Hi Takashi,
looks good, but...
On Apr 6 19:57, Takashi Yano wrote:
> @@ -1685,7 +1700,15 @@ pthread_key::~pthread_key ()
> */
> if (magic != 0)
> {
> - keys.remove (this);
> + LONG64 seq = keys[key_idx].seq;
> + assert (pthread_key::keys_list::ready (seq)
> + && InterlockedCompareExchange64 (&keys[key_idx].seq,
> + seq + 1, seq) == seq);
...do we really want to assert here? Shouldn't this better just skip
the rest of the function?
> + while (InterlockedCompareExchange64 (&keys[key_idx].busy_cnt,
> + INT64_MIN, 0) > 0)
> + yield ();
> + keys[key_idx].key = NULL;
> + InterlockedIncrement64 (&keys[key_idx].seq);
> TlsFree (tls_index);
> }
> }
Corinna
