On Mon, 16 Jun 2014, Satish Balay wrote:

> On Mon, 16 Jun 2014, Barry Smith wrote:
> 
> >    When Vijay, or some one else doing something else says; hey Barry please 
> > take a look at my branch xyz and build it and test and add features a, b, 
> > c,  ALL THEY SHOULD NEED TO TELL ME IS branch xyz, I check it out and start 
> > building. If they need to tell me also get branch wyx of moab and branch 
> > abc of XYZ then we are NOT using tools efficiently (let’s just email the 
> > files back and forth). Each branch should have all the information needed 
> > to build for that branch, this simply means each branch has IN the 
> > appropriate xxxx.py package file the correct information about what other 
> > packages versions to use. This is actually easy except for 
> 
> Note: If I read this correctly you are not editing/fixing commiting to
> moab repo.
> 
> To me it looks like you are 'mercurial subrepo' workflow.  Perhaps an
> equivalent thing exists for git. Jed might chime in.
> 
> And yes you can get equivalent thing with 'xxxx.py:gitcommit=xyz' - if
> this is set correctly [not the gitcommit=branch/head thingy]. I think
> this is what Jed was also advocating..

I should modify this statement.

With hg-subrepo - one would do:

hg update [commitid] to get a consistant snapshot of petsc+moab

With 'xxxx.py:gitcommit=xyz' one would do:

git checkout [commit-id]
rm -rf arch*
./configure --download-moab=1

to get equivelent functionality [resulting in a git clone each time. This will 
presumably be
fixedup in the future]

Satish

> 
> >    merging to next and master where you may not want to drag along exactly 
> > the repository branch info that was used with the branch. (this relates to 
> > Vijay’s question).
> 
> Dragging  gitcommit=xyz to master is not a problem [draging gitcommit=head is 
> a problem]
> 
> >    The tools manage the coordination, people don’t mange coordination with 
> > notes on scraps of paper and rumors in the hallway about what commit of 
> > moab they should use with what branch of petsc.
> 
> I agree one should use tools to simplify as much as possible. But configure 
> should
> not  be the only tool that bears this burden. [to me workflow is partly a 
> social
> contract between team members to co-ordinate in a predetermined way].
> 
> > 
> >    Jed has pointed out that a “branch” is not enough information, so what 
> > should we use in the xxx.py so that the correct thing is built when I build 
> > PETSc with a particular branch??? 
> 
> commit-id, tag.
> 
> Branch is also a 'tag' but its a moving tag - [with non-reproduceable
> state].  I think this is Jed's objection. [and mine as well [wrt
> gitcommit=master-head]
> 
> Satish
> 

Reply via email to