libvips/cplusplus/examples/test_overloads.cpp
John Cupitt ded01df512 c++: add [], change ()
() used to return element zero, just return a vector

add [] overload
2015-03-01 10:43:02 +00:00

283 lines
6.2 KiB
C++

/* Test +/-* overloads with every combination of vector, image and double.
* This isn't a full test suite, look in the Python area for that.
*
* compile with:
*
* g++ -g -Wall test_overloads.cpp `pkg-config vips-cpp --cflags --libs`
*
* run with:
*
* valgrind --leak-check=yes ./a.out ~/pics/k2.jpg ~/pics/shark.jpg
*
*/
/*
#define VIPS_DEBUG
#define VIPS_DEBUG_VERBOSE
*/
#include <stdlib.h>
#include <vips/vips8>
using namespace vips;
bool
equal_vector( std::vector<double> a, std::vector<double> b )
{
for( unsigned int i = 0; i < a.size(); i++ )
if( fabs( a[i] - b[i] ) > 0.001 ) {
printf( "vectors differ at %d: should be [", i );
for( unsigned int i = 0; i < a.size(); i++ ) {
if( i > 0 )
printf( ", " );
printf( "%g", a[i] );
}
printf( "], is [" );
for( unsigned int i = 0; i < a.size(); i++ ) {
if( i > 0 )
printf( ", " );
printf( "%g", a[i] );
}
printf( "]\n" );
return( false );
}
return( true );
}
bool
equal_double( double a, double b )
{
if( fabs( a - b ) > 0.001 ) {
printf( "doubles differ: should be %g, is %g\n", a, b );
return( false );
}
return( true );
}
/* We can't do this with a template, I think we'd need partially-parameterised
* template, which is C++11 only.
*/
/* Only test a few points and only test uchar: we are just testing the C++
* overloads, we rely on the python test suite for testing the underlying
* vips operators.
*/
#define TEST_BINARY( OPERATOR ) \
void \
test_binary_##OPERATOR( VImage left, VImage right ) \
{ \
for( int x = 10; x < 30; x += 10 ) { \
std::vector<double> p_left = left.getpoint( x, x ); \
std::vector<double> p_right = right.getpoint( x, x ); \
std::vector<double> p_result = \
OPERATOR<std::vector<double>, \
std::vector<double>, \
std::vector<double> >(p_left, p_right ); \
\
VImage im_result; \
std::vector<double> p_im_result; \
\
/* test: image = image OP image \
*/ \
im_result = OPERATOR<VImage, VImage, VImage>( left, right ); \
p_im_result = im_result.getpoint( x, x ); \
\
if( !equal_vector( p_result, p_im_result ) ) { \
printf( #OPERATOR \
"(VImage, VImage) failed at (%d, %d)\n", \
x, x ); \
abort(); \
} \
\
/* test: image = image OP vec \
*/ \
im_result = \
OPERATOR<VImage, \
VImage, std::vector<double> >( left, p_right );\
p_im_result = im_result.getpoint( x, x ); \
\
if( !equal_vector( p_result, p_im_result ) ) { \
printf( #OPERATOR \
"(VImage, vector) failed at (%d, %d)\n", \
x, x ); \
abort(); \
} \
\
/* test: image = vec OP image \
*/ \
im_result = \
OPERATOR<VImage, std::vector<double>, \
VImage>( p_left, right ); \
p_im_result = im_result.getpoint( x, x ); \
\
if( !equal_vector( p_result, p_im_result ) ) { \
printf( #OPERATOR \
"(vector, VImage) failed at (%d, %d)\n", \
x, x ); \
abort(); \
} \
\
/* test: image = image OP double \
*/ \
for( unsigned int i = 0; i < p_right.size(); i++ ) { \
im_result = \
OPERATOR<VImage, \
VImage, double>( left, p_right[i] ); \
p_im_result = im_result.getpoint( x, x ); \
\
if( !equal_double( p_result[i], p_im_result[i] ) ) { \
printf( #OPERATOR \
"(VImage, double) failed at " \
"(%d, %d)\n", \
x, x ); \
abort(); \
} \
} \
\
/* test: image = double OP image \
*/ \
for( unsigned int i = 0; i < p_left.size(); i++ ) { \
im_result = \
OPERATOR<VImage, \
double, VImage>( p_left[i], right ); \
p_im_result = im_result.getpoint( x, x ); \
\
if( !equal_double( p_result[i], p_im_result[i] ) ) { \
printf( #OPERATOR \
"(double, VImage) failed at " \
"(%d, %d)\n", \
x, x ); \
abort(); \
} \
} \
} \
}
// eg. double = double + double
// or image = double + image
template <typename A, typename B, typename C>
A test_add( B left, C right )
{
return( left + right );
}
template <typename T>
std::vector<T> operator+(std::vector<T> &v1, const std::vector<T> &v2)
{
std::vector<T> result( v1.size() );
for( unsigned int i = 0; i < v1.size(); i++ )
result[i] = v1[i] + v2[i];
return( result );
}
TEST_BINARY( test_add );
template <typename A, typename B, typename C>
A test_subtract( B left, C right )
{
return( left - right );
}
template <typename T>
std::vector<T> operator-(std::vector<T> &v1, const std::vector<T> &v2)
{
std::vector<T> result( v1.size() );
for( unsigned int i = 0; i < v1.size(); i++ )
result[i] = v1[i] - v2[i];
return( result );
}
TEST_BINARY( test_subtract );
template <typename A, typename B, typename C>
A test_multiply( B left, C right )
{
return( left * right );
}
template <typename T>
std::vector<T> operator*(std::vector<T> &v1, const std::vector<T> &v2)
{
std::vector<T> result( v1.size() );
for( unsigned int i = 0; i < v1.size(); i++ )
result[i] = v1[i] * v2[i];
return( result );
}
TEST_BINARY( test_multiply );
template <typename A, typename B, typename C>
A test_divide( B left, C right )
{
return( left / right );
}
template <typename T>
std::vector<T> operator/(std::vector<T> &v1, const std::vector<T> &v2)
{
std::vector<T> result( v1.size() );
for( unsigned int i = 0; i < v1.size(); i++ )
result[i] = v1[i] / v2[i];
return( result );
}
TEST_BINARY( test_divide );
int
main( int argc, char **argv )
{
GOptionContext *context;
GOptionGroup *main_group;
GError *error = NULL;
if( VIPS_INIT( argv[0] ) )
vips_error_exit( NULL );
context = g_option_context_new( "" );
main_group = g_option_group_new( NULL, NULL, NULL, NULL, NULL );
g_option_context_set_main_group( context, main_group );
g_option_context_add_group( context, vips_get_option_group() );
if( !g_option_context_parse( context, &argc, &argv, &error ) ) {
if( error ) {
fprintf( stderr, "%s\n", error->message );
g_error_free( error );
}
vips_error_exit( NULL );
}
g_option_context_free( context );
{
VImage left = VImage::new_from_file( argv[1] );
VImage right = VImage::new_from_file( argv[2] );
test_binary_test_add( left, right );
test_binary_test_subtract( left, right );
test_binary_test_multiply( left, right );
test_binary_test_divide( left, right );
VImage band_one = left[1];
std::vector<double> point = left(0, 0);
}
vips_shutdown();
return( 0 );
}