6:13 a.m.
Hello everyone,
I would like to ask what do you think about enabling c++11 in library in
F22? Is it permitted by Fedora guidelines?
I am asking because of this bug -
https://bugzilla.redhat.com/show_bug.cgi?id=1274307 . And I want to know
the opinion of someone else, to decide.
Thank you,
Marek
6:23 a.m.
On Mon, Oct 26, 2015 at 11:13 AM, Marek Skalický <mskalick(a)redhat.com> wrote:
Hello everyone,
I would like to ask what do you think about enabling c++11 in library in
F22? Is it permitted by Fedora guidelines?
I am asking because of this bug -
https://bugzilla.redhat.com/show_bug.cgi?id=1274307 . And I want to know
the opinion of someone else, to decide.
It was explicitly not enabled in F-22 because it's not backwards
compatible and required a mass rebuild for all c++ applications. It's
enabled by default, and we had a mass rebuild as a result, in Fedora
23. As a result of the impact we won't be making such changes for a
release that is long GA.
Peter
6:29 a.m.
On 26/10/15 11:23, Peter Robinson wrote:
On Mon, Oct 26, 2015 at 11:13 AM, Marek Skalický
<mskalick(a)redhat.com> wrote:
> Hello everyone,
> I would like to ask what do you think about enabling c++11 in library in
> F22? Is it permitted by Fedora guidelines?
>
> I am asking because of this bug -
> https://bugzilla.redhat.com/show_bug.cgi?id=1274307 . And I want to know
> the opinion of someone else, to decide.
It was explicitly not enabled in F-22 because it's not backwards
compatible and required a mass rebuild for all c++ applications. It's
enabled by default, and we had a mass rebuild as a result, in Fedora
23. As a result of the impact we won't be making such changes for a
release that is long GA.
You're confusing two different things, though without knowing what that
configure patch actually does we can't know what the answer is here.
There is, as far as I know, no problem using -std=c++11 in F22 or any
other version of Fedora, it's just that the result might not be fully
standards compliant.
Specifically in F22 there are two minor changes which require a change
to the library ABI and as F22 was built with the old ABI you won't get
those changes. Neither should affect whether code compiles and works though.
So the only thing you must not do in F22 is to set the define that
selects the new ABI but you would probably find out quite quickly as you
did as you would fail to link if you used any other C++ libraries that
had been built with the old ABI.
Which is why I say we need to know what the configure flag does - my
guess is that it just sets -std=c++11 which is fine. If it also sets the
define to select the new ABI then that might be a problem.
Tom
--
Tom Hughes (tom(a)compton.nu)
http://compton.nu/
7:10 a.m.
On Mon, 2015-10-26 at 11:29 +0000, Tom Hughes wrote:
On 26/10/15 11:23, Peter Robinson wrote:
> On Mon, Oct 26, 2015 at 11:13 AM, Marek Skalický <mskalick(a)redhat.com> wrote:
>> Hello everyone,
>> I would like to ask what do you think about enabling c++11 in library in
>> F22? Is it permitted by Fedora guidelines?
>>
>> I am asking because of this bug -
>> https://bugzilla.redhat.com/show_bug.cgi?id=1274307 . And I want to know
>> the opinion of someone else, to decide.
>
> It was explicitly not enabled in F-22 because it's not backwards
> compatible and required a mass rebuild for all c++ applications. It's
> enabled by default, and we had a mass rebuild as a result, in Fedora
> 23. As a result of the impact we won't be making such changes for a
> release that is long GA.
You're confusing two different things, though without knowing what that
configure patch actually does we can't know what the answer is here.
There is, as far as I know, no problem using -std=c++11 in F22 or any
other version of Fedora, it's just that the result might not be fully
standards compliant.
Specifically in F22 there are two minor changes which require a change
to the library ABI and as F22 was built with the old ABI you won't get
those changes. Neither should affect whether code compiles and works though.
So the only thing you must not do in F22 is to set the define that
selects the new ABI but you would probably find out quite quickly as you
did as you would fail to link if you used any other C++ libraries that
had been built with the old ABI.
Which is why I say we need to know what the configure flag does - my
guess is that it just sets -std=c++11 which is fine. If it also sets the
define to select the new ABI then that might be a problem.
Yes it only set -std=c++11. However there could be some problem with
linking application with -std=c++1 to driver library built without
-std=c++11 - upstream bug https://jira.mongodb.org/browse/CXX-458
F22 is not too old to thinks like this? (to force dependent application
to rebuild,...)?
Or is the mentioned bug only problem of different ABIs and without
changing value of _GLIBCXX_USE_CXX11_ABI there couldn't be any problem?
Thanks,
Marek
Tom
--
Tom Hughes (tom(a)compton.nu)
http://compton.nu/
7:15 a.m.
I'm afraid of some runtime dependencies clash, small amount of
rebuilds to all (r-)deps?
--
Yours sincerely,
Christopher Meng
http://awk.io
7:50 a.m.
Hi.
As the reporter let me briefly phrase the main concerns:
- the driver works only with software which is built for the same ABI,
it is one of the cases where things break immediately otherwise
- we ran into this problem when building 3rd-party software (referee box
of the RoboCup Logistics League competition). This requires C++11 for
other dependencies and therefore does not build with mongo-cxx-driver on
F22.
- we will have this problem again for the upcoming release of Fawkes,
that uses mongodb and requires C++11 in some places
- at this time, we cannot upgrade to anything beyond F22
- there are only two packages which depend on mongocxx-driver:
- iwhd: broken, as soon as MongoDB used crashes. This is not due ABI
incompatibility, but because it is not properly initializing the
library -- but in either case needs an update to actually work
- uwsgi plugins: these are explicitly built with -std=c++11 (cf.
plugins/gridfs/uwsgiplugin.py in the prep'ed package)
I have not run this myself to be fair, but given the different ABI
it will break, cf. https://jira.mongodb.org/browse/CXX-558
Therefore, no depending package currently works with this driver out of
the box. Since we need to make changes to all packages anyway, my
suggestion is to upgrade to the latest upstream release and enable C++11
to be future-proof. Within our eco-systems things are as of now broken
(iwhd, uwsgi) or prevent an upgrade/full feature set (fawkes).
I hope that this is convincing enough to make a clean cut and enable
C++11 in mongo-cxx-driver.
Regards,
Tim
On 26.10.2015 12:13, Marek Skalický wrote:
Hello everyone,
I would like to ask what do you think about enabling c++11 in library in
F22? Is it permitted by Fedora guidelines?
I am asking because of this bug -
https://bugzilla.redhat.com/show_bug.cgi?id=1274307 . And I want to know
the opinion of someone else, to decide.
Thank you,
Marek
7:59 a.m.
On 26/10/15 12:50, Tim Niemueller wrote:
- the driver works only with software which is built for the same
ABI,
it is one of the cases where things break immediately otherwise
You need to be very clear what you mean here. Do you mean built with
-std=c++11 or built with -D_GLIBCXX_USE_CXX11_ABI=1?
The latter is, as I understand it, unacceptable in Fedora. In any event
it's unlikely to even build and link because all the other libraries are
not built with that so if it tries to link to any C++ libraries other
than libstdc++ then the link will likely fail.
Tom
--
Tom Hughes (tom(a)compton.nu)
http://compton.nu/
8:01 a.m.
On 26.10.2015 13:59, Tom Hughes wrote:
On 26/10/15 12:50, Tim Niemueller wrote:
> - the driver works only with software which is built for the same ABI,
> it is one of the cases where things break immediately otherwise
You need to be very clear what you mean here. Do you mean built with
-std=c++11 or built with -D_GLIBCXX_USE_CXX11_ABI=1?
I mean the former, -std=c++11.
Tim
8:09 a.m.
On 26/10/15 13:01, Tim Niemueller wrote:
On 26.10.2015 13:59, Tom Hughes wrote:
> On 26/10/15 12:50, Tim Niemueller wrote:
>
>> - the driver works only with software which is built for the same ABI,
>> it is one of the cases where things break immediately otherwise
>
> You need to be very clear what you mean here. Do you mean built with
> -std=c++11 or built with -D_GLIBCXX_USE_CXX11_ABI=1?
I mean the former, -std=c++11.
Right, so you should stop talking about ABI then, because setting that
doesn't change the ABI. That's kind of the whole point.
All it does is enable C++11 language features.
Now that might cause the ABI of the package you are building to change
if it has ifdefs to add extra functions in that case or something but
it's not what gcc refers to as the "C++11 ABI".
Tom
--
Tom Hughes (tom(a)compton.nu)
http://compton.nu/
8:23 a.m.
https://gcc.gnu.org/wiki/Cxx11AbiCompatibility
I guess this is what is causing problems if mongo-cxx-driver and an
application using it do not use the same language version.
Tim
On 26.10.2015 14:09, Tom Hughes wrote:
On 26/10/15 13:01, Tim Niemueller wrote:
> On 26.10.2015 13:59, Tom Hughes wrote:
>> On 26/10/15 12:50, Tim Niemueller wrote:
>>
>>> - the driver works only with software which is built for the same ABI,
>>> it is one of the cases where things break immediately otherwise
>>
>> You need to be very clear what you mean here. Do you mean built with
>> -std=c++11 or built with -D_GLIBCXX_USE_CXX11_ABI=1?
>
> I mean the former, -std=c++11.
Right, so you should stop talking about ABI then, because setting that
doesn't change the ABI. That's kind of the whole point.
All it does is enable C++11 language features.
Now that might cause the ABI of the package you are building to change
if it has ifdefs to add extra functions in that case or something but
it's not what gcc refers to as the "C++11 ABI".
Tom
7:01 a.m.
On Mon, 2015-10-26 at 14:23 +0100, Tim Niemueller wrote:
https://gcc.gnu.org/wiki/Cxx11AbiCompatibility
I guess this is what is causing problems if mongo-cxx-driver and an
application using it do not use the same language version.
Does someone know it this could be a problem for F22?
Marek
Tim
On 26.10.2015 14:09, Tom Hughes wrote:
> On 26/10/15 13:01, Tim Niemueller wrote:
>> On 26.10.2015 13:59, Tom Hughes wrote:
>>> On 26/10/15 12:50, Tim Niemueller wrote:
>>>
>>>> - the driver works only with software which is built for the same ABI,
>>>> it is one of the cases where things break immediately otherwise
>>>
>>> You need to be very clear what you mean here. Do you mean built with
>>> -std=c++11 or built with -D_GLIBCXX_USE_CXX11_ABI=1?
>>
>> I mean the former, -std=c++11.
>
> Right, so you should stop talking about ABI then, because setting that
> doesn't change the ABI. That's kind of the whole point.
>
> All it does is enable C++11 language features.
>
> Now that might cause the ABI of the package you are building to change
> if it has ifdefs to add extra functions in that case or something but
> it's not what gcc refers to as the "C++11 ABI".
>
> Tom
>
10:33 a.m.
On 27/10/15 13:01 +0100, Marek Skalický wrote:
On Mon, 2015-10-26 at 14:23 +0100, Tim Niemueller wrote:
> https://gcc.gnu.org/wiki/Cxx11AbiCompatibility
>
> I guess this is what is causing problems if mongo-cxx-driver and an
> application using it do not use the same language version.
>
Does someone know it this could be a problem for F22?
Which part?
That page lists several things, many of which are not relevant for GCC
5.2, and many of which are very unliely to be a problem in practice.
This entire thread seems to be completely devoid of details, so it's
no wonder no-one gave a useful answer (although Tom Hughes is
completely correct in his statements about -std=c++11 being distinct
from the new ABI supported by GCC 5.2).
3098
days inactive
3114
days old
11 comments
6 participants
participants (6)
-
Christopher Meng -
Jonathan Wakely -
Marek Skalický -
Peter Robinson -
Tim Niemueller -
Tom Hughes