C++2a synchronisation inefficient in GCC 11

[Jonathan Wakely]

On 03/03/21 09:07 -0800, Thiago Macieira via Libstdc++ wrote:
>On Wednesday, 3 March 2021 06:30:20 PST Jonathan Wakely wrote:
>> Yes, exactly. They could still do that if some additional option/macro
>> was needed to get to those new facilities though.
>>
>> My concern is that if we add a new "I know what I'm doing" option,
>> everybody defines it (because that's what blog posts and youtube
>> videos and conference talks tell them to do, because that gives them
>> the new shiny things) and then it just becomes a default.
>>
>> Maybe Qt wouldn't define it, but plenty would.
>
>Well, if you define the "void warranty" option and you break things, you get
>to keep both pieces.
>
>Another option is to fiddle with the include path.
>
>$ gcc -v -E -xc++ /dev/null
>Using built-in specs.
>[...]
>#include <...> search starts here:
> /usr/lib64/gcc/x86_64-generic-linux/10/../../../../include/c++/10
> /usr/lib64/gcc/x86_64-generic-linux/10/../../../../include/c++/10/x86_64-
>generic-linux
> /usr/lib64/gcc/x86_64-generic-linux/10/../../../../include/c++/10/backward
> /usr/lib64/gcc/x86_64-generic-linux/10/include
> /usr/local/include
> /usr/lib64/gcc/x86_64-generic-linux/10/include-fixed
> /usr/include
>
>That "backward" has always got me thinking. What if we had a command-line
>option like -fexperimental-c++ option that added the $CXXROOT/experimental

We already have an experimental sub-dir for TS headers, so you need to
pick another name. Otherwise <vector> and <experimental/vector> get
mixed up.