Boost logo

Boost :

Subject: Re: [boost] Fwd: Boost.Python build error in 1.58 RC1, 2 and 3 (OK in Beta 1)
From: Jürgen Hunold (jhunold_at_[hidden])
Date: 2015-04-15 09:24:53


Hi Andrey,

Am Mittwoch, 15. April 2015, 15:52:13 schrieb Andrey Semashev:
> On Wednesday 15 April 2015 15:32:12 Vladimir Prus wrote:
> > I don't think there's a primarily technical problem. No matter how release
> > process goes, we either need to assume that 'master' branch of any
> > component is good to be released at all times, or coordinate with
> > maintainers of 120 components to determine which revision of 'master' is
> > really good to be released.
> >
> > You can guess which approach release managers would prefer.
>
> I'm not arguing against taking master as a starting point for a release;
> it's a reasonable compromise. What I'm saying is that once release process
> has started there should be no restriction on updating master in submodules
> (i.e. the freeze), the release should branch off the main code base and get
> finished regardless of the modifications developers make. It is possible
> that master is broken at the point of branching - and part of release
> process is to fix it for the release or revert to a last known good
> version. But the probability of breakage is greatly reduced with this
> approach.

The main problem is the necessary regression testing needed to actually verify
that the regressions
a) are indeed fixed
b) don't introduce new failures.

And our current testing system is hardcoded to "master" and "develop"
branches. And can't test anything else, as it relies on the test runners
admins to set up which branch to test.

> Note that when I say "branch" above I don't necessarily mean "git branch".
> In fact, it may be more convenient to use forks to isolate the code to be
> released from the main code base and use pull requests for patches that
> need to get into the release.

As soon as we get an up-to-date regression testing and reporting system, we
can use any branching scheme you can imagine. And a lot more probably.
That would mean using either Jenkins or BuildBot as a centralized server and
some dashboard reporting tool for visualisation.
But I don't know anyone with enough spare time to set up such a system without
being paid and working full time. Not to mention the hardware necessary for
this...

Yours,

Jürgen

-- 
* Dipl.-Math. Jürgen Hunold  ! 
* voice: ++49 4257 300       ! Fährstraße 1
* fax  : ++49 4257 300       ! 31609 Balge/Sebbenhausen
* jhunold_at_gmx.eu             ! Germany

Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk