[petsc-dev] SNESCOMPOSITE with a VI underneath

[Barry Smith]

> On Apr 15, 2015, at 3:54 AM, Patrick Farrell <patrick.farrell at maths.ox.ac.uk> wrote:
> 
> On 14/04/15 23:14, Barry Smith wrote:
>>    The problem is that since most SNES methods just ignore the bounds if we
>>    do not set the type to a method that supports bounds the user will not be
>>    solving the problem they think they are solving.  Perhaps we should just
>>    have all methods that ignore the bounds error out if the bounds are set
>>    rather than ignore them?
> 
> Should this check live in SNESVISetVariableBounds (where we can compare the
> SNESType against a list of SNESTypes that support bounds), or in every
> SNESSolve_*?

  It has to be in SNESSolve_ because the user might change the SNESType after they call SNESVISetVariableBounds()

> 
> Patrick