www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.bugs - problem with delegates and arrays?

reply Sean Kelly <sean f4.ca> writes:
This code:

void fn( out char[] buf )
{
void fn2()
{
printf( "%i, %i: %.*s\n", buf.length, buf.size, buf );
}
fn2();
printf( "%i, %i: %.*s\n", buf.length, buf.size, buf );
}
int main()
{
char[] buf;
printf( "%i, %i: %.*s\n", buf.length, buf.size, buf );
fn( buf );
return 0;
}

prints:

0, 8:

0, 8:

So as far as I can tell, the stack pointer is messed up for dynamic arrays
within delegates.
Jun 17 2004
parent reply Charlie <Charlie_member pathlink.com> writes:
I dont see any delgates in there ?  I see the nested function ..

This problem goes away when you remove the 'out' and replace with 'in', however
'inout' seems also to fail.

Assiging buf a value fails also.

compiled with -inline it works as expected, weird.

Charlie

In article <casd8k$18ac$1 digitaldaemon.com>, Sean Kelly says...

This code:

void fn( out char[] buf )
{
void fn2()
{
printf( "%i, %i: %.*s\n", buf.length, buf.size, buf );
}
fn2();
printf( "%i, %i: %.*s\n", buf.length, buf.size, buf );
}
int main()
{
char[] buf;
printf( "%i, %i: %.*s\n", buf.length, buf.size, buf );
fn( buf );
return 0;
}

prints:

0, 8:

0, 8:

So as far as I can tell, the stack pointer is messed up for dynamic arrays
within delegates.
Jun 17 2004
parent reply Sean Kelly <sean f4.ca> writes:
In article <casnln$1oe1$1 digitaldaemon.com>, Charlie says...

I dont see any delgates in there ?  I see the nested function ..


That's what I meant.  I should really learn to not post until after I've had
some coffee :)


This problem goes away when you remove the 'out' and replace with 'in', however
'inout' seems also to fail.


Yup, I noticed this too, so perhaps the problem is with parameter specifiers.
However I think there's a related bug that's affecting me too.  I'm still
working on shrinking this repro, but this is the best I've done so far:

interface I
{
void fn( char[] buf );
}

template T()
{
template doFn( CharT )
{
void fn( CharT[] val )
{
size_t	len	= 0;
CharT	ch;

void append( CharT ch )
{
printf( "%i, %i: %.*s\n", val.length, val.size, val );					
if( ++len > val.length )
val.length = len * 2;
val[len-1] = ch;
}
ch = '1';
append( ch );
ch = '2';
append( ch );
ch = '3';
append( ch );
ch = '4';
append( ch );

val.length = len;
return this;
}
}
mixin doFn!(char) doFnC;
alias doFnC.fn fn;
}

class C : I
{
mixin T;
}

int main()
{
char[] buf;
I i = new C();
i.fn( buf );
printf( "%i, %i: %.*s\n", buf.length, buf.size, buf );
return 0;
}

prints:

0, 8:
2, 8: 1
2, 8: 12
6, 8: 123
0, 8:

Basically, in this code I can't use the inner function "append" at all.  If I
add the "out" qualifier then the dynarray is garbage and I get an access
violation, and if I leave it out then the changes don't persist once I leave the
function.  With the above code, if every call to "append" is replaced with the
code inside "append" then everything works as expected.
Jun 17 2004
parent reply Sean Kelly <sean f4.ca> writes:
I just realized that this second problem is intended behavior.  I had mistakenly
thought that dynamic arrays would be governed by the same rules as classes, ie.
that they are reference counted and are always implicitly passed as inout.  It
looks like the out qualifier must be used in order to have changes to dynarrays
persist.  So the only bug I see is the corrupt stack pointer when using inner
classes.
Jun 17 2004
parent "Kris" <someidiot earthlink.dot.dot.dot.net> writes:
"Sean Kelly" wrote ...

 I just realized that this second problem is intended behavior.  I had


mistakenly

 thought that dynamic arrays would be governed by the same rules as


classes, ie.

 that they are reference counted and are always implicitly passed as inout.


It

 looks like the out qualifier must be used in order to have changes to


dynarrays

 persist.  So the only bug I see is the corrupt stack pointer when using


inner

 classes.


A while back I ran into this one also, Sean. I forget the title of the
bug-report, but the resultant code looks like this (note the char[] tmp for
getting around funky nesting stack alignment issues; you can't directly
assign the argument 'x' from within the nested function):

/*****************************************

  Extract a char array from the current read-position

*****************************************/

        IReader get (out char[] x)
        {
                char[]  tmp;
                int     size;

                // callback to assign the prescribed content
                int read (void[] content)
                {
                        tmp = cast(char[]) content;
                        if (! mapped)
                              tmp = tmp.dup;
                        return tmp.length;
                }

                // how many bytes in this array?
                get (size);
                buffer.preserve (size, &read);

                // set output argument and go home
                x = tmp;
                return this;
        }
Jun 17 2004