On 5/6/09 12:59 PM, Ingo Strüwing wrote:
>> I still don't like the two seconds sleep part.
> I understand that. But please be more definite if you want me to get rid
> of it.
> So, after all, I wonder what is left from my patch?
> 1. Checking thd->killed after enter_cond() doesn't help as we have no
> guarantee that we will see the latest change of this variable in another
But eliminates one type of race. Let's keep it and remove the redundant
checks before enter_cond.
> 2. Closing the connection doesn't help as we have no guarantee that this
> will prevent another thread from blocking in read(2).
But let's keep your approach as in for the time being.
> 3. It is not important that we test if we can interrupt read(2) on all
Yes, i don't regard it as important.
> 4. Re-formatting of kill.test is not appreciated.
The re-factoring is appreciated, the re-formatting.. not so much.
> Under these circumstances I don't see a value in this patch any more. I
> draw back my proposal. The remaining question is if to reassign to
> someone else, to declare it "not a bug: KILL is unreliable by design",
> or "won't fix"?
There is still value in your patch. Let's report another bug that "KILL
is broken by design" and someone else will get to fix it.
-- Davi Arnaut