1.8 and then fixing all the issues would be the best way forward. I was thinking along the lines of a one page common upgrade issues that summarised the back-compat changes from the previous releases, separated under each release header.

Anyways, this is slightly off-topic (sorry), I'll move this discussion to a new topic or discuss in IRC.

[Florian Apolloner]

On Saturday, October 18, 2014 1:17:58 AM UTC+2, Josh Smeaton wrote:

 I was thinking along the lines of a one page common upgrade issues that summarised the back-compat changes from the previous releases, separated under each release header.

We already have on page per release for that. Duplicating this into yet another page doesn't seem to be that helpful.

[Mattias Linnap]

My impression from the 1.7 release schedule was that there were many release blockers found in RC stage, and RC with known bugs released because of three combined reasons:

* The release had a number of big new features, including app loading and migrations.

* In alpha and beta stages, the testers tend to be developers working on the new release itself. Once RC is released, the release is seen as mostly bug free, and many more Django users start using and testing the new release in production on new greenfield projects and non-mission critical sites where having the latest and greatest features is better than stumbling on a few bugs. This production testing helps to discover many new bugs, some of them release blockers.

* Due to the high number of release blockers being discovered and fixed, a new version identifier was released to simplify reporting new bugs in Trac against the version that does or does not have these bugs.

I know that the Django project tries to stay conservative and not recommend betas and RCs for production use, since people might get upset if their mission critical sites break. But I think the release and testing process would benefit from the core developers giving their actual "best guess" towards the stability of a beta and RC version, and let users judge the acceptable risk of deploying it in production. There are plenty of hobby projects, internal sites, and new not-yet-released sites out there where a few days of downtime or hacking together workarounds isn't a big problem, but for example data loss would be bad. With a description that is less conservative than "never use it in production", more people might join in on testing and reporting bugs before an RC.

[Tim Graham]

Spot on, that's why I proposed a total feature freeze at alpha instead of beta and becoming more conservative with backports earlier in the release cycle. I hope that advertising 1.8 as the next LTS will help with getting more testers earlier in the release cycle.

[Aymeric Augustin]

Hello,

This timeline was discussed by the core team today at Django under the Hood and approved.

The support period for Django 1.4 will extend for six months after the release of Django 1.8.

The recommended way to migrate from 1.4 to 1.8 is to go through 1.5, 1.6, and 1.7. I encourage you to start this process early if you expect it to take time.

-- 

Aymeric.

--
You received this message because you are subscribed to the Google Groups "Django developers (Contributions to Django itself)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to django-develop...@googlegroups.com.
To post to this group, send email to django-d...@googlegroups.com.
Visit this group at http://groups.google.com/group/django-developers.
To view this discussion on the web visit https://groups.google.com/d/msgid/django-developers/1ace477f-1ead-4554-bb8c-bb86d6347679%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.