Bugzilla Bug 170989

I attach an ebuild for Meep.

Meep is a Maxwell's Equations simulator.  It is research-grade fast and
accurate, easy to use, and supports distributed-memory supercomputers.  It also
produces pretty pictures; check out the website (I am not affiliated with the
developers).

I have tested this ebuild, and it works perfectly.  It also passes all of its
own self-tests when run with FEATURES="test".

Possible issues (all very minor):
1. The ebuild has use-flags for guile and hdf5.  Without these dependencies the
program is essentially only usable as a library.  Unfortunately, there is no
way to make them "on by default".

2. The ebuild has use-flags for gsl and fftw.  If these programs are installed
but the flags are not set, they will not be listed as dependencies, but they
will still be detected by the ./configure and linked in.  If they are later
removed from the system, there will be a linking inconsistency requiring a
revdep-rebuild.
2a. There may be a bug causing the program to link in LAPACK and BLAS if they
are available, even though they are not used.

3. The website claims to support side-by-side installation of MPI and non-MPI
versions of the program.  This feature is not made available in this ebuild.

I should mention that this is a sort of sister program to sci-physics/mpb,
which is in the tree.

(In reply to comment #0)
> I attach an ebuild for Meep.

Did you forget something?

Sébastien

Created an attachment (id=113490) [edit]
ebuild for Meep

I guess so.  I could've sworn...

Hi Ben,

Thanks for your ebuild. I gave a quick look. Here are a few things you could
improve:
- fftw: it is a slotted package. which version >=3 or =2.1* ?
- gsl: is it really an option (fftw as well?) to make it a use flag?
- give honour to the doc and examples by installing at least the few ascii
files, more with a "doc" and/or an "examples" use flags.
- syntax: year in header, missing a space in DEPENDS
- libctl/guile: according to the installation web page, they are strongly
recommended dep. May be you could just drop the USE flag, and force the dep on
libctl.

It also looks like it could greatly benefit from the harminv package from the
same guys, but it is not in the tree.

Sébastien

I have been trying to improve this ebuild (but I also need to spend most of my
time actually using the program for a class project).  I can answer some of
your questions:

fftw and gsl are actually soft dependencies.  fftw adds extracting the spectrum
of the time-series of a field-component at a single location, and gsl adds the
ability to use a bessel function in your initial conditions.  I know that I
have no use for either function.

libctl+guile is more interesting, I think.  You're right that they're strongly
recommended, but definitely optional.  My instinct is to have a "no_guile"
useflag, but I'm told that this is considered harmful.

I'm trying to learn how to implement "doc" and "examples" flags.

Created an attachment (id=118514) [edit]
new ebuild for meep-0.10

This new ebuild includes baseline docs, with useflags for additional (somewhat
obsolete) docs and examples.  It also fixes syntax details.

Behavior of libctl useflag is not yet changed.

Created an attachment (id=118515) [edit]
ebuild for sci-mathematics/harminv-1.3.1

Harminv is a program that is potentially useful for analyzing the simulation
output of Meep.  If Harminv is present, Meep provides Guile functions to access
Harminv's algorithm through Meep's interface.

It is not yet clear to me whether Harminv must be installed before Meep to be
accessible through Meep.

Hi Ben,

I took a deeper look at your ebuilds, and put in the gentooscience overlay. I
had to put a newer version of libctl to make them work on my amd64 box. Test
them as much as you can. I changed a few things, so please give it a look as
well.  
The easiest way to get the gentooscience overlay is to use layman -a science

Sébastien

Vote of CC yourself if you would like to see it in the tree.

meep-0.10.1 now in cvs.