[PATCH] Remove redundant accessors in hash tables

[Jonathan Wakely]

On 16/05/19 11:05 +0100, Jonathan Wakely wrote:
>On 16/05/19 07:47 +0200, François Dumont wrote:
>>On 5/15/19 5:37 PM, Jonathan Wakely wrote:
>>>François,
>>>I noticed that _Hash_code_base and _Hashtable_base have a number of
>>>member functions which are overloaded for const and non-const:
>>>
>>>Â Â  const _Equal&
>>>Â Â  _M_eq() const { return _EqualEBO::_S_cget(*this); }
>>>
>>>Â Â  _Equal&
>>>Â Â  _M_eq() { return _EqualEBO::_S_get(*this); }
>>>
>>>The non-const ones seem to be unnecessary. They're used in the _M_swap
>>>member functions, but all other uses could (and probably should) call
>>>the const overload to get a const reference to the function object.
>>>
>>>If we make the _M_swap members use the EBO accessors directly then we
>>>can get rid of the non-const accessors. That makes overload resolution
>>>simpler for the compiler (as there's only one function to choose from)
>>>and should result in slightly smaller code when inlining is not
>>>enabled.
>>>
>>>Do you see any problem with this patch?
>>>
>>>
>>I think it is more a Pavlov behavior, always providing const and 
>>non-const no matter what.
>>
>>No problem to simplify this.
>
>OK, tested powerpc64le-linux, committed to trunk.

I don't see a need for the _Hashtable_ebo_helper member functions to
be static. They return a member of *this, but are static so *this has
to be passed as a parameter. We could just make them non-static.

It seems to have always been that way since the first version of the
patch that added the helpers:
https://gcc.gnu.org/ml/libstdc++/2011-12/msg00139.html

This patch passes all tests. I plan to commit this to trunk too.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: patch.txt
Type: text/x-patch
Size: 5662 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20190516/fbcb3e63/attachment.bin>