Re: std::forward_list optim for always equal allocator

[Jonathan Wakely <jwakely at redhat dot com>]

On 23/11/17 22:22 +0100, François Dumont wrote:
> Gentle reminder for this patch.
>
> I looked when the constructor got unused and I think it is back in 
> June 2015 in git commit:
>
> commit debb6aabb771ed02cb7256a7719555e5fbd7d3f7
> Author: redi <redi@138bc75d-0d04-0410-961f-82ee72b054a4>
> Date:   Wed Jun 17 17:45:45 2015 +0000
>
>         * include/bits/forward_list.h
>         (_Fwd_list_base(const _Node_alloc_type&)): Change parameter to
>         rvalue-reference.

Hmm, I should have put that same change on the gcc-5-branch too.

> If you fear abi breaking change I can restore it in a 
> !_GLIBCXX_INLINE_VERSION section.

I think if there was a problem here my June 2015 change would already
have caused it (when I changed the _Fwd_list_base constructor
signatures).

So let's assume it's OK to remove the constructor.


> @@ -533,15 +560,13 @@ _GLIBCXX_BEGIN_NAMESPACE_CONTAINER
>
>       /**
>        *  @brief  The %forward_list move constructor.
> -       *  @param  __list  A %forward_list of identical element and allocator
> -       *                  types.
> +       *  @param  A %forward_list of identical element and allocator types.

This change is wrong, you can't just remove the parameter name,
because now Doxygen will document a parameter called "A" (and complain
that there is no such parameter).

It would be better to leave the name __list there and just get the
warning.

Otherwise the patch is OK for trunk (please ensure to update the
Copyright dates in the test files to 2018).

Thanks.