[petsc-dev] PETSc developers who use weird MPI

Barry Smith bsmith at mcs.anl.gov
Fri Nov 1 21:52:12 CDT 2013


On Nov 1, 2013, at 9:37 PM, Satish Balay <balay at mcs.anl.gov> wrote:

> On Fri, 1 Nov 2013, Barry Smith wrote:
> 
>>> 
>>> The fix is to merge latest 'prbrune/sf-sfbasicops' in the same pathway it got into 'barry/saws’
>> 
>>   Thanks. That will likely fix the problem.
>> 
>>   I made the mistake of believing “it is ok to have some branch hang around unmerged with master etc for several months” myth
>> 
> 
> This mode works better when there are no dependencies between multiple
> branches.

   i.e. in fantasy land.
> 
> If there are dependencies like 'prbrune/sf-sfbasicops ->
> prbrune/mat-matcolor -> barry/saws ' we need extra effort to keep
> track of these paths - and make sure the fixes - if any - get
> propogated the same way.
 
   Like how are we going to do that? Every time someone merges to next I check it and decide if that branch also needs to be merged into all my long living branches. Yeah like that is going to happen. 

   The reason I had to merge all that stuff into saws was that saws could not merge into next because those branches so changed next. I had to merge them into saws before I could get saws into next. But I missed 1/2 a one (somehow) getting an outdated verson of the sf-sfbasics into saws.

   Barry




> [and then - these branches can hang arround
> for a long time]
> 
> [I guess these are the costs of having cleaner retraceable history and
> stabler master branch.]...
> 
> Satish




More information about the petsc-dev mailing list