[petsc-dev] workflow diagram

[Barry Smith]

On Apr 30, 2014, at 2:03 PM, Jed Brown <jed at jedbrown.org> wrote:

> Barry Smith <bsmith at mcs.anl.gov> writes:
> 
>>  Should the “Testing/Users” arrow go all the way down next to the
>> next line? Now it looks like maybe it means the testing/users is on
>> the typical feature branch.
> 
> Changed.
> 
>>  The other thing that “bothers” me is the two arrows coming from the
>>  last circle of a branch; one goes “straight” to next which is fine
>>  but the other one goes “straight” to master. I think it would be
>>  clearer if instead that second curve followed the branch line and
>>  then quickly curved up to master. I have shared a stunningly crude
>>  way of doing this (not I don’t intend to have the “bumb” in the
>>  curve I just suck at drawing it.) This still conveys the fact that
>>  the data for both next and master comes from the same circle but
>>  makes it clearer that the data is not “moved” from the final circle
>>  to master until later then it is moved to next; in fact it is not
>>  moved from the branch until it gets into master. Hopefully you can
>>  do that shape better then I.
> 
> Hmm, I like showing that after 'next' is rewound, the graph contains no
> knowledge of 'next' integration (apart from stability).  What I want to
> communicate is that the DECISION to merge to master is based on EVIDENCE
> obtained from the integration in 'next’.

   Right, this is why I proposed delaying the “curve” of the arrow to master until nearly when it goes into master. Currently since the arrow goes straight from the circle to master it looks like the decision WHEN to merge to master was made at the same time the merge was made to next when the decision is actually made after the change has sat in next for a while.

   Barry