DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 15CB3385843B
Mime-Version: 1.0 (Mac OS X Mail 16.0 \(3731.700.6.1.9\))
Subject: Re: g++ 12 -std=gnu++20
In-Reply-To: <575d416a-7ee0-4e85-82f2-d97b4cd511da@SystematicSW.ab.ca>
Date: Thu, 31 Oct 2024 16:34:57 +0100
Cc: Brian Inglis <Brian.Inglis@SystematicSW.ab.ca>
Message-Id: <E1BCA6B2-D979-4166-AA22-729D3B5EEFE0@unified-streaming.com>
References: <48e51b50-63db-4b14-b321-46942e40862e@SystematicSW.ab.ca>
 <3805509C-51BA-46F9-95A3-007CE99B014A@unified-streaming.com>
 <764d17ab-a660-4e89-8fd3-175e93f11507@gmail.com>
 <B7BA05FB-75CE-48DE-92ED-267540E59FD0@unified-streaming.com>
 <575d416a-7ee0-4e85-82f2-d97b4cd511da@SystematicSW.ab.ca>
To: cygwin@cygwin.com
Precedence: list
From: Dimitry Andric via Cygwin <cygwin@cygwin.com>
Reply-To: Dimitry Andric <dimitry@unified-streaming.com>
Content-Type: text/plain; charset="utf-8"
Errors-To: cygwin-bounces~archive-cygwin=delorie.com@cygwin.com
Sender: "Cygwin" <cygwin-bounces~archive-cygwin=delorie.com@cygwin.com>
Content-Transfer-Encoding: 8bit

On 31 Oct 2024, at 06:40, Brian Inglis via Cygwin <cygwin@cygwin.com> wrote:
> 
> On 2024-10-30 10:30, Dimitry Andric via Cygwin wrote:
>> On 30 Oct 2024, at 17:14, Csaba Ráduly via Cygwin wrote:
>>> On 30/10/2024 17:00, Dimitry Andric via Cygwin wrote:
>>>> #include<string>
>>>> #include<cassert>
>>>> 
>>>> constexpr bool foo()
>>>> {
>>>>   std::string str2{"abcwe"};
>>>>   return str2.size()==5;
>>>> }
>>>> 
>>>> static_assert(foo());
>>>> 
>>>> int main()
>>>> {
>>>>   assert(foo());
>>>> }
>>> 
>>> Seems like _GLIBCXX_USE_CXX11_ABI is not defined by default.
>> Indeed, /usr/lib/gcc/x86_64-pc-cygwin/12/include/c++/x86_64-pc-cygwin/bits/c++config.h shows:
>>    317  # define _GLIBCXX_USE_DUAL_ABI 1
>>    318
>>    319  #if ! _GLIBCXX_USE_DUAL_ABI
>>    320  // Ignore any pre-defined value of _GLIBCXX_USE_CXX11_ABI
>>    321  # undef _GLIBCXX_USE_CXX11_ABI
>>    322  #endif
>>    323
>>    324  #ifndef _GLIBCXX_USE_CXX11_ABI
>>    325  # define _GLIBCXX_USE_CXX11_ABI 0
>>    326  #endif
>>  Also, g++ -v shows --with-default-libstdcxx-abi=gcc4-compatible, so I guess this was deliberate on the part of the packagers. Probably for backwards compatibility reasons?
> 
> Okay, sorry, I typoed an extra _ before _GLIBCXX_NOEXCEPT!
> 
> Cygwin gcc/g++ 13.3 is available as a test package; only with .4/.5 are gcc version features and fixes considered stable.
> 
> The build is requesting -std=gnu++20, so why is 11 ABI relevant?
> 
> As I want to use the latest language, why is gcc 4 compatible ABI *default* relevant?
> 
> What can I do in the package build to get g++ to not use the default but whatever is required to support the latest language?
> 
> Where can I find docs on these symbols and their meanings, uses, and settings?

It is a historical artifact that has been dragged along for years. See https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html for more information.

During the gcc 5.1 release, changes were made to libstdc++ that made the ABI incompatible with previous releases, specifically for std::string. You can define _GLIBCXX_USE_CXX11_ABI to 1 to use the 'new' ABI, but then it becomes tricky to link your code to libraries compiled with older versions of libstdc++, or newer versions that define _GLIBCXX_USE_CXX11_ABI to 0.

I think most Linux distributions have switched fully to the new ABI by now, and dropped support for the old ABI, so they configure their gcc's with _GLIBCXX_USE_CXX11_ABI=1 by default.

-Dimitry


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