dtrace discuss - Apr 2006 - Clearing structure-based associative arrays

Consider:

struct str {
        string one;
        int two;
};

struct str foo[string];

BEGIN {
        foo["bar"].one = "baz";
        foo["bar"].two = 1;
}

tick-1s {
        foo["bar"].one = 0;
        foo["bar"].two = 0;
}

This gives:

dtrace: error on enabled probe ID 2 (ID 43463: profile:::tick-1s): invalid
address (0x0) in action #1 at DIF offset 52

which seems to make it impossible to clear associative arrays that contain
structures with string members. Note that you /can/ clear "string
foo[string]",
which makes me think that this is simply a bug. Or at least I''m
completely
missing a way to clear an entry in such an array.

I would have expected ''foo["bar"] = 0;'' to still
work, actually.

regards
john

John,
Setting the individual members of a structure to zero will not cause  
the structure to be deallocated, you''d actually need something like an
undef() function or something of the sort to do this.  There was some  
talk about providing such a function possibly a while ago, but I don''t
think there''s been a big pull for it so I''m not sure if it
ever
officially made it onto the list to be done.
>
> dtrace: error on enabled probe ID 2 (ID 43463: profile:::tick-1s):  
> invalid address (0x0) in action #1 at DIF offset 52
Also, you got an error because you cannot clear a string value by  
setting it to zero.  You can do this by doing:
  	foo["bar"].one = stringof(NULL);


Noel


************************************************************************ 
**

"Question all the answers"
On Apr 11, 2006, at 8:16 AM, John Levon wrote:
>
> Consider:
>
> struct str {
>         string one;
>         int two;
> };
>
> struct str foo[string];
>
> BEGIN {
>         foo["bar"].one = "baz";
>         foo["bar"].two = 1;
> }
>
> tick-1s {
>         foo["bar"].one = 0;
>         foo["bar"].two = 0;
> }
>
> This gives:
>
> dtrace: error on enabled probe ID 2 (ID 43463: profile:::tick-1s):  
> invalid address (0x0) in action #1 at DIF offset 52
>
> which seems to make it impossible to clear associative arrays that  
> contain
> structures with string members. Note that you /can/ clear "string  
> foo[string]",
> which makes me think that this is simply a bug. Or at least I''m  
> completely
> missing a way to clear an entry in such an array.
>
> I would have expected ''foo["bar"] = 0;'' to
still work, actually.
>
> regards
> john
>
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org

On Tue, Apr 11, 2006 at 09:14:07AM -0700, Noel Dellofano wrote:
> Setting the individual members of a structure to zero will not cause  
> the structure to be deallocated
So I suspected... is there a bug on this yet?
> >dtrace: error on enabled probe ID 2 (ID 43463: profile:::tick-1s):  
> >invalid address (0x0) in action #1 at DIF offset 52
> 
> Also, you got an error because you cannot clear a string value by  
> setting it to zero.  You can do this by doing:
>  	foo["bar"].one = stringof(NULL);
> 
> "Question all the answers"
OK then :) This actually does work for simple "string foo[string]"
arrays; why
the difference?

regards
john

> So I suspected... is there a bug on this yet?
I don''t believe so, or at least not that I can track down in bugster...
> OK then :) This actually does work for simple "string
foo[string]"
> arrays; why
> the difference?
Good question :)  I''m not really sure, I''m assuming it was a
design
decision.  Makes sense that you''d have to set a string to a string and
not a string to a number, but the Dtrace team can answer this one more  
definitively than I can.

Noel


************************************************************************ 
**

"Question all the answers"
On Apr 11, 2006, at 9:27 AM, John Levon wrote:
> On Tue, Apr 11, 2006 at 09:14:07AM -0700, Noel Dellofano wrote:
>
>> Setting the individual members of a structure to zero will not cause
>> the structure to be deallocated
>
> So I suspected... is there a bug on this yet?
>
>>> dtrace: error on enabled probe ID 2 (ID 43463: profile:::tick-1s):
>>> invalid address (0x0) in action #1 at DIF offset 52
>>
>> Also, you got an error because you cannot clear a string value by
>> setting it to zero.  You can do this by doing:
>>  	foo["bar"].one = stringof(NULL);
>>
>> "Question all the answers"
>
> OK then :) This actually does work for simple "string
foo[string]"
> arrays; why
> the difference?
>
> regards
> john

> > So I suspected... is there a bug on this yet?
> 
> I don''t believe so, or at least not that I can track down in
bugster...
> 
> > OK then :) This actually does work for simple "string
foo[string]"
> > arrays; why
> > the difference?
> 
> Good question :)  I''m not really sure, I''m assuming it
was a design
> decision.  Makes sense that you''d have to set a string to a string
and
> not a string to a number, but the Dtrace team can answer this one more  
> definitively than I can.
> 
> Noel
As Noel said, we need to add an operator to the compiler (e.g. undef()) or
the ability to assign a structure to the literal zero and have that clear it.
That capability does not currently exist.

-Mike

-- 
Mike Shapiro, Solaris Kernel Development. blogs.sun.com/mws/

On Tue, Apr 11, 2006 at 04:01:20PM -0700, Michael Shapiro wrote:
> As Noel said, we need to add an operator to the compiler (e.g. undef()) or
> the ability to assign a structure to the literal zero and have that clear
it.
> That capability does not currently exist.
I filed:

6411981 need a way to unallocate struct dynamic variables

thanks,
john