On Mon, 25 Nov 2013 08:46:46 -0500, Brian Wignall wrote:
Hello, everyone.
I am new and hoping to make my first contribution. I read through the
Wish
List, and I thought I could help quickly by starting with the special
functions request.
Thanks for your interest in Commons Math.
Contributions are very welcome, but contributors even more so.
By this, I mean that the most important thing is not to increase the
codebase, but to make it so that it is gracefully integrated in the
project.
I did not read your patch in detail, but it is enough to already detect
the drawbacks of automated translation from Fortran.
[We have an unfortunate example of this in the class
"BOBYQAOptimizer".]
A few remarks (in no particular order):
* Existing functionality must be reused (e.g. "factorial").
* Public instance variables are forbidden.
* Variable name must not contain underscores (unless it is a global
constant).
* Variables must be defined at the smallest possible scope and must
have
names as meaningful as possible (unlike customary in Fortran!).
* Commons Math's exception must be used exclusively.
* Global constants must be "final" (and their name must be in all
uppercase).
* Operators must be surrounded by space characters.
* ...
IMO, a very important thing is to make the code as readable as
possible.
I know that you had to work already to obtain this result from the
output
of the translator, but some further improvements are needed so that the
contribution looks like a Java source coded from scratch.
Best regards,
Gilles
I created a JIRA (https://issues.apache.org/jira/browse/MATH-1066)
that
contains a patch for Bessel functions of the first kind. There are
other
Bessel functions that are relevant, but I figured I would post that,
to get
(minor, hopefully) suggestions or tweaks to get it ready, to
streamline the
process for other functions.
The JIRA is a bit sparse, so some more details on the patch follow.
Like several other open source packages, I started from Fortran code
in
Netlib; in this case, http://www.netlib.org/specfun/rjbesl . As a
first
pass, I enabled the double-precision lines with IEEE values already
in the
code and then put the Fortran code through f2j (
http://www.netlib.org/java/f2j/). This handled the straightforward
parts,
but it left in gotos and some other bits that needed cleaning.
Getting the
output of f2j to a decent place took more time and effort than I had
anticipated, but it is passing the unit tests I put in (based on
output
from R, which uses a translation of the same Fortran code).
I am looking forward to hearing thoughts and suggestions.
Kind regards,
Brian
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
