October 2014 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] does next model mess up our histories
by Barry Smith 02 Oct '14

02 Oct '14

On Oct 2, 2014, at 9:44 AM, Jed Brown <jed(a)jedbrown.org> wrote: > Barry Smith <bsmith(a)mcs.anl.gov> writes: >> Yup. Jed is saying our histories should be terrible. > > ;-) > > There is a balancing act between "quality" of long-term history, utility > of near-term history, and development effort/opportunity for mistakes. > I personally think the sweet spot is to prefer not to rebase that which > has been merged to 'next’. But this is just a fluke of our testing approach. If I had 15 machines I could test all myself and not put major broken things into next.

1 0

Re: [petsc-dev] does next model mess up our histories
by Karl Rupp 02 Oct '14

02 Oct '14

>> I second more frequent feature releases. Currently we release every ~12 >> months. Releasing every 6 months is certainly within reach. It should also >> help us with finishing features up on a regular basis rather than having them >> in a half-done state for too long (of course I'm also referring to myself >> here). > > As long as we don't wait for a feature - we should be able to do > frequent timely releases [master is always supporsed to be in a > releaseable state :)] :-) > One issue is: we currently support only 1 previous release at a > time. For faster releases - would this cause grief? [wrt users > requiring frequent upgrades, and us requiring to support more > releases] The rate of new features will remain roughly the same, so the time between eventual interface changes will also be roughly the same. Conversely, it will be easier for users to upgrade, since the differences between versions are smaller. If we keep the current policy of supporting the latest release and occasionally advise users on how to fix their problems with older releases, we should still be fine with a 6 months release cycle. Best regards, Karli

1 0

Re: [petsc-dev] does next model mess up our histories
by Jed Brown 02 Oct '14

02 Oct '14

Satish Balay <balay(a)mcs.anl.gov> writes: > One issue is: we currently support only 1 previous release at a > time. For faster releases - would this cause grief? [wrt users > requiring frequent upgrades, and us requiring to support more > releases] A few more patches might need to be applied to older maint branches. But the rate of API change shouldn't increase just because we release more frequently, so it shouldn't really be harder.

1 0

Re: [petsc-dev] does next model mess up our histories
by Satish Balay 02 Oct '14

02 Oct '14

On Thu, 2 Oct 2014, Jed Brown wrote: > Satish Balay <balay(a)mcs.anl.gov> writes: > > > On Thu, 2 Oct 2014, Jed Brown wrote: > > > >> Satish Balay <balay(a)mcs.anl.gov> writes: > >> > I think is ok to be a bit more messy in next. Sure there is a tradeoff > >> > - but you get better debugging in master :) > >> > >> If bugs make it to 'master', the process has failed. > > > > If we have to do patch releases - the process has failed? > > Of course. Every bug is a failure of internal quality control. It's > not possible to avoid bugs entirely, but we should strive to catch them > early. I.e., saying that we won't worry about bugs in 'next' because we > can fix them in 'master' defeats the purpose of the workflow. I didn't advocate not debugging bugs in next. All I adovcated is: its ok to have a few cases (not allways) reverts (of feature branches) in next that does affect bisection a bit [but I don't think by much - as this is usually recent history debugging] - in favor a cleaner history that helps in long history debugging later on.. Satish

1 0

Re: [petsc-dev] does next model mess up our histories
by Satish Balay 02 Oct '14

02 Oct '14

On Thu, 2 Oct 2014, Karl Rupp wrote: > On 10/02/2014 04:34 PM, Jed Brown wrote: > > Satish Balay <balay(a)mcs.anl.gov> writes: > > > I disagree on 'reset next' every few days as a fix to this problem. I > > > suspect > > > Jed will have a different objection for this reset.. > > > > It makes it harder to follow changes on 'next' (including whether > > certain features get fixed) and users can't simply "git pull". I think > > it's useful to have 'next' persist between releases (we could do more > > frequent feature releases). > > I second more frequent feature releases. Currently we release every ~12 > months. Releasing every 6 months is certainly within reach. It should also > help us with finishing features up on a regular basis rather than having them > in a half-done state for too long (of course I'm also referring to myself > here). As long as we don't wait for a feature - we should be able to do frequent timely releases [master is always supporsed to be in a releaseable state :)] One issue is: we currently support only 1 previous release at a time. For faster releases - would this cause grief? [wrt users requiring frequent upgrades, and us requiring to support more releases] Satish

1 0

Re: [petsc-dev] does next model mess up our histories
by Jed Brown 02 Oct '14

02 Oct '14

Satish Balay <balay(a)mcs.anl.gov> writes: > On Thu, 2 Oct 2014, Jed Brown wrote: > >> Satish Balay <balay(a)mcs.anl.gov> writes: >> > I think is ok to be a bit more messy in next. Sure there is a tradeoff >> > - but you get better debugging in master :) >> >> If bugs make it to 'master', the process has failed. > > If we have to do patch releases - the process has failed? Of course. Every bug is a failure of internal quality control. It's not possible to avoid bugs entirely, but we should strive to catch them early. I.e., saying that we won't worry about bugs in 'next' because we can fix them in 'master' defeats the purpose of the workflow.

1 0

Re: [petsc-dev] does next model mess up our histories
by Jed Brown 02 Oct '14

02 Oct '14

Karl Rupp <rupp(a)iue.tuwien.ac.at> writes: >It should > also help us with finishing features up on a regular basis rather than > having them in a half-done state for too long (of course I'm also > referring to myself here). I'm guilty.

1 0

Re: [petsc-dev] does next model mess up our histories
by Jed Brown 02 Oct '14

02 Oct '14

Barry Smith <bsmith(a)mcs.anl.gov> writes: > Yup. Jed is saying our histories should be terrible. ;-) There is a balancing act between "quality" of long-term history, utility of near-term history, and development effort/opportunity for mistakes. I personally think the sweet spot is to prefer not to rebase that which has been merged to 'next'.

1 0

Re: [petsc-dev] does next model mess up our histories
by Karl Rupp 02 Oct '14

02 Oct '14

On 10/02/2014 04:34 PM, Jed Brown wrote: > Satish Balay <balay(a)mcs.anl.gov> writes: >> I disagree on 'reset next' every few days as a fix to this problem. I suspect >> Jed will have a different objection for this reset.. > > It makes it harder to follow changes on 'next' (including whether > certain features get fixed) and users can't simply "git pull". I think > it's useful to have 'next' persist between releases (we could do more > frequent feature releases). I second more frequent feature releases. Currently we release every ~12 months. Releasing every 6 months is certainly within reach. It should also help us with finishing features up on a regular basis rather than having them in a half-done state for too long (of course I'm also referring to myself here). Best regards, Karli

1 0

Re: [petsc-dev] does next model mess up our histories
by Jed Brown 02 Oct '14

02 Oct '14

Satish Balay <balay(a)mcs.anl.gov> writes: > I disagree on 'reset next' every few days as a fix to this problem. I suspect > Jed will have a different objection for this reset.. It makes it harder to follow changes on 'next' (including whether certain features get fixed) and users can't simply "git pull". I think it's useful to have 'next' persist between releases (we could do more frequent feature releases).

1 0