| 1 |
Hi all, |
| 2 |
|
| 3 |
please review and comment the attached eclass. |
| 4 |
|
| 5 |
Purpose of this eclass is the correct setting of a working fortran |
| 6 |
compiler. There are numerous bugs which suffer from one or the other |
| 7 |
defect here. Anybody who touch a fortran package knows what I am talking |
| 8 |
about. |
| 9 |
Currently we support two fortran compilers in the tree, soonish there |
| 10 |
will be three. But we also like to like to allowed any out-of-tree |
| 11 |
compiler. So depending on gcc[fortran] or virtual/fortran doesn't |
| 12 |
fullfill the needs for one or the other reason, which I will not |
| 13 |
elaborate again. |
| 14 |
|
| 15 |
Our solution: |
| 16 |
1. Depend on virtual/fortran. This will force the ordinary user to use |
| 17 |
gfortran through gcc[fortran]. Or the intel compiler has to be selected |
| 18 |
via FC=ifort. With this also any other solution can be selected. |
| 19 |
|
| 20 |
2. Test whether FC is a working fortran compiler. Why? gcc:4.5[fortran] |
| 21 |
and gcc:4.6[-fortran] can be emerged and gcc-4.6 selected. Thereby |
| 22 |
virtual/fortran dependdencies are fullfiled but no working compiler is |
| 23 |
there. Same happens in many other constellations. |
| 24 |
|
| 25 |
3. Test for openmp support. For a mixture of the above reasons, it is |
| 26 |
impossible to depend on openmp capabilities if user do change anything |
| 27 |
from default. |
| 28 |
|
| 29 |
4. Get_fcomp is needed for some packages which do not work with the full |
| 30 |
name, e.g. seperate makefiles for intel and gnu compiler. |
| 31 |
|
| 32 |
5. Once FC is working, set all other variable possibly defining fortran |
| 33 |
compilers of any flavour to FC. |
| 34 |
|
| 35 |
6. It is still possible without any change to ebuilds to integrate the |
| 36 |
test functions in the toolchain-funcs eclass later, if we decide this is |
| 37 |
a better way to handle those functions. |
| 38 |
|
| 39 |
Thanks for attention, justin |
Archives