Re: [PATCH] Pipeline prefetch build-time control

On 26/05/2013 12:57 p.m., Alex Rousskov wrote:
> On 05/25/2013 01:42 AM, Amos Jeffries wrote:
>
>> So be it. This drops connOkToAddrequest() in favour of a
>> ConnStateData::concurrentRequestQueueFilled() which simply checks the
>> available queue space and replies true/false. All other state handling
>> relies on the existing code to behave properly. Pity anyone who hits a bug.
> Please note that I am not saying that the current code is correct or
> that all of the proposed changes were useless. The earlier patch did
> identify a couple of potentially problematic spots in the current code
> and had good bits that could be massaged into making that code better.
>
> While "this is the last request" changes would not be specific to
> pipelining, and would not require a complex multi-state data member,
> they would still be useful IMO. FWIW, I would welcome their resubmission
> (as a separate patch), perhaps adding a new flag with a setter method
> for debugging.
>

I know. I'm just in doubt about how long it will take to get back to
those parts of the system again during the audit.

pipeline patch applied with some tweaks as trunnk rev.12853.

Amos
Received on Sun May 26 2013 - 02:01:50 MDT