Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

[David Stacey]

On 30/09/15 12:15, JonY wrote:
> gcc-5.2.0-1 has been uploaded for 32bit and 64bit Cygwin.
>
> This is the first series of the 5.x releases, and should be considered
> as experimental as such.

Excellent - looking forward to trying this.

Have you managed to work around the ABI change in gcc-5 [1], or will 
this require a mass rebuild at the point gcc-5 becomes 'current'?

Dave.

[1] - http://developerblog.redhat.com/2015/02/05/gcc5-and-the-c11-abi/


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

[JonY]

[-- Attachment #1: Type: text/plain, Size: 651 bytes --]

On 10/1/2015 00:05, David Stacey wrote:
> On 30/09/15 12:15, JonY wrote:
>> gcc-5.2.0-1 has been uploaded for 32bit and 64bit Cygwin.
>>
>> This is the first series of the 5.x releases, and should be considered
>> as experimental as such.
> 
> Excellent - looking forward to trying this.
> 
> Have you managed to work around the ABI change in gcc-5 [1], or will
> this require a mass rebuild at the point gcc-5 becomes 'current'?
> 
> Dave.
> 
> [1] - http://developerblog.redhat.com/2015/02/05/gcc5-and-the-c11-abi/
> 

Hi,

As far as I know, every gcc release will break C++ ABI, so it would mean
rebuilding everything C++.



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 801 bytes --]

Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

[David Stacey]

On 30/09/15 23:34, JonY wrote:
> On 10/1/2015 00:05, David Stacey wrote:
>> On 30/09/15 12:15, JonY wrote:
>>> gcc-5.2.0-1 has been uploaded for 32bit and 64bit Cygwin.
>>>
>>> This is the first series of the 5.x releases, and should be considered
>>> as experimental as such.
>>
>> Have you managed to work around the ABI change in gcc-5 [1], or will
>> this require a mass rebuild at the point gcc-5 becomes 'current'?
>>
>> [1] -http://developerblog.redhat.com/2015/02/05/gcc5-and-the-c11-abi/
> As far as I know, every gcc release will break C++ ABI, so it would mean
> rebuilding everything C++.

According to the Red Hat blog above, the last time g++ caused an ABI 
change was back in the 3.x days, so it hasn't happened for a while. Ah 
well, we have maintainers for most packages in Cygwin, so we'll have to 
co-ordinate a rebuild.

Dave.


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

[cyg Simple]

On 9/30/2015 7:36 PM, David Stacey wrote:
> On 30/09/15 23:34, JonY wrote:
>> On 10/1/2015 00:05, David Stacey wrote:
>>> On 30/09/15 12:15, JonY wrote:
>>>> gcc-5.2.0-1 has been uploaded for 32bit and 64bit Cygwin.
>>>>
>>>> This is the first series of the 5.x releases, and should be considered
>>>> as experimental as such.
>>>
>>> Have you managed to work around the ABI change in gcc-5 [1], or will
>>> this require a mass rebuild at the point gcc-5 becomes 'current'?
>>>
>>> [1] -http://developerblog.redhat.com/2015/02/05/gcc5-and-the-c11-abi/
>> As far as I know, every gcc release will break C++ ABI, so it would mean
>> rebuilding everything C++.
> 
> According to the Red Hat blog above, the last time g++ caused an ABI
> change was back in the 3.x days, so it hasn't happened for a while. Ah
> well, we have maintainers for most packages in Cygwin, so we'll have to
> co-ordinate a rebuild.

Regardless, JonY is correct.  Every C++ release, regardless of the
vendor, causes an ABI break with shared libraries and the naming of the
object elements (mangled names).  It is the nature created by technical
reason.  Not to mention that the first paragraph of the release states
that there is an ABI break.  And then there is this paragraph: "So the
plan for this ABI change has been to leave the soname (and the existing
binary interface) alone, and express the new ABI using different mangled
names."

-- 
cyg Simple

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

[Christian Franke]

cyg Simple wrote:
> On 9/30/2015 7:36 PM, David Stacey wrote:
>> On 30/09/15 23:34, JonY wrote:
>>> On 10/1/2015 00:05, David Stacey wrote:
>>>> On 30/09/15 12:15, JonY wrote:
>>>>> gcc-5.2.0-1 has been uploaded for 32bit and 64bit Cygwin.
>>>>>
>>>>> This is the first series of the 5.x releases, and should be considered
>>>>> as experimental as such.
>>>> Have you managed to work around the ABI change in gcc-5 [1], or will
>>>> this require a mass rebuild at the point gcc-5 becomes 'current'?
>>>>
>>>> [1] -http://developerblog.redhat.com/2015/02/05/gcc5-and-the-c11-abi/
>>> As far as I know, every gcc release will break C++ ABI, so it would mean
>>> rebuilding everything C++.
>> According to the Red Hat blog above, the last time g++ caused an ABI
>> change was back in the 3.x days, so it hasn't happened for a while. Ah
>> well, we have maintainers for most packages in Cygwin, so we'll have to
>> co-ordinate a rebuild.
> Regardless, JonY is correct.  Every C++ release, regardless of the
> vendor, causes an ABI break with shared libraries and the naming of the
> object elements (mangled names).

Probably not in this 4.X -> 5.X case. Otherwise the new cygstdc++-6.dll 
should IMO be renamed to -6.1, -7 or similar.

A diff of exported symbols of cygstdc++-6.dll 4.9.3-1 and 5.2.0 shows 
that the new version only adds new symbols.


Results of a quick build check with smartmontools package:

- Existing package which was built with g++ 4.9.2 runs with 5.2.0 
cygstdc++-6.dll and cyggcc_s-1.dll.

- A rebuild with g++ 5.2.0 does not run with 4.9.3 cygstdc++-6.dll due 
to missing symbols. This is as expected because 5.X adds a new C++11 
conforming std::string which is used by default 
(https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html).

- A rebuild with g++ 5.2.0 and CXXFLAGS=-D_GLIBCXX_USE_CXX11_ABI=0 runs 
with old 4.9.3 cygstdc++-6.dll and cyggcc_s.dll.

So there is probably no need to rebuild existing C++ packages.


BTW: segment sizes differ considerably:

$ size */smartctl.exe
    text    data     bss     dec     hex filename
  459725    6956   66320  533001   82209 gcc-4-old-strings/smartctl.exe
  451525    7424   66260  525209   80399 gcc-5-new-strings/smartctl.exe
  517337    8912   66196  592445   90a3d gcc-5-old-strings/smartctl.exe

Christian


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

[Ken Brown]

On 10/6/2015 12:26 PM, Christian Franke wrote:
> cyg Simple wrote:
>> On 9/30/2015 7:36 PM, David Stacey wrote:
>>> On 30/09/15 23:34, JonY wrote:
>>>> On 10/1/2015 00:05, David Stacey wrote:
>>>>> On 30/09/15 12:15, JonY wrote:
>>>>>> gcc-5.2.0-1 has been uploaded for 32bit and 64bit Cygwin.
>>>>>>
>>>>>> This is the first series of the 5.x releases, and should be
>>>>>> considered
>>>>>> as experimental as such.
>>>>> Have you managed to work around the ABI change in gcc-5 [1], or will
>>>>> this require a mass rebuild at the point gcc-5 becomes 'current'?
>>>>>
>>>>> [1] -http://developerblog.redhat.com/2015/02/05/gcc5-and-the-c11-abi/
>>>> As far as I know, every gcc release will break C++ ABI, so it would
>>>> mean
>>>> rebuilding everything C++.
>>> According to the Red Hat blog above, the last time g++ caused an ABI
>>> change was back in the 3.x days, so it hasn't happened for a while. Ah
>>> well, we have maintainers for most packages in Cygwin, so we'll have to
>>> co-ordinate a rebuild.
>> Regardless, JonY is correct.  Every C++ release, regardless of the
>> vendor, causes an ABI break with shared libraries and the naming of the
>> object elements (mangled names).
>
> Probably not in this 4.X -> 5.X case. Otherwise the new cygstdc++-6.dll
> should IMO be renamed to -6.1, -7 or similar.
>
> A diff of exported symbols of cygstdc++-6.dll 4.9.3-1 and 5.2.0 shows
> that the new version only adds new symbols.
>
>
> Results of a quick build check with smartmontools package:
>
> - Existing package which was built with g++ 4.9.2 runs with 5.2.0
> cygstdc++-6.dll and cyggcc_s-1.dll.
>
> - A rebuild with g++ 5.2.0 does not run with 4.9.3 cygstdc++-6.dll due
> to missing symbols. This is as expected because 5.X adds a new C++11
> conforming std::string which is used by default
> (https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html).
>
> - A rebuild with g++ 5.2.0 and CXXFLAGS=-D_GLIBCXX_USE_CXX11_ABI=0 runs
> with old 4.9.3 cygstdc++-6.dll and cyggcc_s.dll.
>
> So there is probably no need to rebuild existing C++ packages.

What about the following scenario: Package P links against library L. 
Library L is rebuilt to use the new ABI.  Isn't it possible that package 
P will then have to be rebuilt as well?

Ken


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

[David Stacey]

On 06/10/15 18:51, Ken Brown wrote:
> On 10/6/2015 12:26 PM, Christian Franke wrote:
>> cyg Simple wrote:
>>> On 9/30/2015 7:36 PM, David Stacey wrote:
>>>> On 30/09/15 23:34, JonY wrote:
>>>>> On 10/1/2015 00:05, David Stacey wrote:
>>>>>> On 30/09/15 12:15, JonY wrote:
>>>>>>> gcc-5.2.0-1 has been uploaded for 32bit and 64bit Cygwin.
>>>>>>>
>>>>>>> This is the first series of the 5.x releases, and should be
>>>>>>> considered
>>>>>>> as experimental as such.
>>>>>> Have you managed to work around the ABI change in gcc-5 [1], or will
>>>>>> this require a mass rebuild at the point gcc-5 becomes 'current'?
>>>>>>
>>>>>> [1] 
>>>>>> -http://developerblog.redhat.com/2015/02/05/gcc5-and-the-c11-abi/
>>>>> As far as I know, every gcc release will break C++ ABI, so it would
>>>>> mean
>>>>> rebuilding everything C++.
>>>> According to the Red Hat blog above, the last time g++ caused an ABI
>>>> change was back in the 3.x days, so it hasn't happened for a while. Ah
>>>> well, we have maintainers for most packages in Cygwin, so we'll 
>>>> have to
>>>> co-ordinate a rebuild.
>>> Regardless, JonY is correct.  Every C++ release, regardless of the
>>> vendor, causes an ABI break with shared libraries and the naming of the
>>> object elements (mangled names).
>>
>> Probably not in this 4.X -> 5.X case. Otherwise the new cygstdc++-6.dll
>> should IMO be renamed to -6.1, -7 or similar.
>>
>> A diff of exported symbols of cygstdc++-6.dll 4.9.3-1 and 5.2.0 shows
>> that the new version only adds new symbols.
>>
>>
>> Results of a quick build check with smartmontools package:
>>
>> - Existing package which was built with g++ 4.9.2 runs with 5.2.0
>> cygstdc++-6.dll and cyggcc_s-1.dll.
>>
>> - A rebuild with g++ 5.2.0 does not run with 4.9.3 cygstdc++-6.dll due
>> to missing symbols. This is as expected because 5.X adds a new C++11
>> conforming std::string which is used by default
>> (https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html).
>>
>> - A rebuild with g++ 5.2.0 and CXXFLAGS=-D_GLIBCXX_USE_CXX11_ABI=0 runs
>> with old 4.9.3 cygstdc++-6.dll and cyggcc_s.dll.
>>
>> So there is probably no need to rebuild existing C++ packages.
>
> What about the following scenario: Package P links against library L. 
> Library L is rebuilt to use the new ABI.  Isn't it possible that 
> package P will then have to be rebuilt as well?

My understanding is that if 'L' is built with -D_GLIBCXX_USE_CXX11_ABI=0 
then P wouldn't need to be rebuilt (I haven't tested that, though). 
However, this compiler switch loses some C++11 features of the STL, so 
it isn't a magic cure.

Dave.


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

[David Stacey]

On 06/10/15 17:26, Christian Franke wrote:
> cyg Simple wrote:
>> On 9/30/2015 7:36 PM, David Stacey wrote:
>>> On 30/09/15 23:34, JonY wrote:
>>>> On 10/1/2015 00:05, David Stacey wrote:
>>>>> On 30/09/15 12:15, JonY wrote:
>>>>>> gcc-5.2.0-1 has been uploaded for 32bit and 64bit Cygwin.
>>>>>>
>>>>>> This is the first series of the 5.x releases, and should be 
>>>>>> considered
>>>>>> as experimental as such.
>>>>> Have you managed to work around the ABI change in gcc-5 [1], or will
>>>>> this require a mass rebuild at the point gcc-5 becomes 'current'?
>>>>>
>>>>> [1] -http://developerblog.redhat.com/2015/02/05/gcc5-and-the-c11-abi/
>>>> As far as I know, every gcc release will break C++ ABI, so it would 
>>>> mean
>>>> rebuilding everything C++.
>>> According to the Red Hat blog above, the last time g++ caused an ABI
>>> change was back in the 3.x days, so it hasn't happened for a while. Ah
>>> well, we have maintainers for most packages in Cygwin, so we'll have to
>>> co-ordinate a rebuild.
>> Regardless, JonY is correct.  Every C++ release, regardless of the
>> vendor, causes an ABI break with shared libraries and the naming of the
>> object elements (mangled names).
>
> Probably not in this 4.X -> 5.X case. Otherwise the new 
> cygstdc++-6.dll should IMO be renamed to -6.1, -7 or similar.

Bumping the DLL number wouldn't necessarily fix the problem. You'd still 
run into conflicts if one executable loaded two DLLs, each linked 
against different versions of libstdc++.

Dave.


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Updated: gcc-5.2.0-1 (Test x86/x86_64)

[Csaba Raduly]

On Wed, Oct 7, 2015 at 12:12 AM, David Stacey  wrote:
>
> On 06/10/15 18:51, Ken Brown wrote:

(snip)

>>
>>
>> What about the following scenario: Package P links against library L. Library L is rebuilt to use the new ABI.  Isn't it possible that package P will then have to be rebuilt as well?
>
>
> My understanding is that if 'L' is built with -D_GLIBCXX_USE_CXX11_ABI=0 then P wouldn't need to be rebuilt (I haven't tested that, though). However, this compiler switch loses some C++11 features of the STL, so it isn't a magic cure.
>

If you use -D_GLIBCXX_USE_CXX11_ABI=0, you're stuck with the COW (the
old std::string implementation in libstdc++ which uses copy-on-write;
this is not standard-conforming anymore). std::list::size() remains
O(n) instead of the newly mandated O(1).

AFAIK, that's it.

Csaba
-- 
GCS a+ e++ d- C++ ULS$ L+$ !E- W++ P+++$ w++$ tv+ b++ DI D++ 5++
The Tao of math: The numbers you can count are not the real numbers.
Life is complex, with real and imaginary parts.
"Ok, it boots. Which means it must be bug-free and perfect. " -- Linus Torvalds
"People disagree with me. I just ignore them." -- Linus Torvalds

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple