Cleanup several things. #17
Conversation
We just use GitHub issues now.
Most of this comes from NR and should be replaced anyway.
The underscore situation should be detected by running ./configure.
These are already defined in math.h.
For system headers, they basically do nothing but take up extra lines.
Since the exectuables were never mentioned by full path, they would be rebuilt unconditionally.
Without the correct dependencies, parallel builds fail due to missing modules.
It's only used in the mesher, and the search path was wrong when used here anyway.
Also, move them away from the NR functions.
Just placing the word exit there does nothing.
* MAX can be replaced by fmax. * SIGN can be replaced by copysign. * PYTHAG can be replaced by hypot. The last two require C99, but since we require Fortran2003, I don't think that's a big deal.
No more of this pseudo-Fortran indexing. It's odd and causes unnecessary work (to be fixed).
ctype.h is required for isspace.
The status is not used, so this saves us an extra variable or two. As suggested by @jedbrown in another PR.
Every time you configured with internal scotch, you'd get modifications to its Makefiles which git would show as modified. This is troublesome whenever you want to commit or change branches, because you'd have to remember not to keep those changes. The changes to the scotch Makefiles are independent of any options you might pass to configure (except that they aren't done if you aren't using internal scotch). We can just make them permanently in git, especially because the changes have been made upstream in scotch as well.
It's a bit confusing to call it 'makefile' when it's just a shell script and not intended to be used with Make at all.
This should be more portable, but most scripts are really just a list of commands, making little difference. Tested with bash and dash (which is POSIX compliant, but doesn't implement most of bash's extensions).
I wonder if we should just remove this commented code, though.
|
Gosh darn it, this is opened to the wrong branch again! If you're merging, please change it to the QA branch first, since it doesn't appear I can change it myself. |
Without this dependency, parallel builds fail due to missing scotchf.h.
The Pyrized version of the code was removed long ago, so these checks don't do anything except change the default targets.
|
We should try to make QA the default otherwise I am sure many users will On 01/12/2014 02:57 AM, Elliott Sales de Andrade wrote:
Dimitri Komatitsch |
|
It is possible to make QA the default for when people clone the repo. See
My preference goes to 2 (developers make the change on their forks). But I Best, On Sun, Jan 12, 2014 at 3:58 PM, Dimitri Komatitsch <
David Luet |
|
What I said above is inconsistent with the workflow we established. The On Mon, Jan 13, 2014 at 10:53 AM, David Luet luet@princeton.edu wrote:
David Luet |
attenuation_compute_param.cas mentioned in remove copyrighted routines from attenuation_compute_param.c #13, though some of the NR code still needs replacing.In related terms, I'm wondering if I can:
DATA/Database00000_left_edge_only,DATA/Database00000_right_edge_only? What are these even used for? I see they're read in the solver, but I'm not sure how they're produced.UTILS/visualization/plot_wavefield.pl? @carltape?