Most of the docs also don't expect you run a pointer out of the bounds of an array, either, but the compiler trusts you know what you're doing.

I see what you mean in the first question. Can we make a patch to test it dynamically?

On Fri, Oct 2, 2015 at 12:36 AM, Jonathan Wilkes <jancsika@yahoo.com> wrote:
For the first question, here's what I'm thinking:
1) create [boat(---[float]
2) click [boat(
3) error associated with [float].  (string ".x1234567" associated with [float] gets saved in Pd window of GUI)
4) delete [float]
5) create [clip]
6) somehow the OS happens to use addy "1234567" for [clip]
7) click the error in the Pd window
8) GUI sends "pd findinstance .x1234567" to Pd
9) Pd assigns "1234567" to error_object
10) Pd compares [clip] object's "1234567" to old [float] addy "1234567"
11) It's a match!
12) [clip] is falsely accused

For #2-- I guess I'm just anxious since most of the docs I've read assume
a void* is either NULL or pointing to something that exists.

-Jonathan



On Friday, October 2, 2015 12:15 AM, Matt Barber <brbrofsvl@gmail.com> wrote:


​So, if you deleted [float] and created [clip], isn't it going to bash whatever c string was associated with [float] and associate it with [clip]? Moreover, that kind of error has to occur in an object that exists in the current state (I think?), so the old [float] (once it's gone) could never be part of an error in the first place.

For 2) I think it depends on what you're going to do with the void pointer(s). Your compare_pointers() function could actually be read as the meat of a "guess my address!" roulette game. If you decided to write to *bar in case it matched, it might be a "guess my address!" Russian roulette game. I'm not sure what the compiler would say about that since I think you'd have to cast *bar back to something you could write.



On Thu, Oct 1, 2015 at 11:57 PM, Jonathan Wilkes <jancsika@yahoo.com> wrote:
Well, two questions I guess:
1) false positives-- if I deleted [float] and create [clip], can't malloc use the addy
that belonged to [float]?  In that case [clip] could get associated with an error
it had nothing to do with.
2) Is it undefined behavior to check void* garbage for equality?
And just for the heck of it...
3) is there a way to create something like [readpd~] which would take indices
as input and output the corresponding bytes of the running Pd instance? :)

-Jonathan





On Thursday, October 1, 2015 11:22 PM, Matt Barber <brbrofsvl@gmail.com> wrote:


The left side is still determined by the current state of the patch, though -- it's only going to check objects that are still there, which any garbage on the right won't ever match (right?). If there is a match, it's going to be because the state on both sides of the == was updated when the object was created. We never really have to worry about false positives, so I'm not sure the random long is the same situation.

On Thu, Oct 1, 2015 at 9:54 PM, Jonathan Wilkes <jancsika@yahoo.com> wrote:
But if you trace error_object back, you'll see it gets created from an sscanf
of a c string.  And that string was stored in the tcl/tk "text" widget as state bound
to a <ctrl-click> proc (or in post-1980s version of Pd that I work on, a
"hyperlink").

That state can persist well past the life of the object it referred to.  For example,
the error_object could have been deleted by the user.

That's why I was generating a random long in my contrived example.  If we
cast garbage to void* and put it to the right of the equals sign, isn't libc
technically allowed to respond by serving Pd users a listicle of the top 10 C
programming references available from Amazon with free shipping?

-Jonathan



On Thursday, October 1, 2015 7:32 PM, Matt Barber <brbrofsvl@gmail.com> wrote:


As I understand it, you can compare void pointers because they just store addresses.

In g_editor.c:


static int glist_dofinderror(t_glist *gl, void *error_object)
{
    t_gobj *g;
    for (g = gl->gl_list; g; g = g->g_next)
    {
        if ((void *)g == error_object)
        {
            /* got it... now show it. */
            glist_noselect(gl);
            canvas_vis(glist_getcanvas(gl), 1);
            canvas_editmode(glist_getcanvas(gl), 1.);
            glist_select(gl, g);
            return (1);
        }
        else if (g->g_pd == canvas_class)
        {
            if (glist_dofinderror((t_canvas *)g, error_object))
                return (1);
        }
    }
    return (0);
}


this function takes a pointer to void (storing the address of an object) as its second argument void *error_object. It can't know what the type of that object is, because it's being called from somewhere else, and it could be any kind of Pd object. That somewhere else knows what kind of object it is and (more importantly) where that object's address is, and just passes that address in. Then t_gobj *g; traverses the canvas, and since the address of each object is known, each of those can be compared to the object address passed in until there's a match. This is kind of a way of getting around the usual strong typing in c; as long as we know from both ends of the transaction that we can get valid addresses of what we're interested in, there's no problem just comparing those addresses to see if we've found the same object.



On Thu, Oct 1, 2015 at 4:45 PM, Jonathan Wilkes via Pd-list <pd-list@lists.iem.at> wrote:
Hi list,

int compare_pointers(t_pd *foo)
{
     long bar = generate_random_long();
     return (((void *)foo) == ((void *)bar));
}

(I probably have unnecessary parens there...)

Is the check for equality a case of undefined behavior?

If so, doesn't glob_findinstance of s_print.c also lead to the same
undefined behavior?

-Jonathan
 

_______________________________________________
Pd-list@lists.iem.at mailing list
UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list