2.31: glibc master is now kinda, sorta, frozen

2.31: glibc master is now kinda, sorta, frozen

[Siddhesh Poyarekar]

Hello,

It's almost the end of the year where I am (and already 2020 further
east, hope your new year resolutions have not broken already!) and the
much anticipated release freeze is here, kinda.

If your patch has been reviewed and acked, please commit it as soon as
you can.  There are 4 release blockers in the list and we need to try to
get them off the list one way or another by the end of the week.

- iconv: rewrite option parsing
    https://sourceware.org/ml/libc-alpha/2019-12/msg00613.html

Arjun, how's this going?  I see a v2 but no reviews.  I'll review this
on Thursday if nobody gets to it by then.

- New DSO sorting algorithm and test script
    https://sourceware.org/ml/libc-alpha/2019-07/msg00472.html
    https://sourceware.org/ml/libc-alpha/2019-07/msg00473.html

Chung-Lin, based on Carlos' review comments, a new revision looks like a
fair amount of work, which makes this unsuitable as a release blocker.
I suggest we drop this from the blocker list so that you're not rushed
into pushing a patchset.  Please do post the patch for review in the
meantime since we're still OK to do reviews; just that you can only
commit after the 2.31 release is done.

Of course, if you want (and if the patch does not have ABI implications)
we can always backport to the 2.31 branch after the release.

- vDSO refactor and fixes
    https://sourceware.org/ml/libc-alpha/2019-12/msg00597.html

Adhemerval, I see some reviews and change requests.  Are you waiting for
reviews on the rest of your patches or do you intend to post a v3
patchset?  I can help with reviews on Thursday if nobody gets to it by then.

- Add libm_alias_finite for _finite symbols
    https://sourceware.org/ml/libc-alpha/2019-10/msg00627.html
    It moves the _finite symbol to compat symbols and not enable them
    for newer architectures.

Adhemerval, would you be OK with dropping this off the blocker list if a
review suggests that the patchset needs an update?  I'll do a review on
Thursday and if there are no changes needed, you can push it in.
However if the patchset needs changes, I'll help with reviews of the
updated patchset too, but you can commit after the 2.31 release.  Does
that sound reasonable?

Siddhesh

Re: 2.31: glibc master is now kinda, sorta, frozen

[Adhemerval Zanella]

On 31/12/2019 13:29, Siddhesh Poyarekar wrote:
> 
> - vDSO refactor and fixes
>     https://sourceware.org/ml/libc-alpha/2019-12/msg00597.html
> 
> Adhemerval, I see some reviews and change requests.  Are you waiting for
> reviews on the rest of your patches or do you intend to post a v3
> patchset?  I can help with reviews on Thursday if nobody gets to it by then.

I will check the review already done and send updated versions, thanks
for checking on it.

> 
> - Add libm_alias_finite for _finite symbols
>     https://sourceware.org/ml/libc-alpha/2019-10/msg00627.html
>     It moves the _finite symbol to compat symbols and not enable them
>     for newer architectures.
> 
> Adhemerval, would you be OK with dropping this off the blocker list if a
> review suggests that the patchset needs an update?  I'll do a review on
> Thursday and if there are no changes needed, you can push it in.
> However if the patchset needs changes, I'll help with reviews of the
> updated patchset too, but you can commit after the 2.31 release.  Does
> that sound reasonable?

I will be around on Thursday and Friday, so if see anything need change
we can try to sort this out by Friday and if by then it is not already
ready I will postpone to 2.31.

Re: 2.31: glibc master is now kinda, sorta, frozen

[Adhemerval Zanella]

On 31/12/2019 13:53, Adhemerval Zanella wrote:
> 
> 
> On 31/12/2019 13:29, Siddhesh Poyarekar wrote:
>>
>> - vDSO refactor and fixes
>>     https://sourceware.org/ml/libc-alpha/2019-12/msg00597.html
>>
>> Adhemerval, I see some reviews and change requests.  Are you waiting for
>> reviews on the rest of your patches or do you intend to post a v3
>> patchset?  I can help with reviews on Thursday if nobody gets to it by then.
> 
> I will check the review already done and send updated versions, thanks
> for checking on it.

Hi Siddhesh,

The latest version [1] received some sparse review that I fixed locally
and I don't think it is worth to send an updated version (I can if you
prefer).

[1] https://sourceware.org/ml/libc-alpha/2019-12/msg00597.html

Re: 2.31: glibc master is now kinda, sorta, frozen

[Rafal Luzynski]

Hi Siddhesh,

So I am about half an hour late.  Now I will not commit without a
permission,
so what is your opinion about this one?

https://sourceware.org/ml/libc-alpha/2019-12/msg00819.html

Regards,

Rafal

Re: 2.31: glibc master is now kinda, sorta, frozen

[Chung-Lin Tang]

On 2020/1/1 12:29 AM, Siddhesh Poyarekar wrote:
> - New DSO sorting algorithm and test script
>      https://sourceware.org/ml/libc-alpha/2019-07/msg00472.html
>      https://sourceware.org/ml/libc-alpha/2019-07/msg00473.html
> 
> Chung-Lin, based on Carlos' review comments, a new revision looks like a
> fair amount of work, which makes this unsuitable as a release blocker.
> I suggest we drop this from the blocker list so that you're not rushed
> into pushing a patchset.  Please do post the patch for review in the
> meantime since we're still OK to do reviews; just that you can only
> commit after the 2.31 release is done.

Hi Siddhesh,
as Carlos' review suggests, and what I'm working on right now, can this still
make 2.31 as a non-default tunable option? Since the new algorithm does have
some slight behavior difference, it may be better to make available earlier
than later for users to try.

I will try to update the patches and respond to Carlos' review in detail within
the week.

Thanks,
Chung-Lin

Re: 2.31: glibc master is now kinda, sorta, frozen

[Siddhesh Poyarekar]

On 06/01/20 9:11 pm, Chung-Lin Tang wrote:
> as Carlos' review suggests, and what I'm working on right now, can this
> still
> make 2.31 as a non-default tunable option? Since the new algorithm does
> have
> some slight behavior difference, it may be better to make available earlier
> than later for users to try.
> 
> I will try to update the patches and respond to Carlos' review in detail
> within
> the week.

I'm afraid that's going to be too long and too late.  The change is
non-trivial and it needs more time to settle in master.  Please do
continue the review process though since, like I said, you can always
backport to 2.31 after the release.  New tunables can be backported
since they're not considered part of ABI.

Siddhesh

Re: 2.31: glibc master is now kinda, sorta, frozen

[Arjun Shankar]

Hi Siddhesh,

> - iconv: rewrite option parsing
>     https://sourceware.org/ml/libc-alpha/2019-12/msg00613.html
> 
> Arjun, how's this going?  I see a v2 but no reviews.  I'll review this
> on Thursday if nobody gets to it by then.

Florian has reviewed earlier versions of it with me, and it has two tests
which exercise the front end. Given this, I feel pretty confident about the
patch, overall.

I'm just waiting for reviews to come in on libc-alpha, and then I'll get to
work on a follow-up or check-in depending on the response(s). Carlos is
planning on reviewing it, so I should have some feedback soon. Of course,
I'll be happy to have your review if you have time to spare.

Cheers,
Arjun

Re: 2.31: glibc master is now kinda, sorta, frozen

[Siddhesh Poyarekar]

On 06/01/20 9:48 pm, Arjun Shankar wrote:
> Florian has reviewed earlier versions of it with me, and it has two tests
> which exercise the front end. Given this, I feel pretty confident about the
> patch, overall.
> 
> I'm just waiting for reviews to come in on libc-alpha, and then I'll get to
> work on a follow-up or check-in depending on the response(s). Carlos is
> planning on reviewing it, so I should have some feedback soon. Of course,
> I'll be happy to have your review if you have time to spare.

I think you missed the review I did last week :)

Siddhesh

Re: 2.31: glibc master is now kinda, sorta, frozen

[Arjun Shankar]

Hi Siddhesh,

> I think you missed the review I did last week :)

Sorry, I did. I'm reading it now :)

Re: 2.31: glibc master is now kinda, sorta, frozen

[Chung-Lin Tang]

On 2020/1/6 11:46 PM, Siddhesh Poyarekar wrote:
> On 06/01/20 9:11 pm, Chung-Lin Tang wrote:
>> as Carlos' review suggests, and what I'm working on right now, can this
>> still
>> make 2.31 as a non-default tunable option? Since the new algorithm does
>> have
>> some slight behavior difference, it may be better to make available earlier
>> than later for users to try.
>>
>> I will try to update the patches and respond to Carlos' review in detail
>> within
>> the week.
> 
> I'm afraid that's going to be too long and too late.  The change is
> non-trivial and it needs more time to settle in master.  Please do
> continue the review process though since, like I said, you can always
> backport to 2.31 after the release.  New tunables can be backported
> since they're not considered part of ABI.
> 
> Siddhesh
> 
Okay I see, thanks for letting me know. I will still try to update the patches
for review soon.

Chung-Lin

Re: 2.31: glibc master is now kinda, sorta, frozen

[Carlos O'Donell]

On 1/7/20 4:07 AM, Chung-Lin Tang wrote:
> 
> 
> On 2020/1/6 11:46 PM, Siddhesh Poyarekar wrote:
>> On 06/01/20 9:11 pm, Chung-Lin Tang wrote:
>>> as Carlos' review suggests, and what I'm working on right now, can this
>>> still
>>> make 2.31 as a non-default tunable option? Since the new algorithm does
>>> have
>>> some slight behavior difference, it may be better to make available earlier
>>> than later for users to try.
>>>
>>> I will try to update the patches and respond to Carlos' review in detail
>>> within
>>> the week.
>>
>> I'm afraid that's going to be too long and too late.  The change is
>> non-trivial and it needs more time to settle in master.  Please do
>> continue the review process though since, like I said, you can always
>> backport to 2.31 after the release.  New tunables can be backported
>> since they're not considered part of ABI.
>>
>> Siddhesh
>>
> Okay I see, thanks for letting me know. I will still try to update the patches
> for review soon.

If I can help please tell me.

I agree with Siddhesh in that I think we can backport this change.

Keep in mind that when this langs in master we'll immediately pull it into
Fedora Rawhide for testing (within the week). This is really the gold standard
for making sure we didn't break anything :-)

Even Florian's changes to NODELETE which seemed innocuous revealed all sorts
of latent expectations from Firefox, Chrome, and Calibre.

-- 
Cheers,
Carlos.