Re: 2.29 freeze update: Last fortnight

[Carlos O'Donell <carlos at redhat dot com>]

On 1/21/19 10:28 PM, Siddhesh Poyarekar wrote:
> On 16/01/19 9:17 AM, Siddhesh Poyarekar wrote:
>> On 16/01/19 1:47 AM, Carlos O'Donell wrote:
>>> I'm working at trying to fix the rwlock bug. I'd like to see the bug
>>> fix included in the release. I think I need the rest of the week to
>>> fix this. However, we can push out to 2.30 if we want.
>>
>> We can hold off till 23rd for the hard freeze; would that be enough time
>> for you?  That gives me enough time to run tests over the weekend and do
>> the release the following week.
>>
>> That should also give HJ enough time to fix the x32 bug he just found.
> 
> Carlos, HJ, are we doing OK for time?  Would you be OK for a hard freeze
> by the end of this week, i.e. Sunday?  That will give me all of next
> week to run final tests for the release.
> 
> How's the strftime change doing?  It looks like y'all are on the last
> couple of patches now.
> 
> As for the ChangeLog script, I've worked on incorporating it into gnulib
> and am targeting end of my day tomorrow to post a patch for review on
> the gnulib ML.  I believe we are on target for dropping the ChangeLog
> requirement for 2.30 development, so I'll be requesting review of wiki
> changes next week indicating the change in workflow.

I just posed the complete rwlock fix with regression tests:
https://www.sourceware.org/ml/libc-alpha/2019-01/msg00557.html

Analysis with execution ordering showing failures:
https://sourceware.org/bugzilla/show_bug.cgi?id=23844#c14

I don't expect anyone to give substantive review. Torvald was the
expert, and he mostly wrote the patch with input from Rik Prohaska.
I basically did the review, and created the test cases from my own
review, and carried out testing on x86_64 to determine how fast the
stall can be triggered (to set test timeouts and sleeps). Reviewing
this requires a complete understanding of the new rwlock algorithm.
However, at this point it is much better than the stalls we have and
so it should go in ASAP.

I expect to commit the fix as soon as Torvald and Rik let me use
their Signed-off-by.

-- 
Cheers,
Carlos.