New iso_c_binding interface about opencalphad HOT 6 CLOSED

Dear Teslos

I have downloaded your modifications and as it is a long time since I used
linux I have some difficulties testing your modification on Windows. I
will continue working on it, hopefully I can sort it out as it looks very
nice.

My email is [email protected] if you want to contact me directly.

Best wishes

Bosse

2015-01-11 14:21 GMT+01:00 ∃teslos [email protected]:

Dear Prof. Sundman,

I have finished the iso_c_binding for the new TQ interface. The new zip
file you can download from my git repo
https://github.com/Teslos/opencalphad. The main iso binding interface is
in liboctqisoc.F90.
I am also new to git versioning system, so I hope that I properly merged
your version with my modified version.
There still some open issuses with the with deprecated functions for
example tqgcom, which currently don't work in the this version of TQ
interface. But I suppose you will update these functions or removed them in
a new version.
The module variables which are now global to liboctq interface are
exported in iso_c_binding as global variables in C program. I think this is
not a good design decision, but iso_c_binding standard does not give any
other option to do this (at least I don't know of any). Currently, all
global variables and function are prefixed with c_ for example c_ntup. This
reduces the probability that it will clash with names of other variables in
C program but it doesn't eliminate it. Thank you very much for the
opencalphad project!!

Best regards,
∃t

—
Reply to this email directly or view it on GitHub
#11.

from opencalphad.

Dear Prof. Sundman,

The first test version of Python interface to TQ using cython is finished. You can test it by building the library with python setup.py build_ext --inplace. Also, liboceq.a has to be in local directory. This creates pyoctq.so that can be imported in python script. For building I have used Cython version 0.21.2 and Python 2.7.9 on mac os x. Running python test_pyoctq.py gives this output:

System with 2 elements: ['FE', 'NI']
and 2 phases: 
['LIQUID', 'FCC_A1']
 Mode :            1
 Composition set(s) created:            1
 Grid minimization:    16 gridpoints   4.6200E-04 s and       0 clockcycles
Phase change: its/add/remove:     7    1    0
Phase change: its/add/remove:    11    0    2
Equilibrium calculation   16, its,   3.3430E-03 s and       3 clockcycles
Amount of 2 phases:
0.000000000
1.000000000
Phase : LIQUID 
Stable phase : LIQUID, amount: 1.000000
 Phase :            1           1
 FE : 0.243154, 
 NI : 0.756846, 
Phase : FCC_A1 
Stable phase : FCC_A1, amount: 0.300000
 Phase :            2           1
 FE : 0.500000, 
 NI : 0.500000, 
Component, mole fraction and chemical potential
FE        0.500000        -48015.506985
NI        0.500000        -54794.509223

I am still working on the problem of one and zero array offset, as defaults in F90 and python are different..So I think it would be wise to set the index of the phases in python from 0--n and not 1--n as in the fortran code.

Best regards,
∃t

from opencalphad.

@Teslos For this case, have you considered using f2py?

from opencalphad.

Hi Richard,
Yes I have considered using f2py, but it seems that it is not actively developed. The Cython seems to be good alternative because after doing C interface with iso_c_binding you can get Cython support almost for free. So you get bindings for C and Python...
Do you think that it is better to use f2py? There is also fwrap library that looks good...

Best regards,
∃t

from opencalphad.

I do not have very much experience with any of them. I only ask about f2py because it's what the developers of numpy use and it's a very well-known and high-quality Python library.

from opencalphad.

Yes you are right, but I think fwrap was developed to replace f2py, but it is also not updated for several years...

Best regards,
∃t

from opencalphad.