[petsc-dev] consequences of setting KSP_NORM_TYPE to NONE

[Barry Smith]

> On Jun 12, 2016, at 10:53 PM, Jed Brown <jed at jedbrown.org> wrote:
> 
> Barry Smith <bsmith at mcs.anl.gov> writes:
> 
>>   This looks like a reasonable fix for me. The manual page should indicate that it changes the convergence test if you change from no norm.
> 
> Alternatively, KSPConvergedDefault could just exit immediately instead
> of erroring with KSP_NORM_NONE?  Why do these options need to be
> non-orthogonal such that they need this kind of manipulation?

  Less code is, of course, always most appealing. Currently KSPConvergedDefault() has

 if (normtype == KSP_NORM_NONE) SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_ARG_WRONGSTATE,"Use KSPConvergedSkip() with KSPNormType of KSP_NORM_NONE");

presumably this is so that a user won't mistakenly set the convergence test after setting the norm to none and assume that the test will work? But does this really provide anything of value?