Re: new ODE functions

[c.]

On 16 Dec 2016, at 19:18, John W. Eaton <address@hidden> wrote:

> On 12/16/2016 12:40 PM, Rik wrote:
> 
>>> On 29 Nov 2016, at 15:56, c. <address@hidden> wrote:
>>> 
>>> I will now start working more thoroughly on improving the coding style,
>>> but I would really appreciate if someone could look in detail at this
>>> code, especially the autoconf part, and suggest changes/improvements.
> 
> I made a number of style fixes for the C++ code.
> 
> The Debian packages for Sundials do not have KLU enabled.  I installed my own 
> local copy with it enabled, but it might be nice if we could disable whatever 
> features KLU provides if it is not found while still allowing the rest to 
> build.  I could work on that if you agree.
> 
> jwe


JWE,

Yes, that makes sense.
But changing the config test will not be enough.
Whenever the Jacobian is provided as sparse, __ode15__ 
will invoke the KLU solver, so you should also remove that
call if KLU is not available.

I would also like to have a very clear warning when SUNDIALS
is available but KLU is not.

In preliminary tests ode15i is more than 10 faster than DASPK
for systems with sparse jacobian when using KLU, while it gives
negligible advantages without it.

c.