389-ds-base and freeipa on 32 bit arches

[Randy Barlow]

Greetings gcc maintainers!

A FESCo issue[0] has been filed due to the dropping of 389-ds-base and
freeipa on 32 bit arches for Fedora 28. This was done without a change
request being filed, so FESCo is trying to decide how best to handle it.

It seems there are some concerns about whether the C tooling correctly
handles some cases for 32-bit arches that led to the decision by the IPA
maintainers to drop support for 32-bit. FESCo would like to ask for the
GCC maintainers' input on the issue.

Thanks in advance for your feedback!


[0] https://pagure.io/fesco/issue/1845
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1544386

[Jeff Law]

On 02/23/2018 09:58 AM, Randy Barlow wrote:
>  Greetings gcc maintainers!
>  
>  A FESCo issue[0] has been filed due to the dropping of 389-ds-base and
>  freeipa on 32 bit arches for Fedora 28. This was done without a change
>  request being filed, so FESCo is trying to decide how best to handle it.
>  
>  It seems there are some concerns about whether the C tooling correctly
>  handles some cases for 32-bit arches that led to the decision by the IPA
>  maintainers to drop support for 32-bit. FESCo would like to ask for the
>  GCC maintainers' input on the issue.
>  
>  Thanks in advance for your feedback! 
GCC, binutils, etc all support 32-bit arches
just fine.  THe current
thinking is there are serious concurrency concerns in the freeipa code
which are leading to the failures they were seeing on 32 bit platforms
(i686 in particular).

Jeff

[Alexander Bokovoy]

On pe, 23 helmi 2018, Jeff Law wrote:
> On 02/23/2018 09:58 AM, Randy Barlow wrote:
> > Greetings gcc maintainers!
> >
> > A FESCo issue[0] has been filed due to the dropping of 389-ds-base and
> > freeipa on 32 bit arches for Fedora 28. This was done without a change
> > request being filed, so FESCo is trying to decide how best to handle it.
> >
> > It seems there are some concerns about whether the C tooling correctly
> > handles some cases for 32-bit arches that led to the decision by the IPA
> > maintainers to drop support for 32-bit. FESCo would like to ask for the
> > GCC maintainers' input on the issue.
> >
> > Thanks in advance for your feedback!
> GCC, binutils, etc all support 32-bit arches just fine.  THe current
> thinking is there are serious concurrency concerns in the freeipa code
> which are leading to the failures they were seeing on 32 bit platforms
> (i686 in particular). 
To be clear there are concerns in 389-ds code that lead to
data
corruptions on i686. They aren't seen on other 32-bit platforms Fedora
supports according to 389-ds developers.

FreeIPA has no choice as 389-ds-base was moved to exclude i686
architecture according to the process defined in Fedora.
-- 
/ Alexander Bokovoy

[Jeff Law]

On 02/23/2018 10:29 AM, Alexander Bokovoy wrote:
>  On pe, 23 helmi 2018, Jeff Law wrote:
> > On 02/23/2018 09:58 AM, Randy Barlow wrote:
> >> Greetings gcc maintainers!
> >>
> >> A FESCo issue[0] has been filed due to the dropping of 389-ds-base and
> >> freeipa on 32 bit arches for Fedora 28. This was done without a change
> >> request being filed, so FESCo is trying to decide how best to handle it.
> >>
> >> It seems there are some concerns about whether the C tooling correctly
> >> handles some cases for 32-bit arches that led to the decision by the IPA
> >> maintainers to drop support for 32-bit. FESCo would like to ask for the
> >> GCC maintainers' input on the issue.
> >>
> >> Thanks in advance for your feedback!
> > GCC, binutils, etc all support 32-bit arches just fine.  THe current
> > thinking is there are serious concurrency concerns in the freeipa code
> > which are leading to the failures they were seeing on 32 bit platforms
> > (i686 in particular).
>  To be clear there are concerns in 389-ds code that lead to data
>  corruptions on i686. They aren't seen on other 32-bit platforms Fedora
>  supports according to 389-ds developers. 
Sorry to mis-characterize the problem. 
It's the 389-ds code with the
concurrency concerns, not freeipa.

The fact that the 389-ds code works on other architectures does not
allow us to draw any conclusions at this point.  It really needs to go
through a root cause analysis to determine exactly why it is not working
correctly.  That will in turn tell us if it's the 389-ds code or
something in the compiler (or elsewhere) that is the problem.


Jeff

[Jared K. Smith]

On Fri, Feb 23, 2018 at 12:32 PM, Jeff Law <law(a)redhat.com&gt; wrote:

>  The fact that the 389-ds code works on other architectures does not
>  allow us to draw any conclusions at this point.
>  
Can you give any more details on 32-bit ARM in particular -- is it just
i686 that is experiencing issues, or have you seen similar memory
corruption issues on 32-bit ARM as well?

--
Jared Smith

[Jared K. Smith]

On Fri, Feb 23, 2018 at 6:39 PM, Jared K. Smith <jsmith(a)fedoraproject.org&gt;
wrote:

>  Can you give any more details on 32-bit ARM in particular -- is it
> just
>  i686 that is experiencing issues, or have you seen similar memory
>  corruption issues on 32-bit ARM as well?
>  
Just another gentle ping -- do we have any more information here that can
help FESCo as it makes a decision about this issue?

-Jared