[petsc-dev] PETSc release soon, request for input on needed fixes or enhancements

[Barry Smith]

> On Mar 16, 2016, at 9:35 PM, Matthew Knepley <knepley at gmail.com> wrote:
> 
> On Wed, Mar 16, 2016 at 2:40 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
>    Thanks. This is definitely a bug, but very different from my understanding of what was reported (which is why one should always tell someone how to reproduce a problem in a bug report). It is because the PushFormat is done regardless of whether it is a new viewer or just the same viewer being set again.
> 
>    Absolutely needs to be fixed. I don't like the idea of cluttering up the object landscape of PETSc even more by adding new objects such as "viewer plus format" as was suggested.
> 
>    I need to think about a way to fix this that results in less cluttered code, nor more cluttered code.
> 
> To me it seems like this PushFormat is occurring in completely the wrong place. For example
> 
> PetscErrorCode  KSPMonitorSetFromOptions(KSP ksp,const char name[],const char help[], const char manual[],PetscErrorCode (*monitor)(KSP,PetscInt,PetscReal,void*))
> {
>   ierr = PetscOptionsGetViewer(PetscObjectComm((PetscObject)ksp),((PetscObject)ksp)->prefix,name,&viewer,&format,&flg);CHKERRQ(ierr);
>   if (flg) {
>     ierr = PetscViewerPushFormat(viewer,format);CHKERRQ(ierr);
>     ierr = KSPMonitorSet(ksp,monitor,viewer,(PetscErrorCode (*)(void**))PetscViewerDestroy);CHKERRQ(ierr);
>   }
> }
> 
> This clearly cannot be what we want because two monitors of different formats cannot coexist on the same Viewer,
> even if its a shared Viewer like PETSC_VIEWER_ASCII_WORLD. I think the format should be passed to MonitorSet,
> and put into the monitor struct (someone likes arrays more then structs):
> 
>     ksp->monitorformats[ksp->numbermonitors]          = format;
> 
> and then the monitor call should push and pop the format around the monitor routine.

  Sure, it is just a question of writing the simplest clean code that does this.


> 
>    Matt
>  
> 
>   Barry
> 
> 
> 
> > On Mar 16, 2016, at 2:18 PM, Lawrence Mitchell <lawrence.mitchell at imperial.ac.uk> wrote:
> >
> >
> >
> > On 16/03/16 18:56, Barry Smith wrote:
> >>
> >>  Please send me a command line option and example (one in the repository?) where this problem happens.
> >
> > I had slightly misread the code (MG setup doesn't call
> > KSPSetFromOptions more than once even when PCSetUP_MG is called
> > again), but here is an example:
> >
> > cd src/ts/examples/tutorials
> > make ex26
> > ./ex26 -da_refine 1  -ts_max_steps 10 -ts_rtol 1e-3 -ts_atol 1e-3
> > -ts_type beuler -pc_type asm -pc_asm_blocks 30 -sub_ksp_type
> > richardson -sub_ksp_monitor
> >
> > Cheers,
> >
> > Lawrence
> >
> >
> 
> 
> 
> 
> -- 
> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
> -- Norbert Wiener