Re: List of known_good_options

[Carlo De Falco]

On 16 Dec 2015, at 21:40, Rik <address@hidden> wrote:

> The issue is that "TimeStepSize" and "TimeStepNumber" have been added to
> the list of known options.  However, I don't see these in the Matlab
> documentation for odeset.  When odeset was moved into the core I got rid of
> options which were not Matlab compatible, but downgraded any usage of an
> unknown option from an error to a warning so they can always be used. 
> Should these two stay as known options?

In the long term these options should go.
They are used to control 'integrate_constant'
and 'integrate_n_steps' stepping strategies
that we want removed so they shouldn't be there.

In the patch applied yesterday I put them 
back because the ode solvers currently assume 
they may be given, so they check for them via odeget.

odeget currently issues a warning when user asks for
an option with an unexpected name, so any time ode23/ode45
were run a warning about asking for "TimeStepSize" and
"TimeStepNumber" would be raised.

I'm thinking about a clean solution about what to do
when odeget is passed an unexpected option name.

Matlab considers this an error but we agreed that we want 
to allow Octave odeset to set options not included in the list
we therefore also need odeget to be able to query them.

I thought adding back "TimeStepSize" and "TimeStepNumber"
would be easiest approach to avoid warnings while I was 
thinking about what to do with odeget/odeset.

I'll work on removing 'integrate_n_steps' and 'integrate_constant'
steppers from ode/private and therefore checking for the related
options in ode23/ode45 asap. In the mean time, if you are annoyed
by the test failure I could either mark it as an expected failure
or increase the expected number of options returned by odeset to 25
in the test.

c.