On Thu, Sep 01, 2005 at 01:21:24PM +0100, Glynn Clements wrote:
Markus Neteler wrote:
> > Forget about r.cats (and categories in general). This:
> >
> > > r.stats -1 one
> > > 120
> > > 156
> > > 99
> > > 100
> > > 120
> > > 156
> > > 99
> > > 100
> >
> > is a lot more interesting (and relevant).
>
> OK:
>
> in r.stats:
>
> main.c
> G_get_range_min_max (&range, &min, &max);
> min =1
> max =1
>
> ok so far.
> raw_stats (fd, verbose, with_coordinates, with_xy, with_labels);
> fd 0 0 0 0
>
> Entering raw_stats():
> raw_stats.c:49
> rastp[i] = rast[i];
> i=0
> rastp[0]= void
>
> raw_stats.c:98, when i is 0, then
> *((CELL *) rastp[i]) is 120.
>
> Then it prints 120 and so forth.
>
> Should I add some G_debug() in raw_stats?
I'd start by examining the pointers rastp[0] and rast[0], and the
contents of the buffer to which they point, before entering the
"for (col = ..." loop. I.e.:
print rast[0]
print rastp[0]
print *(CELL*)(rast[0])@50
The @ is dislike (not sure what it does). This is what I get:
GRASS 6.1.cvs (spearfish60):~/grass61/raster/r.stats > cvs diff raw_stats.c
Index: raw_stats.c
RCS file: /grassrepository/grass6/raster/r.stats/raw_stats.c,v
retrieving revision 2.1
diff -u -r2.1 raw_stats.c
--- raw_stats.c 13 Jul 2005 15:55:40 -0000 2.1
+++ raw_stats.c 1 Sep 2005 13:10:08 -0000
@@ -52,6 +52,11 @@
if (with_coordinates)
G_format_northing (G_row_to_northing(row+.5, &window), nbuf, -1);
+
+ G_debug(0, "rast[%d]: %d", rast[i], i);
+ G_debug(0, "rastp[%d]: %d", rastp[i], i);
+ G_debug(0, "*(CELL*)(rast[0]): %d", *(CELL*)(rast[0]));
+ G_debug(0, "*(CELL*)(rast[%d]): %d", *(CELL*)(rast[i]));
for (col = 0; col < ncols; col++)
{
if (no_nulls || no_nulls_all)
GRASS 6.1.cvs (spearfish60):~/grass61/raster/r.stats > r.stats -1 one
D0/0: rast[0]: 1
D0/0: rastp[0]: 1
D0/0: *(CELL*)(rast[0]): 120
Segmentation fault
Apparently the last G_debug does not make sense, but the mysterious (?) 120
is there.
BTW:
gcc version 3.4.3 20050227 (Red Hat 3.4.3-22.1)
I just see that the damn optimization creeped in again.
When I do
CFLAGS="-g -Wall" ./configure ...
checking host system type... i686-pc-linux-gnu
checking for gcc... gcc
checking whether the C compiler (gcc -g -Wall ) works... yes
checking whether the C compiler (gcc -g -Wall ) is a cross-compiler... no
...
C compiler: gcc -g -Wall
C++ compiler: c++ -g -O2
...
But:
grep CFLAGS_OPTIMIZE ~/grass61/include/Make/Platform.make
CFLAGS_OPTIMIZE = -O2
I have to manually fix that. Seems to be evil in aclocal.m4
around line 756:
Linux*)
SHLIB_CFLAGS="-fPIC"
SHLIB_LD_LIBS='${LIBS}'
SHLIB_SUFFIX=".so"
CFLAGS_OPTIMIZE=-O2
# egcs-2.91.66 on Redhat Linux 6.0 generates lots of warnings
# when you inline the string and math operations. Turn this off to
# get rid of the warnings.
#CFLAGS_OPTIMIZE="${CFLAGS_OPTIMIZE} -D__NO_STRING_INLINES -D__NO_MATH_INLINES"
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
The r.stats debug result remains however the same.
Markus
