[gmx-developers] compiling with c++

[David van der Spoel]

On 2010-06-26 23.21, Sander Pronk wrote:
> I tried it a couple of days ago and didn't find anything that couldn't be fixed. The recent power outage in Stockholm has prevented me from fixing things, though.
> Did you find anything particularly worrisome?
>
 > Sander
 >

It seems hard to believe that anyone successfully compiled the whole 
gromacs with c++ recently...

In single precision, warnings

/Users/spoel/GG/gromacs/src/gmxlib/selection/compiler.c: In function 
‘void init_root_item(t_selelem*, 
gmx_ana_index_t*)’:/Users/spoel/GG/gromacs/src/gmxlib/selection/compiler.c:2289: 
warning: comparison between ‘enum e_selelem_t’ and ‘enum e_selvalue_t’

and some more for assignment of char * to constant strings.

Resolution: ignored

Errors:

/Users/spoel/GG/gromacs/src/gmxlib/selection/mempool.c: In function 
‘void _gmx_s
el_mempool_free(gmx_sel_mempool_t*, void*)’:
/Users/spoel/GG/gromacs/src/gmxlib/selection/mempool.c:169: error: 
invalid conversion from ‘void*’ to ‘char*’

/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c: In function ‘void 
qsort_swapfunc(void*, void*, size_t, int)’:
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c:43: error: invalid 
conversion from ‘void*’ to ‘int*’
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c:44: error: invalid 
conversion from ‘void*’ to ‘int*’
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c:54: error: invalid 
conversion from ‘void*’ to ‘char*’
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c:55: error: invalid 
conversion from ‘void*’ to ‘char*’
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c: In function ‘void* 
qsort_med3(void*, void*, void*, int (*)())’:
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c:72: error: too many 
arguments to function
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c:74: error: too many 
arguments to function
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c:76: error: too many 
arguments to function
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c:83: error: too many 
arguments to function
/Users/spoel/GG/gromacs/src/gmxlib/gmx_sort.c:85: error: too many 
arguments to function

etc.

In both mempool.c and gmx_sort.c I've added type casts, however these 
are not trivial, so it would be good if the authors (Teemu?, Berk?) 
check these.

/Users/spoel/GG/gromacs/src/gmxlib/mtxio.c: In function ‘void 
gmx_mtxio_write(const char*, int, int, real*, 
gmx_sparsematrix_t*)’:/Users/spoel/GG/gromacs/src/gmxlib/mtxio.c:135: 
error: cannot convert ‘bool*’ to
  ‘int*’ for argument ‘2’ to ‘bool gmx_fio_doe_int(t_fileio*, int*, 
const char*, const char*, 
int)’/Users/spoel/GG/gromacs/src/gmxlib/mtxio.c: In function ‘void 
gmx_mtxio_read(con
st char*, int*, int*, real**, gmx_sparsematrix_t**)’:
/Users/spoel/GG/gromacs/src/gmxlib/mtxio.c:219: error: cannot convert 
‘bool*’ to ‘int*’ for argument ‘2’ to ‘bool gmx_fio_doe_int(t_fileio*, 
int*, const char*, const char*, int)’

These are more tricky: the C++ type bool is not compatible with int. 
Removing the check for booleans
AC_CHECK_TYPES([bool])
in configure.ac makes matters worse rather than better.

I introduced new functions gmx_fio_doe_bool and  gmx_fio_ndoe_bool that 
converts the bool to an int before writing/reading (in order to keep 
compatible files!).

I replaced the gmx_fio_do_int function by gmx_fio_do_bool also in tpxio.c:
/Users/spoel/GG/gromacs/src/gmxlib/tpxio.c:464: error: cannot convert 
‘bool*’ to ‘int*’ for argument ‘2’ to ‘bool gmx_fio_doe_int(t_fileio*, 
int*, const char*, const char*, int)’
/Users/spoel/GG/gromacs/src/gmxlib/tpxio.c:466: error: cannot convert 
‘bool*’ to ‘int*’ for argument ‘2’ to ‘bool gmx_fio_doe_int(t_fileio*, 
int*, const char*, const char*, int)’
and so on...

/Users/spoel/GG/gromacs/include/smalloc.h: In function ‘void 
_snew_aligned(const char*, const char*, int, T*&, size_t, size_t, 
size_t) [with T = real]’:
/Users/spoel/GG/gromacs/src/mdlib/tables.c:914:   instantiated from here
/Users/spoel/GG/gromacs/include/smalloc.h:195: error: invalid conversion 
from ‘const char*’ to ‘char*’
/Users/spoel/GG/gromacs/include/smalloc.h:195: error:   initializing 
argument 1 of ‘void* save_calloc_aligned(char*, char*, int, unsigned 
int, size_t, size_t)’
/Users/spoel/GG/gromacs/include/smalloc.h:195: error: invalid conversion 
from ‘const char*’ to ‘char*’
/Users/spoel/GG/gromacs/include/smalloc.h:195: error:   initializing 
argument 2 of ‘void* save_calloc_aligned(char*, char*, int, unsigned 
int, size_t, size_t)’

Fixed this by twice adding a typecast to (char *) in the template code 
in smalloc.h line 195.

/Users/spoel/GG/gromacs/src/mdlib/qm_orca.c: In function ‘real 
read_orca_output(
real (*)[3], real (*)[3], int, t_forcerec*, t_QMrec*, 
t_MMrec*)’:/Users/spoel/GG/gromacs/src/mdlib/qm_orca.c:248: warning: 
format ‘%s’ expects type ‘char*’, but argument 3 has type ‘int*’

Replace %s by %d

g_membed wouldn't compile because gmx_ana.h wasn't included leading to 
name mangling problems.

Then there is a clash between X-windows and MPI: both define a typedef 
Status. Duh! Removed g_showcol from compilation list to side-step this 
issue.

Another tip: I usually compile in separate object directories rather 
than in the source tree. It is a good idea to remove the src etc. before 
you run configure. Now I had a src/gmxlib/version.h from february lying 
around there which was incorrect and caused linking errors.


Now in double precision:
Undefined symbols:
   "_nb_kernel430_x86_64_sse2", referenced from:
       kernellist_x86_64_sse2       in 
libgmx_mpi_d.a(nb_kernel_x86_64_sse2.o)
   "_nb_kernel410_x86_64_sse2", referenced from:
       kernellist_x86_64_sse2       in 
libgmx_mpi_d.a(nb_kernel_x86_64_sse2.o)
   "_nb_kernel400_x86_64_sse2", referenced from:
       kernellist_x86_64_sse2       in 
libgmx_mpi_d.a(nb_kernel_x86_64_sse2.o)


This is a name mangling problem and it is fixed by including the 
appropriate header file into the c-files. By the way the comment in this 
file still says:
  * This file is generated automatically at compile time
  * by the program mknb in the Gromacs distribution.

I don't think that is correct anymore, is it?


So please test compiling in single and double precision as well as with 
with C compile and C++ compiler.

I have committed these patches, and now everything compiles.





-- 
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205.
spoel at xray.bmc.uu.se    http://folding.bmc.uu.se