Loïc Bartoletti
README.TYPE
.\" use -ms
.LP
There were two flavors of changes made to structures in gis.h
and to calls to routines in the library:
.IP 1.
Variables in structures, parameters in argument lists to library routines,
and return values from library routines that really represented
CELL data but were declared as int have been changed to CELL.
The following structures were changed:
.DS
Categories [gis.h]
Cell_stats
Colors
Histogram
Range
Reclass
.DE
The argument type for some parameters for the following routines was changed:
.DS
G_init_cats [cats.c]
G_get_cats
G_set_cats
G_get_color [color.c]
G_set_color
G_make_random_colors [make_colr2.c]
G_make_color_wave
G_make_grey_scale
G_make_color_ramp
G_make_aspect_colors
G_add_histogram [histogram.c]
G_set_histogram
G_update_range [range.c]
.DE
The return value for the following routines was changed:
.DS
G_number_of_cats [cats.c]
G_get_histogram_cat [histogram.c]
.DE
This change does not, at present, require changes to existing code since
the CELL type is currently defined as int.
In the future, if we (or someone else) want to
redefine CELL to long (say on a 16 bit machine), the application code
will compile ok, but will not run correctly since the calling seqences
(including printf and scanf) will not match anymore.
To upgrade GRASS code for this is a big deal.
It consists of
finding all places that ints are used in assignments with
CELL variables,
modifying all the calls to the above routines,
and modifying all calls to other routines that don't know about the CELL
type (in particular printf, scanf, and Vask code).
The way to handle printf for CELL data is to use %ld and cast the
CELL value to long:
.DS
CELL x
printf ("%ld", (long) x);
.DE
The way to handle scanf for CELL data is to use %ld into a long
and assign to the CELL variable:
.DS
CELL x;
long t;
scanf ("%ld", &t);
x = (CELL) t;
.DE
For Vask calls, assign the CELL value to a long, call Vask will the long,
then assign the long to the CELL variable:
.DS
CELL x;
long t;
t = x;
V_clear();
V_ques (&t, 'l', ...);
V_call();
x = t;
.DE
The long vs int issue for CELL also presents a subtle, but
potentially fatal, problem with malloc(n) which requires n to be int.
Suppose we want to allocate an array for reclass, from
cats min to max. What we do now is
.DS
char *table;
CELL min,max;
int len;
len = max - min + 1;
table = malloc (len);
.DE
This will fail if CELL is long and the expression max-min+1 yields
a value which overflows an int. Therefore, we should do
.DS
long len;
len = max - min + 1;
if (len != (int)len)
error("Too many categories");
table = malloc ((int)len);
.DE
.IP 2.
Variables in structures, parameters in argument lists to library routines,
and return values from library routines that should have been long
but were declared as int have been changed to long.
The following structures were changed:
.DS
Cell_stats [gis.h]
Histogram
.DE
The argument type for some parameters for the following routines was changed:
.DS
G_add_histogram [histogram.c]
G_set_histogram
G_find_cell_stat [cell_stats.c]
G_next_cell_stat
.DE
The return value for the following routines was changed:
.DS
G_get_histogram_count [histogram.c]
.DE
These changes required immediate upgrade.
I think this update is done, but I want to review it again.
.LP
There were two flavors of changes made to structures in gis.h
and to calls to routines in the library:
.IP 1.
Variables in structures, parameters in argument lists to library routines,
and return values from library routines that really represented
CELL data but were declared as int have been changed to CELL.
The following structures were changed:
.DS
Categories [gis.h]
Cell_stats
Colors
Histogram
Range
Reclass
.DE
The argument type for some parameters for the following routines was changed:
.DS
G_init_cats [cats.c]
G_get_cats
G_set_cats
G_get_color [color.c]
G_set_color
G_make_random_colors [make_colr2.c]
G_make_color_wave
G_make_grey_scale
G_make_color_ramp
G_make_aspect_colors
G_add_histogram [histogram.c]
G_set_histogram
G_update_range [range.c]
.DE
The return value for the following routines was changed:
.DS
G_number_of_cats [cats.c]
G_get_histogram_cat [histogram.c]
.DE
This change does not, at present, require changes to existing code since
the CELL type is currently defined as int.
In the future, if we (or someone else) want to
redefine CELL to long (say on a 16 bit machine), the application code
will compile ok, but will not run correctly since the calling seqences
(including printf and scanf) will not match anymore.
To upgrade GRASS code for this is a big deal.
It consists of
finding all places that ints are used in assignments with
CELL variables,
modifying all the calls to the above routines,
and modifying all calls to other routines that don't know about the CELL
type (in particular printf, scanf, and Vask code).
The way to handle printf for CELL data is to use %ld and cast the
CELL value to long:
.DS
CELL x
printf ("%ld", (long) x);
.DE
The way to handle scanf for CELL data is to use %ld into a long
and assign to the CELL variable:
.DS
CELL x;
long t;
scanf ("%ld", &t);
x = (CELL) t;
.DE
For Vask calls, assign the CELL value to a long, call Vask will the long,
then assign the long to the CELL variable:
.DS
CELL x;
long t;
t = x;
V_clear();
V_ques (&t, 'l', ...);
V_call();
x = t;
.DE
The long vs int issue for CELL also presents a subtle, but
potentially fatal, problem with malloc(n) which requires n to be int.
Suppose we want to allocate an array for reclass, from
cats min to max. What we do now is
.DS
char *table;
CELL min,max;
int len;
len = max - min + 1;
table = malloc (len);
.DE
This will fail if CELL is long and the expression max-min+1 yields
a value which overflows an int. Therefore, we should do
.DS
long len;
len = max - min + 1;
if (len != (int)len)
error("Too many categories");
table = malloc ((int)len);
.DE
.IP 2.
Variables in structures, parameters in argument lists to library routines,
and return values from library routines that should have been long
but were declared as int have been changed to long.
The following structures were changed:
.DS
Cell_stats [gis.h]
Histogram
.DE
The argument type for some parameters for the following routines was changed:
.DS
G_add_histogram [histogram.c]
G_set_histogram
G_find_cell_stat [cell_stats.c]
G_next_cell_stat
.DE
The return value for the following routines was changed:
.DS
G_get_histogram_count [histogram.c]
.DE
These changes required immediate upgrade.
I think this update is done, but I want to review it again.