OpenEuphoria

1. Problems with C floats

• Posted by irv Jan 30, 2010
• 2459 views

Are there known bugs when passing Eu values to C_FLOAT ? I have several routines in EuGTK which worked fine with Eu 3, but no longer work properly now with 4.0. Each of these routines has one thing in common, they pass floats to c_func or c_proc.

3. Re: Problems with C floats

• Posted by irv Jan 30, 2010
• 2486 views

For a test, in the demos there's callc.exu - copying the last test, and changing the C_DOUBLE to a C_FLOAT, we have this code:

puts(1, "\n> printf test:\n") 
s = define_c_proc(libc, "printf", {C_POINTER, C_INT, C_DOUBLE, C_INT, C_DOUBLE}) 
format = allocate_string("Here are 4 numbers: %d, %.3f, %d, %e\n") 
c_proc(s, {format, 111, 222.222, 333, 444.4444}) 
 
puts(1, "\n> printf test:\n") 
s = define_c_proc(libc, "printf", {C_POINTER, C_INT, C_DOUBLE, C_INT, C_FLOAT}) 
format = allocate_string("Here are 4 numbers: %d, %.3f, %d, %e\n") 
c_proc(s, {format, 111, 222.222, 333, 444.4444}) 
 

and get this printed:

> printf test: 
Here are 4 numbers: 111, 222.222, 333, 4.444444e+02 
 
> printf test: 
Here are 4 numbers: 111, 222.222, 333, 1.332670e-269 

Now, I may be doing something foolishly wrong, but the results I'm getting from my GTK programs indicate that something isn't getting through properly, and it only happens when C_FLOATs are involved.

5. Re: Problems with C floats

• Posted by mattlewis (admin) Jan 30, 2010
• 2499 views

I don't think so. Can you post some more details? I did this test:

// irv.c 
// build with: $ gcc -shared -fPIC -ldl irv.c -o irv.so 
#include <stdio.h> 
 
void test_float( double foo, float bar ){ 
	printf( "double: %f\nfloat: %f\n", foo, bar ); 
} 
 

...and then called it using:

-- irv.ex 
include std/dll.e 
 
constant 
	irv = open_dll( "./irv.so" ), 
	test_float = define_c_proc( irv, "test_float", {C_DOUBLE, C_FLOAT}) 
 
c_proc( test_float, { 0.25, 0.75}) 

Then I get:

$ eui irv 
double: 0.250000 
float: 0.750000 

So it looks good to me.

Matt

7. Re: Problems with C floats

• Posted by jimcbrown (admin) Jan 30, 2010
• 2503 views

puts(1, "\n> printf test:\n") 
s = define_c_proc(libc, "printf", {C_POINTER, C_INT, C_DOUBLE, C_INT, C_FLOAT}) 
format = allocate_string("Here are 4 numbers: %d, %.3f, %d, %e\n") 
c_proc(s, {format, 111, 222.222, 333, 444.4444}) 

The test used %e format, not %f format.

But OTOH I think all of printf's formats might expect doubles.. as I posted already in this thread.

9. Re: Problems with C floats

• Posted by mattlewis (admin) Jan 31, 2010
• 2417 views

irv@irv-laptop ~ $ eui irv.ex 
double: 0.250000 
float: 0.000000 

Yeah, I understand. printf is just the easiest way to see what's getting passed. I was running r3054, but I get the expected results after updating to 3061, too. I'm using gcc 4.4.1, BTW.

I can't get it to break. I switched order, changed the format to %e, added some other variables to pass...Can anyone else reproduce this?

Matt

11. Re: Problems with C floats

• Posted by irv Jan 31, 2010
• 2420 views

Yes, the routine in question needs the following: C_POINTER, C_FLOAT, C_FLOAT, C_FLOAT, C_INT No varargs involved, and the same routine in GTK works correctly when called by a C program.

13. Re: Problems with C floats

• Posted by irv Jan 31, 2010
• 2407 views

WooHoo! The eui from above (version r3023) works fine! (both with the test program and with GTK) Something changed after that.

15. Re: Problems with C floats

• Posted by mattlewis (admin) Jan 31, 2010
• 2438 views

I don't think it's anything that changed in the source. I tried the eubins version of r3023, and I got the same bad results. I think there's something funky about the build environment for eubins.

Matt

17. Re: Problems with C floats

• Posted by irv Jan 31, 2010
• 2406 views

HOWEVER - while sending floats now works, getting them back from a c routine still fails!

19. Re: Problems with C floats

• Posted by mattlewis (admin) Jan 31, 2010
• 2381 views

I updated my test to:

// irv.c 
// build with $ gcc -shared -fPIC -ldl irv.c -o irv.so 
#include <stdio.h> 
 
void test_float( float foo, double bar, int baz ){ 
	printf( "double: %e\nfloat: %e\nint: %d\n", foo, bar, baz ); 
} 
 
float test_irv( int *ptr, float f1, float f2, float f3, int i ){ 
	printf( "ptr: %p\nf1: %f (%x)\nf2: %f (%x)\nf3: %f (%x)\ni: %d\n", 
			ptr,  
			f1, *(int*)(&f1), 
			f2, *(int*)(&f2), 
			f3, *(int*)(&f3), 
			i ); 
	return f1 + f2; 
} 

...and...

-- irv.ex 
include std/dll.e 
 
constant 
	irv = open_dll( "./irv.so" ), 
	test_float = define_c_proc( irv, "test_float", {C_FLOAT, C_DOUBLE, C_INT}), 
	test_irv   = define_c_func( irv, "test_irv", {C_POINTER, C_FLOAT, C_FLOAT, C_FLOAT, C_INT}, C_FLOAT ) 
 
c_proc( test_float, { 0.25, 0.75, 9}) 
? c_func( test_irv, { 0xDEADBEEF, 1.1, 2.3, 0.5, 8008135} ) 

And I get:

$ eui irv 
double: 2.500000e-01 
float: 7.500000e-01 
int: 9 
ptr: 0xdeadbeef 
f1: 1.100000 (3f8ccccd) 
f2: 2.300000 (40133333) 
f3: 0.500000 (3f000000) 
i: 8008135 
3.399999976 

Or did you mean as a callback? That shouldn't have ever worked. A euphoria callback routine assumes that it's getting back all 32-bit integers, so you'd need to convert the float:

function my_callback( atom the_float ) 
    poke4( buffer, the_float ) 
    the_float = float32_to_atom( peek( { buffer, 4 } ) ) 
    return 0 
end function 

Matt

21. Re: Problems with C floats

• Posted by mattlewis (admin) Jan 31, 2010
• 2376 views

void                gtk_window_get_default_size         (GtkWindow *window, 
                                                         gint *width, 
                                                         gint *height); 

Not sure what you're doing as far as passing floats:

// irv.c 
// build with $ gcc -shared -fPIC -ldl irv.c -o irv.so 
#include <stdio.h> 
 
void pbr_float( float * pf ){ 
	*pf += 5.5f; 
} 

...and then...

-- irv.ex 
include std/dll.e 
include std/machine.e 
include std/convert.e 
 
constant 
	irv = open_dll( "./irv.so" ), 
	pbr_float  = define_c_proc( irv, "pbr_float", {C_POINTER} ) 
 
atom ptr = allocate( 4 ) 
poke( ptr, atom_to_float32( 1.5 ) ) 
 
c_proc( pbr_float, {ptr} ) 
 
? float32_to_atom( peek( ptr & 4 ) ) 

From that, I get:

$ eui irv 
7 

...which is correct.

Matt

23. Re: Problems with C floats

• Posted by jimcbrown (admin) Jan 31, 2010
• 2379 views

void                gtk_window_get_default_size         (GtkWindow *window, 
                                                         gint *width, 
                                                         gint *height); 

You know, it'd help a lot if you posted example code demonstrating the bug...

Anyways, I updated the tests here:

http://malcom.unkmar.com/lostsouls/irv3.c http://malcom.unkmar.com/lostsouls/irv3.ex http://malcom.unkmar.com/lostsouls/irv3.so

And I get the expected results. So this looks like it works fine, or at least I can't reproduce the issue...

25. Re: Problems with C floats

• Posted by irv Jan 31, 2010
• 2369 views

If I knew enough C to write the tests, I wouldn't need Euphoria. I'll just wait until someone else has the same problem, and it gets fixed.

27. Re: Problems with C floats

• Posted by mattlewis (admin) Jan 31, 2010
• 2369 views

The problem is, from what you've said, the bug appears to be in your euphoria code, not euphoria itself. We've come up with test scenarios to try the things you've discovered. Admittedly, there appears to be something wrong with the eubins build, but the problem appears to be there, not with the euphoria code. I'm happy to help figure it out, but I think we need an example that demonstrates the issue.

Matt

29. Re: Problems with C floats

• Posted by irv Feb 01, 2010
• 2293 views

// irv.c  
// build with $ gcc -shared -fPIC -ldl irv3.c -o irv3.so  
#include <stdio.h> 
 
static int a = 12; 
static double b = 34567890; 
static float c = 67.890123; 
 
void test_float(int *foo, double *bar, float *baz){ 
 
	printf( "C in:\n\tfoo: %d\n\tbar: %e\n\tbaz: %f\n\n", *foo,*bar,*baz); 
	printf("C routine changes values & should return \n\t12, \t34567890, \t67.890123\n\n"); 
 
	*foo = a;  
	*bar = b;  
	*baz = c;  
	 
	printf( "C out:\n\tfoo: %d\n\tbar: %e\n\tbaz: %f\n\n", *foo,*bar,*baz); 
}  
 

-- irv3.ex 
include std/dll.e 
include std/memory.e 
include std/convert.e as convert 
 
constant 
        irv = open_dll( "./irv3.so" ), 
        test1 = define_c_proc( irv, "test_float", {C_POINTER,C_POINTER,C_POINTER}), 
        test2 = define_c_proc( irv, "test_float", {C_INT,C_DOUBLE,C_FLOAT}) 
         
atom 
		ptr1 = allocate(4), 
		ptr2 = allocate(8), 
		ptr3 = allocate(8) 
		 
poke(ptr1,convert:int_to_bytes(123)) 
poke(ptr2,convert:atom_to_float64(123456789)) 
poke(ptr3,convert:atom_to_float32(3.14159)) 
 
puts(1,"Eu sends: 123, 123456789, 3.14159 to C routine:\n") 
 
-- change the following to test2, and you get a machine exception: 
c_proc( test1, {ptr1,ptr2,ptr3})  
 
puts(1,"Eu received:\n") 
puts(1,"\tINT:") ? peek(ptr1)  
puts(1,"\tDBL:") ? convert:float64_to_atom(peek({ptr2,8}))  
puts(1,"\tFLT:") ? convert:float32_to_atom(peek({ptr3,4})) 

This is the best I can do without knowing any C. Test 1 successfully passes values both ways. Test 2 crashes every time, no matter what values are sent.

31. Re: Problems with C floats

• Posted by irv Feb 01, 2010
• 2299 views

It would be a lot more help and take less of your time if you would just write the correct way to send and receive values using test2. Don't change the dll, just the Eu calling code.

33. Re: Problems with C floats

• Posted by irv Feb 01, 2010
• 2298 views

void 
gtk_misc_get_alignment (GtkMisc *misc, 
		        gfloat  *xalign, 
			gfloat  *yalign) 
{ 
  g_return_if_fail (GTK_IS_MISC (misc)); 
 
  if (xalign) 
    *xalign = misc->xalign; 
  if (yalign) 
    *yalign = misc->yalign; 
} 

35. Re: Problems with C floats

• Posted by irv Feb 01, 2010
• 2279 views

Thanks, I'll give that a try.

37. Re: Problems with C floats

• Posted by irv Feb 01, 2010
• 2260 views

OK, I think that solved the last problem remaining with GTK. I'll send the update to RDS tomorrow. Thanks for the assistance.

39. Re: Problems with C floats

• Posted by jimcbrown (admin) Mar 20, 2010
• 2092 views

Someone reported issues with EuGTK crashing with c_func()/c_proc(). I think the issue with eubins is still there.

double: 4.602295e-34 
float: 7.500000e-01 
int: 9 
ptr: 0xdeadbeef 
f1: 2.300000 (40133333) 
f2: 0.500000 (3f000000) 
f3: 0.000000 (818efff) 
i: 8008135 
2.799999952 

double: 2.500000e-01 
float: 7.500000e-01 
int: 9 
ptr: 0xdeadbeef 
f1: 1.100000 (3f8ccccd) 
f2: 2.300000 (40133333) 
f3: 0.500000 (3f000000) 
i: 8008135 
3.399999976 

41. Re: Problems with C floats

• Posted by irv Apr 02, 2010
• 1954 views

Since the problem with floats still exists on r3132 from the eubins for Linux, I have compiled a version which does not have the problem, and will try to keep it in sync with the svn.

Get the eubins here if you need it: http://etcwebspace.com/users/irvm