On Mar 3, 2012, at 4:54 PM, Jed Brown wrote:

> On Tue, Feb 28, 2012 at 14:51, Barry Smith <bsmith at mcs.anl.gov> wrote:
> Let me restate the plan:
> 
> We want to retain the current hierarchy: IS, Vec, Mat, DM, PC/KSP, SNES, TS. 
> To do this we "cheat" slightly by having a PC/KSP part of the DM, a SNES part 
> of DM and a TS part of DM that are implemented outside of the base _p_DM. In 
> a sense they are subclasses of the base DM that support PC/KSP, SNES, and TS. 
>  The fact that they are subclasses is represented in the method names 
> DMSNESSetFunction() etc (note that Jed has it wrong above by putting the Set 
> before the XXXX).
> 
> This is mostly implemented now, I'll push soon.
> 
> What was the rationale behind the silent early exit from 
> SNESComputeJacobian() if snes->ops->computejacobian as not set? I would 
> prefer explicitly setting a Jacobian routine that does nothing to this 
> behavior which I think is error prone.

  If no compute Jacobian routine was provided it just always uses the original 
matrix that was set into the SNESSetJacobian.  I can live with removing this 
feature, I agree it is a little dangerous.




   Barry



Reply via email to