10/11/16
Carlo,
I'm concerned that the latest changes to the ode functions made just five
days ago are too speculative for the stable branch and should be re-located
to the development branch. As Carnë noted in a separate e-mail, feature
freeze has already happened and the stable branch is now just for bug fixes
and documentation updates.
I'm not saying that the changes are bad, but it's clear that there hasn't
been enough time to test things. The first commit broke 'make check' and
that had to be quickly patched. If I look at the code for odedefaults.m I
see that there is no indentation for code within the function block, single
quotes are used instead of double quotes, and there is no space between the
function name and the opening parenthesis. This is small stuff, and can be
corrected, but it is an indication that a thorough review has not
happened. Similarly, ode23.m has a typo in the docstring:
<@qcode{"AbsTol"},.>. The comma before the period is not grammatically
correct. With more testing time I'm sure this would be caught. Finally, I
did some benchmarking of odeset because I remember that I had to do some
tricks to get it to run quickly. The following code
tic; x = odeset (); toc
runs in 0.00015 seconds with the old code, but now takes .0058 seconds.
That's a 39X slowdown. Of course the absolute time of five milliseconds is
probably unimportant to a human, but I want to understand whether we must
accept that slowdown, or whether there is a better way to code this.
Perhaps the use of persistent variables in odedefaults would help.
If you have a plan on how we could become comfortable with the idea of
keeping this code on the stable branch we will listen. Otherwise, it seems
like the best course is to move the changes to the default branch where
they can simply have more time to mature into solid code.
--Rik