Re: ode45 inclusion into core

[Richard Crozier]

On 06/10/15 17:52, Olaf Till wrote:
> On Tue, Oct 06, 2015 at 04:09:54PM +0000, Carlo De Falco wrote:
> > On 6 Oct 2015, at 18:02, Richard Crozier <address@hidden> wrote:
> >
> > > I know this is deprecated, but I much prefer it to the recommended
> > > anonymous function method. Please don't remove it if it already
> > > exists.
> > >
> > > I unfortunately can't remember the exact reasons, but there are
> > > issues with saving anonymous functions and their data to files
> > > that make this way much less efficient in my application
>
> The (probably sole) issues are:
>
> http://savannah.gnu.org/bugs/?45972
>
> http://savannah.gnu.org/bugs/?45969
>
> both of which have supplied patches, so there is hope for next release
> series.
>
> > (I use
> > > the multicore package, or rather a slightly modified version of
> > > it).
> > >
> > > Richard
>
> With parcellfun, only the results, which probably don't contain
> anonymous functions, need to be saved (over a pipe, not to
> files). Since your problem obviously concerns additional arguments,
> not results, it would be absent if you used parcellfun. Which, I
> think, you really should do ...

Prcelfun, as far as I can tell, does not allow me to utilise multiple 
separate computers on a network with access to a common directory, 
whereas multicore does. It also won't work in both Octave and ML which 
is a goal for me. I've also modified multicore to do a few things, like 
run a preprocessing function, and a function between iterations of 
checking for results (which I use to monitor slave resource availability 
for instance on an HTCondor cluster), my version also provides a 
mechanism to kill slaves by deleting certain files in the multicore 
shared directory, and also lets you run a pool of slaves continuously 
and have more than one master process submit jobs to the same pool 
without mixing them up. It also easily lets you see which slave process 
is working on which job as jobs and slaves have a unique ID. Multicore 
is also helpful in that it resubmits jobs that don't complete within a 
pre-specified time limit, or have the master process do them when the 
don't complete within the time limit.

So parcelfun is much too limited for my use case I'm afraid.

> > OK, if this feature has a reason to be I won't remove it from where
> > it is implemented already, but I think we should not insist that new
> > implementations of other ode solvers also implement it.
> >
> > c.

I suppose this is reasonable, since it's not documented.

> The feature is ugly and should be weeded out where possible,
> IMO. Sorry.
>
> Olaf

Beauty is in the eye of the beholder.

Richard

--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.