On Sat, May 21, 2005 at 07:49:04PM +0100, Dave Mitchell wrote:
> On Sat, May 21, 2005 at 04:25:10PM -0000, Nicholas Clark wrote:
> > There is a bug in duping - GvHV() can end up pointing at garbage.
>
>
>
> I can reproduce it with:
>
> our $foo : unique = 'abc';
> use threads;
> threads->create(sub { 1} )->join();
>
> It seems to be down to the code in S_gv_share(), which at clone time
> basically fills all unused slots (eg gv_hv) in the :unique GV with
> &PL_sv_no from the *parent* thread:
Surely that logic is pants, in that:
a: it should be PL_sv_no from the child thread
and
b: even PL_sv_no is wrong, because too much other code assumes that GVs
point to the "correct" type.
Can a parent thread go away before its child? What happens if a child thread
creates a new unique GV, spawns a grandchild and then quits? Does this mess
with the assumptions of the :unique implementation?i
Nicholas Clark
