www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - [D1] gc safety

reply %u <e ee.com> writes:
How gc unfriendly is an union of objects and sizet_t?

union{
  size_t arr[10];
  Class obj[10];
}


And, if multiple arrays contain exclusively the same objects, is it then
safe/useful to mark all but the smallest array with gc.hasNoPointers?
Any object removal/addition happens simultaneous across all the arrays.

void remove(size_t i){
  size_t i2 = arr[i].i2;
  arr2[i2] = null;
  arr[i] = null;
}

ps. I was tempted to make the subject "[D1,gc(safety),union]"..
Oct 10 2010
next sibling parent %u <e ee.com> writes:
Please tell me if anything is unclear/stupid :)

== Quote from %u (e ee.com)'s article

 How gc unfriendly is an union of objects and sizet_t?
 union{
   size_t arr[10];
   Class obj[10];
 }
 And, if multiple arrays contain exclusively the same objects, is it then
 safe/useful to mark all but the smallest array with gc.hasNoPointers?
 Any object removal/addition happens simultaneous across all the arrays.
 void remove(size_t i){
   size_t i2 = arr[i].i2;
   arr2[i2] = null;
   arr[i] = null;
 }
Oct 12 2010
prev sibling next sibling parent reply "Simen kjaeraas" <simen.kjaras gmail.com> writes:
%u <e ee.com> wrote:


 How gc unfriendly is an union of objects and sizet_t?

 union{
   size_t arr[10];
   Class obj[10];
 }



All elements of this union will be considered pointers by the GC.



 And, if multiple arrays contain exclusively the same objects, is it then
 safe/useful to mark all but the smallest array with gc.hasNoPointers?
 Any object removal/addition happens simultaneous across all the arrays.



As long as there is at least one GC registered pointer to the allocated
objects, it will not be collected. Hence, if some arrays are permutations
or subsets of another array, only the original array need to be marked as
having pointers.

-- 
Simen
Oct 12 2010
parent reply %u <e ee.com> writes:
== Quote from Simen kjaeraas (simen.kjaras gmail.com)'s article

 %u <e ee.com> wrote:

 How gc unfriendly is an union of objects and sizet_t?

 union{
   size_t arr[10];
   Class obj[10];
 }






this to the GC?


 And, if multiple arrays contain exclusively the same objects, is it then
 safe/useful to mark all but the smallest array with gc.hasNoPointers?
 Any object removal/addition happens simultaneous across all the arrays.



 objects, it will not be collected. Hence, if some arrays are permutations
 or subsets of another array, only the original array need to be marked as
 having pointers.



The GC will probably really like this as one of the arrays is cubed the size of
the smallest one :D
Oct 12 2010
parent %u <e ee.com> writes:
== Quote from Simen kjaeraas (simen.kjaras gmail.com)'s article

 %u <e ee.com> wrote:

 == Quote from Simen kjaeraas (simen.kjaras gmail.com)'s article

 %u <e ee.com> wrote:

 How gc unfriendly is an union of objects and sizet_t?

 union{
   size_t arr[10];
   Class obj[10];
 }






 to explain
 this to the GC?



 elements will be objects, and the rest will be size_ts that are not
 pointers?
Oct 13 2010
prev sibling next sibling parent "Simen kjaeraas" <simen.kjaras gmail.com> writes:
%u <e ee.com> wrote:


 == Quote from Simen kjaeraas (simen.kjaras gmail.com)'s article

 %u <e ee.com> wrote:

 How gc unfriendly is an union of objects and sizet_t?

 union{
   size_t arr[10];
   Class obj[10];
 }






 to explain
 this to the GC?



Perhaps. Is there a way to explain it to me? Do you mean the first x
elements will be objects, and the rest will be size_ts that are not
pointers?

-- 
Simen
Oct 12 2010
prev sibling parent "Simen kjaeraas" <simen.kjaras gmail.com> writes:
%u <e ee.com> wrote:


 Perhaps. Is there a way to explain it to me? Do you mean the first x
 elements will be objects, and the rest will be size_ts that are not
 pointers?






No. The GC only cares about whole blocks, not singular locations within
each. That said, there is a bug report with a GC that uses precise
scanning (http://d.puremagic.com/issues/show_bug.cgi?id=3463). As you
are using D1, I believe it should work for you.

-- 
Simen
Oct 13 2010