*switching back to English*

Both are correct. The documentation for gmp_gcdext does not define the "correct" answer uniquely. It computes g, s and t such that

g = gcd(a,b) = s a + t b

and |s| <= |b| and |t| <= |a|

If you, for example, want s to never be nonegative, you can do something like

if (s < 0) { s += b; t -= a }

(and then there are some additional borderline cases when you can have s = b or s = 0).

Then the testcase should check that these four conditions hold, rather than comparing to a fixed result.

Done.

OTOH we are not really interested in testing the GMP library.

To test that we can communicate with the Gmp lib. For most functions that is done by comparing with a predefined value, but if that can't be done, just testing that the returned value is reasonable is enough. E.g. we don't test that the returned value of random() is actually random.

I just found a bug in the GTK2.Container:

object a=GTK2.Window(GTK2.WindowToplevel); object b=GTK2.Button(); a->add(b); object c=GTK2.Label("text"); b->add(c);

c->add(b);

(Pike GTK:31904): Gtk-CRITICAL **: gtk_container_add: assertion `GTK_IS_WIDGET (widget)' failed

But, GTK2.Label is a GTK2.Widget. (it inherits form GTK2.Misc, which inherits from GTK2.Widget).

I may have to customize the add() method to do some checks and allow the add, rather than autogenerating the code.

Hm, how does this differ from the generated code? Maybe the code generation is wrong?

This is the generated code:

void pgtk2_container_add( INT32 args ) #line 25 "/home/riffraff/src/Pike-v7.8.306/src/post_modules/GTK2/source/gtkcontainer.pre" { GtkWidget *a0; if( args < 1 ) Pike_error(/* "Too few arguments, %d required, got %d\n" */ (PSTR+0x2e), 1, args);

if( Pike_sp[ 0 - args ].type != PIKE_T_OBJECT ) a0 = NULL; else a0 = GTK_WIDGET(get_pg2object(Pike_sp[0-args].u.object, pgtk2_widget_program)); pgtk2_verify_inited(); gtk_container_add( GTK_CONTAINER(THIS->obj), GTK_WIDGET(a0) ); RETURN_THIS(); }

The only real difference I see is the call to get_pg2object, pgtk2_widget_program above vice pg2_object_program for my fix (because get_gobject() gets an object as a pg2_object_program).

I could rewrite my fix to grab pgtk2_widget_program instead and see what happens.

----- Original Message ---- From: Lance Dillon riffraff169@yahoo.com To: pike-devel@lists.lysator.liu.se Sent: Thursday, June 25, 2009 9:16:32 PM Subject: Re: GTK2.Container

That is a good question. I'm doing a test now. I knew what the problem was, and my solution fixed it, but I didn't look at the generated code.

----- Original Message ---- From: "Marcus Comstedt (ACROSS) (Hail Ilpalazzo!) @ Pike (-) developers forum" 10353@lyskom.lysator.liu.se To: pike-devel@lists.lysator.liu.se Sent: Thursday, June 25, 2009 7:00:02 PM Subject: Re: GTK2.Container

Hm, how does this differ from the generated code? Maybe the code generation is wrong?

It is very probable that the answering entity was a GMP -- but knock yourself out, if you want to be really sure it's a non-buggy GMP too.

But what I want to know is why it is more probable that the entity is GMP if the answer if ({32,1,1}) than if the answer is ({1,2,3}), i.e. why should we treat one incorrect answer as an error, but not another one? If all we're interrested in is that we get an answer, and don't care if it is correct or not, we should accept any answer (within the range of the function).

When you have g, s and t, which are claimed to satisfy

g = gcd(a, b) = s * a + t * b

then a sufficient test (besides the range checks) would be that

g == s*a + t * b a % g == 0 b % g == 0

Then it follows that g = gcd(a, b);