Something about std::list<> member functions.

[Jerry Quinn]

Gabriel Dos Reis writes:
 > Nathan Myers <ncm-nospam@cantrip.org> writes:
 > 
 > | On Sat, Aug 09, 2003 at 08:55:16PM +0200, Gabriel Dos Reis wrote:
 > | > Dhruv Matani <dhruvbird@gmx.net> writes:
 > | > 
 > | > [...]
 > | > 
 > | > | form At&T syntax, but I thought (and still do think) that compiler
 > | > | technology should have evolved by now so as to optimize such stuff into
 > | > | nothingness...... And, looking at g++ optimizations, I definitely
 > | > | thought that it could (at least now) optimize away the layer.
 > | > 
 > | > See recent discussions on gcc@gcc.gnu.org about inlining.
 > | 
 > | Without consulting those, I recall Jason Merrill mentioning that
 > | you can get enormous optimization improvements just from declaring
 > | the inline functions' arguments "const", unnecessarily.  Is that 
 > | advice still current?  Since it doesn't affect the ABI or standard 
 > | conformance, that seems worth doing where it makes a difference.
 > 
 > We already pass the arguments by const reference.

If I remember right, the issue isn't passing by const ref.  It's that
early on in parsing(?) the input, when the argument itself isn't
const, the compiler fails to see that inlining is OK.  Which means you
would have to have something like:

int function(const T& x const)

in order to work around the problem.  Even though declaring x const is
silly.

Richard Henderson was the one who told us about it.  It came up when
looking at performance problems in std::copy.

Jerry Quinn